Greenhouse environment control method and system based on swarm intelligence

Abstract

The application discloses a greenhouse environment control method and system based on swarm intelligence, and relates to the technical field of intelligent control of greenhouse environment, which comprises the following steps: acquiring greenhouse environment state, crop canopy state, outdoor weather, equipment operation and historical feedback data; determining a leaf transpiration-dew double-side safety domain, predicting a canopy environment trajectory interval and a safety boundary crossing time window; identifying index level action timing and unit control quantity adjustment contribution of an environment adjustment device, determining an environment correction amount and generating a candidate control action sequence; constructing a historical response calculation scene, determining an effective action arrival time window and a conservative predicted canopy environment trajectory envelope; iteratively evaluating the candidate sequence by using a swarm intelligence algorithm, determining an optimal or emergency control action sequence and executing; and updating a prediction error boundary, a historical response sample and a device response parameter according to an actual canopy environment trajectory. The application can identify canopy environment risks in advance, improve control timeliness, reliability and working condition adaptability.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for greenhouse environments, specifically to a method and system for controlling greenhouse environments based on swarm intelligence. Background Technology

[0002] As facility agriculture develops towards digitalization, precision, and autonomy, greenhouse environmental control has gradually shifted from single-variable regulation based on fixed thresholds to intelligent control that integrates sensor monitoring, crop physiological state analysis, environmental prediction models, and multi-device collaborative optimization. Existing greenhouses typically collect data on air temperature, humidity, solar irradiance, outdoor wind speed, and crop canopy temperature, and regulate the canopy microenvironment through ventilation windows, circulating fans, heating equipment, misting humidification equipment, and shading devices. The application of model predictive control, fuzzy control, neural network control, and swarm intelligence optimization methods enables the control system to assess environmental change trends within a certain prediction time domain and generate control strategies based on energy consumption, temperature and humidity deviations, and equipment operating constraints. This provides a technological foundation for improving the stability of the crop growth environment and the collaborative operation capabilities of environmental regulation equipment.

[0003] However, existing greenhouse environmental control methods typically rely primarily on air temperature, relative humidity, or deviations from their setpoints, failing to simultaneously describe the transpiration and dew safety status of the crop canopy from two perspectives: leaf vapor pressure deficit and leaf dew margin. This makes it difficult to identify situations where the canopy environment has not yet clearly crossed the safety boundary but is already showing a tendency to do so. While some predictive control methods can predict changes in environmental parameters, they usually output a single predictive trajectory, making it difficult to characterize the trajectory uncertainty caused by meteorological forecast errors, equipment response fluctuations, and interactions among multiple devices. Furthermore, it is difficult to accurately determine the timeframe in which safety risks may occur. In addition, ventilation, heating, humidification, and shading equipment exhibit varying degrees of start-up delay, rise time, and residual effects between changes in control parameters and an effective response from the canopy environment. Moreover, the regulatory contribution per unit of control varies with greenhouse conditions, control direction, and equipment status. Existing methods typically use fixed equipment parameters or instantaneous environmental deviations to directly allocate control quantities, making it difficult to determine whether the control action can generate sufficient environmental correction before the safety boundary is crossed.

[0004] For coordinated operations of multiple devices, existing optimization methods may still rely solely on central prediction results for evaluation, lacking joint constraints on control timing margin, effective regulation margin, and conservative environmental safety envelope. This can easily lead to problems such as theoretically effective control strategies arriving too late in practice, insufficient regulation, or inadequate consideration of device coupling effects. Furthermore, the environmental response after actual execution cannot be effectively used to update the prediction error boundary and device response parameters, making it difficult for the control model to continuously adapt to seasonal changes, equipment aging, and changes in greenhouse operating conditions. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by this invention is that existing greenhouse environmental control methods have the following drawbacks: they are difficult to simultaneously characterize the safety risks of crop canopy transpiration and leaf dew formation; they are difficult to accurately determine the timing of the effects of different environmental regulation devices on various canopy environmental indicators and the adjustment contribution of unit control quantities; they are difficult to ensure the timeliness and sufficiency of control actions under conditions of equipment response fluctuations, multi-device interaction, and prediction errors; and they are also concerned with how to achieve multi-device collaborative, predictive, and rolling self-correcting control under canopy environmental safety constraints based on swarm intelligence optimization and actual control feedback.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, embodiments of the present invention provide a greenhouse environment control method based on swarm intelligence, comprising the following steps: S1: Acquire greenhouse environmental status data, crop canopy status data, outdoor meteorological data, environmental regulation equipment operation data, and historical control feedback data for the current control cycle; S2: Determine the leaf transpiration-condensation bilateral safety domain based on the greenhouse environmental status data and crop canopy status data. Based on the greenhouse environmental status data, crop canopy status data, outdoor meteorological data, and historical control feedback data, predict the canopy environment trajectory range to maintain the current equipment control state, and determine the leaf safety boundary crossing time window. S3: Based on the operating data of the environmental control equipment and the historical control feedback data, determine the timing of the action of each environmental control equipment on the index-level equipment and the environmental control contribution per unit control quantity for the current operating conditions of the blade vapor pressure deficit and blade condensation margin. S4: Determine the environmental correction amount and generate candidate environmental control action sequences based on the deviation between the canopy environment trajectory interval and the bilateral safety domain. Construct a historical response calculation scenario based on the index-level time series and unit contribution, and determine the effective arrival time window, representative predicted canopy environment trajectory, and conservative predicted canopy environment trajectory envelope for each candidate sequence. S5: The candidate sequences are iteratively updated using a swarm intelligence algorithm. In each iteration, the effective arrival window, the representative predicted canopy environment trajectory, and the conservative predicted canopy environment trajectory envelope of each candidate environmental control action sequence are redefined. The control timing margin is determined based on the crossing and arrival windows, and the effective adjustment margin is determined based on the cumulative effective correction and environmental correction before the start time of the crossing window. The strategy is evaluated by combining the conservative predicted envelope, timing margin, and adjustment margin. After the iteration terminates, the best candidate sequence is selected from those whose conservative envelope satisfies the safety domain constraint and whose timing margin and adjustment margin are not lower than the corresponding threshold. If there is no candidate sequence that satisfies the constraints, the emergency sequence is determined based on the degree of constraint violation. S6: Select the preferred or emergency sequence as the sequence to be executed and output the equipment control command. After obtaining the actual canopy environment trajectory, update the prediction error and boundary according to the deviation between the actual control path and the predicted trajectory. Update the historical samples according to the identifiable actual response segments, correct the index-level time series and unit contribution, and use the updated results for the next cycle.

[0008] As a preferred embodiment of the greenhouse environment control method based on swarm intelligence described in this invention, wherein: in step S1: The greenhouse environmental status data includes air temperature and relative humidity. Crop canopy status data includes crop canopy temperature; Outdoor meteorological data includes outdoor temperature, outdoor relative humidity, solar irradiance, and outdoor wind speed; Environmental control equipment operation data includes the current control variables and historical control variable sequences for each device; Historical control feedback data includes historical actual canopy environment trajectories and historical prediction error data.

[0009] As a preferred embodiment of the greenhouse environment control method based on swarm intelligence described in this invention, wherein: in step S2, determining the bilateral security domain includes: Determine the actual water vapor pressure and dew point temperature of the air based on the air temperature and relative humidity, and determine the saturated water vapor pressure of the canopy based on the crop canopy temperature. The difference between the saturated vapor pressure of the canopy and the actual vapor pressure of the air is taken as the leaf vapor pressure deficit, and the difference between the crop canopy temperature and the air dew point temperature is taken as the leaf dew margin. The range of conditions in which the vapor pressure deficit of the blades is not lower than a preset lower limit and not higher than a preset upper limit, and the condensation margin of the blades is not lower than a preset lower limit, is defined as the bilateral safety domain.

[0010] As a preferred embodiment of the greenhouse environment control method based on swarm intelligence described in this invention, step S2, predicting the canopy environment trajectory interval and determining the crossing time window, includes: Acquire the prediction sequence of outdoor meteorological data within the preset prediction time domain, and combine it with the greenhouse status, crop canopy status, current control values ​​of each device and historical actual trajectory to predict the baseline trajectory of leaf vapor pressure deficit and leaf dew margin to maintain the current control state. Based on historical prediction errors, the error range for each prediction time is determined, forming the upper and lower bound trajectories of the two indicators, thus obtaining the canopy environment trajectory range. The earliest possible crossing time of each safety boundary is defined as the moment when the lower boundary of the leaf vapor pressure deficit first falls below its lower limit, the upper boundary first rises above its upper limit, and the lower boundary of the leaf condensation margin first falls below its lower limit. The moment when the upper limit of leaf vapor pressure deficit first falls below its lower limit, the moment when the lower limit first rises above its upper limit, and the moment when the upper limit of leaf condensation margin first falls below its lower limit are taken as the latest possible crossing time of each safety boundary. The earliest time among the earliest possible crossing times corresponding to each safety boundary with predicted crossing is taken as the start time of the blade safety boundary crossing time window, and the latest time among the latest possible crossing times is taken as the end time of the blade safety boundary crossing time window.

[0011] As a preferred embodiment of the greenhouse environment control method based on swarm intelligence described in this invention, step S3 specifically includes: Based on the historical control quantity sequence of each device, determine the timing of control quantity changes, the magnitude of signed changes, and the direction of changes, and extract trajectory segments within the preset response observation period from the historical actual trajectory to form historical control response samples; For leaf vapor pressure deficit and leaf condensation margin, the starting time of the effect relative to the dynamic baseline trend before the change, the time when the effect reaches the preset effective response threshold, and the time when the effect subsides after the control quantity is restored are determined respectively. Based on the above time, the respective indicator-level action start delay, action rise time and residual action time after stopping of the two indicators are determined to form the indicator-level equipment action timing sequence. Based on the signed change amplitude and the effective change amount within the corresponding effective response statistical period, determine the environmental regulation contribution of each device to the unit control quantity of the two indicators; The time series and unit contribution of the index are saved according to the direction of change. The samples in the same direction are matched and weighted according to the difference between the current working condition and the working conditions of each sample to obtain the time series and unit contribution of the index under the current working condition.

[0012] As a preferred embodiment of the greenhouse environment control method based on swarm intelligence described in this invention, step S4 specifically includes: Based on the deviation of the canopy environmental trajectory interval relative to the bilateral safety domain during the correction demand assessment period starting from the start time of the crossing time window, the leaf vapor pressure deficit correction amount and the leaf dew margin correction amount formed no later than the start time are determined to constitute the environmental correction amount. Candidate environmental control action sequences are generated based on the current control variables and adjustable ranges of each device. Each sequence includes the action start time, signed change amplitude, and duration of each device. For each candidate sequence, sample-level index-level time series and unit contribution are selected from the matching samples of the devices participating in the action in the same direction, and combined to form a historical response calculation scenario; In each scenario, the cumulative effective correction amount of the two indicators is calculated based on the indicator-level time sequence and unit contribution of each device, and the arrival time to reach the corresponding required correction amount is determined. Based on this, the earliest and latest arrival times of the candidate sequence and the proportion of reachable scenarios are determined, forming an effective arrival window. Based on the baseline trajectory maintained under the current control state and the cumulative correction amount of each scenario, a scenario prediction trajectory is generated, and a representative predicted canopy environment trajectory is obtained by weighting according to the scenario weight. By combining the predicted trajectories for each scenario, the historical prediction error boundaries for different prediction lead times, the uncertainties of device interaction, and the extrapolation uncertainties of duration, a conservative prediction envelope for the canopy environment trajectory is formed.

[0013] As a preferred embodiment of the greenhouse environment control method based on swarm intelligence described in this invention, step S5, determining the control timing margin and effective adjustment margin, includes: The control timing margin is obtained by subtracting the latest arrival time of the arrival time window from the start time of the travel time window. When there is a need for correction of leaf vapor pressure deficit, the cumulative correction amount of leaf vapor pressure deficit at the beginning of each scenario is projected in the direction of the correction amount, and the minimum value of the projection result of each scenario is taken as the conservative effective correction amount of leaf vapor pressure deficit. When there is a need to correct the condensation margin, the minimum cumulative correction amount of the condensation margin at the beginning of each scenario is taken as the conservative effective correction amount of the condensation margin. Divide the difference between each conservative effective correction amount and the corresponding required correction amount by the larger of the absolute value of the required correction amount and the preset normalized protection value to obtain the normalized adjustment margin of the two indicators. The minimum value among the margins of indicators with correction needs is taken as the effective adjustment margin. When the environmental correction is zero, the timing margin and adjustment margin are not considered as feasibility constraints.

[0014] As a preferred embodiment of the greenhouse environment control method based on swarm intelligence described in this invention, step S5, which involves iteratively updating and evaluating strategies using a swarm intelligence algorithm, includes: Each candidate sequence is treated as an individual, and the start time, change range, and duration of each device's action are used as control variables. Each iteration updates variables based on information interaction between individuals, limiting the magnitude of change to an adjustable range and limiting the timing and duration of actions to the prediction time domain; The scenario is reconstructed for each updated sequence, and the arrival window, reachability ratio, representative trajectory and conservative envelope, as well as the temporal margin and adjustment margin are redefined. The minimum normalized safety distance is determined based on the distance between the conservative envelope and each safety boundary. After normalizing the timing margin, adjustment margin, and control action cost, the strategy evaluation result is determined, where the control action cost is determined based on the change magnitude, duration, and number of switching operations. After the iteration terminates, the best strategy evaluation sequence is selected from the candidate sequences that satisfy the safety domain, time margin and adjustment margin not lower than the threshold and meet the accessibility constraint and equipment operation constraint. If no sequence satisfies all constraints, an emergency sequence is determined based on the degree of violation of safety, timing, adjustment, and reachability constraints, and an identifier for no fully feasible control strategy and a result of constraint violation are generated.

[0015] As a preferred embodiment of the greenhouse environment control method based on swarm intelligence described in this invention, step S6 specifically includes: The preferred environmental control action sequence, or the emergency environmental control action sequence when no fully feasible control strategy identifier is generated, is determined as the environmental control action sequence to be executed in the current control cycle; The control command corresponding to the current cycle is sent to each device for execution, and the actual feedback control quantity, air temperature, air relative humidity and crop canopy temperature are collected to form the actual canopy environment trajectory; The path difference between the actual control process and the planned process is determined based on the actual feedback control quantity. When the difference does not exceed the preset threshold, the representative trajectory corresponding to the sequence to be executed is used as the predicted trajectory corresponding to the actual control process. Otherwise, the actual action sequence is generated based on the actual feedback control quantity and the predicted trajectory is regenerated. The actual trajectory is compared with the predicted trajectory corresponding to the actual control process according to the prediction lead time to obtain the prediction error of the two indicators. The historical prediction error set and its upper and lower deviation boundaries of each prediction lead time are updated for the formation of the trajectory interval in the next cycle. Establish equipment response tracking records based on the actual feedback control quantity change process, extract actual control response segments from closed records, add segments that meet the single device identification conditions to the historical sample set according to device, change direction and effective response index, and add segments with overlapping effective effects of multiple devices that cannot be separated to the joint sample set. Based on the segments that meet the identifiable conditions, determine the actual indicator-level effect start-up delay, rise time, residual effect time after stopping, and actual unit contribution of the two indicators, and perform confidence weighting and single change limit update on the directional initial parameters in the same direction; The actual trajectory, historical prediction error set, deviation boundary, updated historical sample set, joint sample set, and directional initial parameters are used as historical control feedback data for the next cycle. In the next cycle, the operating condition matching is re-executed to determine the index-level time series and unit contribution.

[0016] Secondly, embodiments of the present invention provide a greenhouse environment control system based on swarm intelligence, comprising: Data acquisition module: Acquires greenhouse environmental status data, crop canopy status data, outdoor meteorological data, environmental regulation equipment operation data, and historical control feedback data for the current control cycle; Safety risk prediction module: Based on the greenhouse environmental status data and crop canopy status data, determine the leaf transpiration-condensation bilateral safety domain; based on the greenhouse environmental status data, crop canopy status data, outdoor meteorological data and historical control feedback data, predict the canopy environment trajectory range to maintain the current equipment control state, and determine the leaf safety boundary crossing time window; Equipment Operation Parameter Determination Module: Based on the operating data of the environmental regulation equipment and historical control feedback data, determine the timing of the operation of each environmental regulation equipment and the environmental regulation contribution per unit control quantity to the indicator-level equipment for the current operating conditions of the blade vapor pressure deficit and blade condensation margin. Candidate control action generation module: Determines the environmental correction amount and generates candidate environmental control action sequences based on the deviation between the canopy environment trajectory interval and the bilateral safety domain; Constructs a historical response calculation scenario based on the index-level time series and unit contribution; Determines the effective arrival time window, representative predicted canopy environment trajectory, and conservative predicted canopy environment trajectory envelope for each candidate sequence. The control strategy optimization module utilizes a swarm intelligence algorithm to iteratively update candidate sequences. In each iteration, the effective arrival window, the representative predicted canopy environment trajectory, and the conservative predicted canopy environment trajectory envelope of each candidate environmental control action sequence are redefined. The control timing margin is determined based on the crossing and arrival windows, and the effective regulation margin is determined based on the cumulative effective correction and environmental correction before the start of the crossing window. The strategy is evaluated by combining the conservative predicted envelope, timing margin, and regulation margin. After the iteration terminates, the best candidate sequence is selected from those whose conservative envelope satisfies the safety domain constraint and whose timing margin and regulation margin are not lower than the corresponding thresholds. If no candidate sequence satisfies the constraints, an emergency sequence is determined based on the degree of constraint violation. Control execution and feedback correction module: It takes the preferred or emergency sequence as the sequence to be executed and outputs the equipment control command. After obtaining the actual canopy environment trajectory, it updates the prediction error and boundary according to the deviation between the actual control path and the predicted trajectory. It updates the historical samples according to the identifiable actual response segments, corrects the index-level time series and unit contribution, and uses the updated results for the next cycle.

[0017] The beneficial effects of this invention are as follows: By constructing a bilateral safety domain for leaf transpiration and condensation and a leaf safety boundary crossing time window, this invention enables early identification of canopy environmental risks; by identifying the timing of the action of each environmental regulation device at the index level and the environmental regulation contribution per unit control quantity, it can evaluate the arrival time and effective regulation capability of candidate control actions; by calculating scenarios based on historical responses and conservatively predicting the canopy environmental trajectory envelope, it reduces the impact of device response fluctuations and interaction uncertainties on control results; by utilizing swarm intelligence algorithms to comprehensively optimize environmental safety, control timing margin, effective regulation margin, and control action cost, and updating prediction errors and device response parameters based on actual execution results, it improves the timeliness, reliability, and adaptability of greenhouse environmental control. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 The first embodiment of the present invention provides an overall flowchart of a greenhouse environment control method based on swarm intelligence; Figure 2 The third embodiment of the present invention provides a module connection diagram of a greenhouse environment control system based on swarm intelligence. Detailed Implementation

[0019] To make the above-mentioned objects, features, and advantages of the present invention more readily understood, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0020] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for controlling the environment of a greenhouse based on swarm intelligence is provided.

[0021] S1: Obtain greenhouse environmental status data, crop canopy status data, outdoor meteorological data, environmental regulation equipment operation data, and historical control feedback data for the current control cycle.

[0022] The greenhouse environmental status data includes air temperature and relative humidity. Crop canopy status data includes crop canopy temperature; Outdoor meteorological data includes outdoor temperature, outdoor relative humidity, solar irradiance, and outdoor wind speed; Environmental control equipment operation data includes the current control variables and historical control variable sequences for each device; Historical control feedback data includes historical actual canopy environment trajectories and historical prediction error data.

[0023] In this embodiment, step S1 is used to acquire greenhouse environmental status data, crop canopy status data, outdoor meteorological data, environmental regulation equipment operation data, and historical control feedback data for the current control cycle, so as to form current status input, external disturbance input, equipment control input, and historical feedback input with a unified time reference, providing a data foundation for subsequent determination of leaf transpiration-condensation bilateral safety domain, prediction of canopy environmental trajectory interval, identification of equipment action parameters, generation of candidate environmental control action sequences, and strategy evaluation.

[0024] Specifically, the greenhouse is divided into at least one independent control zone. Each independent control zone is equipped with corresponding environmental status acquisition points and environmental control equipment, and the greenhouse controller outputs control commands to the environmental control equipment. This embodiment uses one independent control zone as an example. The environmental control equipment includes ventilation windows, circulating fans, heating equipment, atomizing humidification equipment, and shading equipment. It should be noted that the atomizing humidification equipment can also be replaced with wet curtains, spray equipment, or other equipment capable of changing the humidity content of the greenhouse air.

[0025] The time interval between two adjacent environmental control decision moments is defined as a control cycle, and the control cycle currently in which data acquisition and control decisions are being executed is defined as the current control cycle. The length of the control cycle is determined based on the rate of change of the internal temperature and humidity of the greenhouse, the response time of the environmental regulation equipment, and the computing power of the controller, and can be set to 3 to 15 minutes. In this embodiment, the control cycle is set to 5 minutes. The sensor sampling interval is shorter than the control cycle length and can be set to 30 to 60 seconds. In this embodiment, it is set to 30 seconds, so that multiple continuous sampling values ​​can be generated within one control cycle to support the determination of the current state and the identification of changes in equipment control response.

[0026] The greenhouse environmental data includes air temperature and relative humidity. Air temperature and relative humidity are collected by air temperature and humidity sensors deployed near the crop canopy. In this embodiment, three air temperature and humidity sampling points are set up within an independent control zone. These three sampling points are located near the air inlet, in the center of the control zone, and near the air outlet, respectively, and are situated 0.2 to 0.4 meters above the crop canopy. The distance between each air temperature and humidity sampling point and the crop leaves, greenhouse frame, heating pipes, and atomizing nozzles is no less than 0.3 meters to reduce the impact of local leaf transpiration, frame heat conduction, heating radiation, and droplet adhesion on the measurement results.

[0027] The air temperature sensor's measurement range can be set from -20°C to 60°C, with a measurement error of no more than ±0.3°C; the air relative humidity sensor's measurement range can be set from 0% to 100%, with a measurement error of no more than ±2 percentage points. Each sampling point outputs air temperature and relative humidity sampling values ​​with sampling timestamps according to a preset sampling interval.

[0028] The validity of the data output from each air temperature and humidity sampling point is checked. If the air temperature sample value exceeds the sensor's measurement range, the air relative humidity sample value exceeds 0% to 100%, or the sensor outputs a communication verification error flag, the corresponding sample value is marked as invalid. For multiple air temperature samples at the same sampling time, the median air temperature is calculated. If the difference between any air temperature sample value and the median air temperature is greater than a preset temperature spatial deviation threshold, that air temperature sample value is determined to be a spatial outlier. For multiple air relative humidity samples at the same sampling time, the median air relative humidity is calculated. If the difference between any air relative humidity sample value and the median air relative humidity is greater than a preset humidity spatial deviation threshold, that air relative humidity sample value is determined to be a spatial outlier.

[0029] In this embodiment, the preset temperature spatial deviation threshold is set to 2 degrees Celsius, and the preset humidity spatial deviation threshold is set to 8 percentage points. These thresholds are determined based on sensor measurement accuracy, the normal spatial fluctuation range within the greenhouse control zone, and pre-collected data. In other embodiments, they can be recalibrated based on the greenhouse area, equipment layout, and airflow organization characteristics.

[0030] After removing invalid sampled values ​​and spatial outliers, the remaining air temperature sampled values ​​at the same sampling time are arithmetically averaged to obtain the zone air temperature at that sampling time; the remaining relative humidity sampled values ​​are also arithmetically averaged to obtain the zone relative humidity at that sampling time. Furthermore, the zone air temperature and relative humidity are averaged separately over a preset state statistical period before the end of the current control cycle to obtain the air temperature and relative humidity for the current control cycle. The preset state statistical period can be set to 1 to 3 minutes before the end of the control cycle; in this embodiment, it is set to 2 minutes. Using multiple sampled values ​​at the end of the control cycle to determine the current state can suppress random fluctuations in a single sampling while ensuring that the obtained greenhouse environmental state data closely approximates the actual environmental state at the current control decision time.

[0031] The crop canopy status data includes crop canopy temperature. Crop canopy temperature is acquired using a non-contact infrared temperature sensor. In this embodiment, two infrared temperature sensors are installed within an independent control zone, each facing a different location within the control zone's crop canopy. The infrared temperature sensors are installed 0.8 to 1.5 meters above the top of the canopy, with the angle between the measurement direction and the vertical direction set to 20 to 45 degrees, ensuring that the infrared temperature sensor's field of view primarily covers the crop leaves.

[0032] Furthermore, the area occupied by soil, supports, heating pipes, and irrigation pipes within the measurement field of view of the infrared temperature sensor does not exceed 10%. The emissivity parameter of the infrared temperature sensor is set to 0.95 to 0.99 based on the radiation characteristics of crop leaf surfaces; in this embodiment, it is set to 0.98. By limiting the measurement field of view and emissivity parameters, the influence of deviations in the emissivity settings for non-leaf targets and leaf surfaces on the canopy temperature measurement results can be reduced.

[0033] Two infrared temperature sensors output canopy temperature samples at the same sampling interval as the air temperature and humidity sensors. The canopy temperature samples are then checked for measurement range and rate of change. If the canopy temperature sample exceeds -20°C to 80°C, it is marked as an abnormal sample. Similarly, if the canopy temperature change between two adjacent sampling times exceeds a preset canopy temperature change threshold, and the corresponding air temperature change does not exceed a preset air temperature change threshold, the canopy temperature sample is also marked as an abnormal sample.

[0034] In this embodiment, the preset canopy temperature change threshold is set to 5 degrees Celsius, and the preset air temperature change threshold is set to 1 degree Celsius. These are used to identify abnormal fluctuations caused by foreign objects in the field of view, water droplet reflections, or short-term measurement interference. After removing abnormal sampled values, the effective canopy temperature sampled values ​​output by the two infrared temperature sensors within the preset state statistical period are averaged to obtain the crop canopy temperature for the current control cycle. By measuring different canopy locations and averaging over time, measurement deviations caused by single leaf orientation, differences in canopy density, and local differences in solar irradiance can be reduced.

[0035] The outdoor meteorological data includes outdoor temperature, outdoor relative humidity, solar irradiance, and outdoor wind speed. Outdoor temperature and relative humidity are collected by outdoor temperature and humidity sensors installed inside the ventilation louvers on the outside of the greenhouse; solar irradiance is collected by a total radiation sensor installed in an unobstructed location, measured in watts per square meter; outdoor wind speed is collected by a wind speed sensor installed above the greenhouse roof or in an unobstructed location on the windward side of the greenhouse, measured in meters per second. A preset distance is maintained between the outdoor meteorological sensors and the greenhouse enclosure structure to reduce the impact of heat storage, ventilation, and shading from the enclosure structure on the meteorological measurement results. In this embodiment, the outdoor temperature and humidity sensors and the wind speed sensor are at least 2 meters horizontally from the outer wall of the greenhouse, and there are no structures higher than their installation position that continuously obstruct the total radiation sensor.

[0036] The outdoor temperature and humidity sensor, total radiation sensor, and wind speed sensor use the same sampling time reference as the sensors inside the greenhouse. The outdoor meteorological sampling values ​​are checked for measurement range and communication status, and data exceeding the measurement range of the corresponding sensor or with communication error indicators are discarded. The effective sampling values ​​within the preset statistical period before the end of the current control cycle are averaged to obtain the outdoor temperature, outdoor relative humidity, solar irradiance, and outdoor wind speed for the current control cycle.

[0037] It should be noted that the outdoor meteorological data obtained in step S1 is used to characterize the external boundary conditions and external disturbances of the greenhouse during the current control cycle. The prediction sequence of the outdoor meteorological data used in subsequent steps within the preset prediction time domain can be obtained from external meteorological forecast services, or it can be determined by a time series prediction model based on the current outdoor meteorological data and outdoor meteorological data from previous control cycles.

[0038] The environmental control equipment operation data includes the current control values ​​and historical control value sequences for each piece of equipment. The current control values ​​for each piece of equipment are preferentially read from the equipment driver, position feedback unit, speed feedback unit, flow meter, power detection interface, or valve position feedback interface to obtain the actual execution status of the corresponding equipment. For example, the current control value of a ventilation window is expressed as a percentage of the actual opening degree relative to the maximum opening degree; the current control value of a circulating fan is expressed as a percentage of the actual speed relative to the rated speed; the current control value of a heating device is expressed as a percentage of the actual output power or the actual opening degree of a heating valve relative to its corresponding maximum value; the current control value of a misting humidifier is expressed as a percentage of the actual flow rate or output power relative to its corresponding maximum value; and the current control value of a shading device is expressed as a percentage of the actual unfolded length of the shading curtain relative to its maximum unfolded length.

[0039] When the environmental control equipment is not configured with an actual status feedback interface, the corresponding control command value is read from the control register of the greenhouse controller or programmable logic controller, and this control command value is used as the current control quantity. Simultaneously, a command substitution identifier is added to this current control quantity to distinguish it from the actual control quantity obtained from the equipment feedback interface. When subsequently extracting historical control response samples, data with actual control quantity feedback can be prioritized; when using data with command substitution identifiers, the sample reliability can be reduced or the corresponding operating condition matching requirements can be increased.

[0040] The control quantities of each environmental control device are normalized to a range of 0% to 100%, where 0% represents the device at its minimum permissible operating state and 100% represents its maximum permissible operating state. When an environmental control device only has start-stop control states, the control quantity corresponding to the off state is set to 0%, and the control quantity corresponding to the on state is set to 100%. When an environmental control device has multiple discrete operating levels, each operating level is mapped to a control quantity range of 0% to 100% based on its actual output capacity. For example, a circulating fan with low, medium, and high operating levels can be mapped to 30%, 60%, and 100%, respectively. By unifying the expression of control quantities, it is easier to determine the range of changes in device control quantities and generate candidate environmental control action sequences. However, the actual regulation capacity between different devices is still distinguished by the environmental regulation contribution per unit control quantity.

[0041] The historical control quantity sequence is formed by arranging the actual control quantities or control command values ​​with command substitution identifiers corresponding to multiple sampling times of each environmental control device before the current control cycle in chronological order. Each control quantity record includes at least the device identifier, sampling time, control quantity, operating status, and data validity marker. Subsequently, by analyzing the differences in control quantities between adjacent sampling times, the time of control quantity change, the magnitude of control quantity change, and the time when the control quantity recovers to the pre-change level are determined, without requiring the above-mentioned change time and recovery time to be included in the original control quantity record in advance.

[0042] The response observation window covered by the historical control sequence is no shorter than the estimated longest duration of action for each environmental control device. The estimated longest duration of action includes the possible start-up delay, rise time, hold time, and residual duration after the control quantity changes. This duration can be predetermined based on equipment commissioning tests, equipment technical parameters, or historical calibration results of similar equipment. In this embodiment, the response observation window is set to 60 minutes to cover the main environmental response process after a single control change for ventilation windows, circulating fans, heating equipment, atomizing humidifiers, and shading devices.

[0043] In addition to the aforementioned response observation window, a historical control response sample storage window is also established to save historical control response samples formed under different greenhouse operating conditions. The historical control response sample storage window can be set to the most recent 7 to 30 days, or it can save a preset number of recent valid historical control response samples. This embodiment saves valid historical control response samples formed within the most recent 14 days to take into account diurnal environmental differences and meteorological changes within a certain range. Data exceeding the storage window can be archived and saved, but it will not participate in the priority operating condition matching for the current cycle.

[0044] The historical control feedback data includes historical actual canopy environment trajectories and historical prediction error data. The historical actual canopy environment trajectories are sequences of actual leaf vapor pressure deficit and leaf dew margin values ​​determined based on actual collected air temperature, relative humidity, and crop canopy temperature during previous control cycles. Each historical actual canopy environment trajectory is associated and stored with the corresponding control cycle identifier, sampling time, outdoor meteorological data, and control values ​​of various environmental regulation devices to support subsequent extraction of historical control response samples according to current operating conditions.

[0045] The specific method for determining the leaf vapor pressure deficit and leaf dew condensation margin is further explained in step S2. In step S1, the actual value sequence of leaf vapor pressure deficit and the actual value sequence of leaf dew condensation margin formed according to a uniform sampling time are stored as historical actual canopy environment trajectories, thereby avoiding repeated reading and processing of raw sensor data during subsequent equipment response identification.

[0046] The historical prediction error data is obtained by comparing the predicted canopy environment trajectory of previous control cycles with the corresponding historical actual canopy environment trajectory. For any historical control cycle and any prediction time, the predicted leaf vapor pressure deficit is subtracted from the actual leaf vapor pressure deficit at that prediction time to obtain the leaf vapor pressure deficit prediction error; the predicted leaf dew condensation margin is subtracted from the actual leaf dew condensation margin at that prediction time to obtain the leaf dew condensation margin prediction error. A positive prediction error indicates that the actual value is higher than the predicted value, and a negative prediction error indicates that the actual value is lower than the predicted value.

[0047] Furthermore, historical prediction error data are saved separately according to different prediction lead times. For example, when the control cycle is 5 minutes and the preset prediction time domain is 30 minutes, the prediction errors for leaf vapor pressure deficit and leaf condensation margin corresponding to 5 minutes, 10 minutes, 15 minutes, 20 minutes, 25 minutes, and 30 minutes in advance are saved separately. By using the method of saving data according to prediction lead time, subsequent steps can determine the prediction error range for different prediction times, avoiding the distortion of the trajectory range caused by mixing prediction errors with those of short and long lead times.

[0048] When the system is first started, or when the number of valid historical prediction errors corresponding to any prediction lead time is lower than a preset error sample number threshold, the corresponding preset initial error boundary is used as the historical prediction error boundary parameter for that prediction lead time. The preset initial error boundary is determined based on model validation data before the greenhouse is put into operation; when model validation data is lacking, it is set based on the sensor comprehensive measurement error and the initial validation error of the prediction model.

[0049] In this embodiment, the preset error sample number threshold is set to 60 valid historical prediction errors corresponding to each prediction lead time. When the number of valid historical prediction errors corresponding to a certain prediction lead time reaches the threshold, the error boundary is updated using the 5th percentile and 95th percentile of the actual historical prediction error corresponding to that prediction lead time; prediction lead times that do not reach the threshold continue to use the corresponding preset initial error boundary. Step S2 forms the canopy environment trajectory interval based on the historical prediction error boundary parameters corresponding to each prediction lead time.

[0050] Furthermore, the greenhouse environmental status data, crop canopy status data, outdoor meteorological data, environmental regulation equipment operation data, and historical control feedback data are synchronized in time. Each sensor, greenhouse controller, equipment driver, and data storage device is time-synchronized via network time synchronization or the greenhouse controller's internal clock, ensuring that the allowable deviation between timestamps from different data sources does not exceed a preset time synchronization deviation threshold. In this embodiment, the preset time synchronization deviation threshold is set to 2 seconds, which is less than one-tenth of the 30-second sampling interval, reducing the probability of incorrect correspondence between changes in equipment control quantities and canopy environmental responses.

[0051] Various types of sampled data are resampled to a unified time series according to a preset sampling interval. When the number of consecutively missing sampling points does not exceed the preset number of interpolation points, linear interpolation is performed using the valid data before and after the missing sampling points; when the number of consecutively missing sampling points exceeds the preset number of interpolation points, the corresponding data segment is marked as an invalid data segment. In this embodiment, the preset number of interpolation points is set to 2, that is, linear interpolation is performed when the consecutive missing time does not exceed 60 seconds, and no interpolation is performed when the consecutive missing time exceeds 60 seconds. Furthermore, the corresponding data segment is not used for historical control response sample extraction, equipment action timing correction, and unit control quantity environmental adjustment contribution correction.

[0052] It should also be noted that the five types of data obtained in step S1 have different uses in subsequent steps. Greenhouse environmental status data and crop canopy status data are used to determine the current leaf vapor pressure deficit, leaf dew margin, and leaf transpiration-dew bilateral safety domain; outdoor meteorological data are used to characterize the boundary conditions of external disturbances to the greenhouse and canopy environmental trajectory prediction; environmental regulation equipment operation data are used to identify the current equipment control status, control variable change process, and historical control actions; historical actual canopy environmental trajectories are used to extract the actual environmental response corresponding to the environmental regulation equipment control actions; and historical prediction error data are used to determine the uncertainty range of the canopy environmental trajectory prediction results at different prediction times.

[0053] By performing multi-point acquisition, outlier removal, spatial fusion, control variable normalization, time synchronization, missing value processing, and historical data classification and storage on the above data, sensor environmental data, equipment control data, and historical prediction feedback data can be correlated according to a unified time axis. This reduces the impact of local measurement point anomalies, inconsistencies between the actual execution state of the equipment and the control commands, and time misalignment of multi-source data on subsequent canopy environment trajectory interval prediction and environmental regulation equipment response identification.

[0054] S2: Determine the leaf transpiration-condensation bilateral safety domain based on the greenhouse environmental status data and crop canopy status data. Based on the greenhouse environmental status data, crop canopy status data, outdoor meteorological data, and historical control feedback data, predict the canopy environment trajectory range to maintain the current equipment control state, and determine the leaf safety boundary crossing time window.

[0055] Determining the bilateral security domain includes: The actual vapor pressure and dew point temperature of the air are determined based on the air temperature and relative humidity, and the saturated vapor pressure of the canopy is determined based on the crop canopy temperature. The difference between the saturated vapor pressure of the canopy and the actual vapor pressure of the air is taken as the leaf vapor pressure deficit, and the difference between the crop canopy temperature and the air dew point temperature is taken as the leaf dew margin. The range of conditions in which the vapor pressure deficit of the blades is not lower than a preset lower limit and not higher than a preset upper limit, and the condensation margin of the blades is not lower than a preset lower limit, is defined as the bilateral safety domain.

[0056] Predicting the canopy environment trajectory range and determining the crossing window includes: Acquire the prediction sequence of outdoor meteorological data within the preset prediction time domain, and combine it with the greenhouse status, crop canopy status, current control values ​​of each device and historical actual trajectory to predict the baseline trajectory of leaf vapor pressure deficit and leaf dew margin to maintain the current control state. Based on historical prediction errors, the error range for each prediction time is determined, forming the upper and lower bound trajectories of the two indicators, thus obtaining the canopy environment trajectory range. The earliest possible crossing time of each safety boundary is defined as the moment when the lower boundary of the leaf vapor pressure deficit first falls below its lower limit, the upper boundary first rises above its upper limit, and the lower boundary of the leaf condensation margin first falls below its lower limit. The moment when the upper limit of leaf vapor pressure deficit first falls below its lower limit, the moment when the lower limit first rises above its upper limit, and the moment when the upper limit of leaf condensation margin first falls below its lower limit are taken as the latest possible crossing time of each safety boundary. The earliest time among the earliest possible crossing times corresponding to each safety boundary with predicted crossing is taken as the start time of the blade safety boundary crossing time window, and the latest time among the latest possible crossing times is taken as the end time of the blade safety boundary crossing time window.

[0057] In this embodiment, step S2 is used to determine the leaf transpiration-dew bilateral safety domain, which simultaneously constrains leaf transpiration status and leaf dew risk, based on the air temperature, relative humidity, and crop canopy temperature of the current control cycle; and, while maintaining the current control values ​​of each environmental regulation device unchanged, to predict the possible range of variation of leaf vapor pressure deficit and leaf dew margin within a preset prediction time domain, and to determine the leaf safety boundary crossing time window based on the relationship between the possible range of variation and each safety boundary in the leaf transpiration-dew bilateral safety domain. Step S2 includes two parts: determining the leaf transpiration-dew bilateral safety domain, and determining the canopy environmental trajectory interval and the leaf safety boundary crossing time window.

[0058] Part 1: Determining the bilateral safety domain of leaf transpiration and condensation.

[0059] Based on the air temperature, relative humidity, and crop canopy temperature obtained in step S1, determine the actual air vapor pressure, air dew point temperature, and canopy saturated vapor pressure, respectively.

[0060] Let the air temperature in the current control cycle be... The unit is degrees Celsius; the relative humidity is... ,use to The percentage is expressed as a percentage; crop canopy temperature is The unit is degrees Celsius. The air temperature is... relative humidity and crop canopy temperature All are state values ​​obtained by fusing the valid sampled values ​​within the preset state statistical period before the end of the current control cycle in step S1.

[0061] The partial pressure of water vapor at which water vapor reaches saturation at a given temperature is determined using the saturated vapor pressure function. This embodiment uses the following formula to calculate the temperature. Corresponding saturated vapor pressure: ; In the formula, Indicates temperature The corresponding saturated vapor pressure, in kilopascals; Represents an exponential function with the natural constant as its base; and The coefficient is calculated empirically for saturated water vapor pressure, where The unit is Celsius. This embodiment is below zero. Celsius The above-mentioned saturated water vapor pressure function is used within a temperature range of degrees Celsius.

[0062] air temperature Substituting the saturated water vapor pressure function, we obtain the saturated water vapor pressure of air: ; In the formula, This represents the saturated vapor pressure of the air corresponding to the current air temperature, in kilopascals.

[0063] crop canopy temperature Substituting the saturated vapor pressure function, we obtain the canopy saturated vapor pressure: ; In the formula, This represents the canopy saturated vapor pressure corresponding to the current crop canopy temperature, expressed in kilopascals (kPa). Under approximate conditions where the water vapor inside the crop leaves is nearly saturated, the canopy saturated vapor pressure is used to characterize the water vapor pressure level on the crop leaf side.

[0064] Convert relative humidity of air to The relative humidity ratio between the two: ; In the formula, This represents the relative humidity of the air and is a dimensionless value. For example, when the relative humidity of the air... for At that time, the relative humidity ratio of the air It is 0.80.

[0065] Determine the actual water vapor pressure of the air based on the relative humidity ratio and the saturated water vapor pressure of the air: ; In the formula, This represents the actual water vapor pressure of the air during the current control period, expressed in kilopascals. It is used to characterize the actual water vapor partial pressure present in the greenhouse air at the moment.

[0066] Further calculate the air dew point temperature based on the actual air vapor pressure. First, determine the variables for dew point calculation: ; Then, determine the air dew point temperature based on the aforementioned dew point calculation variables: ; In the formula, This represents the variable used in dew point calculation; it is a dimensionless numerical value. Represents the natural logarithm function; This indicates the air dew point temperature, expressed in degrees Celsius. The air dew point temperature represents the temperature at which air reaches saturation when cooled, provided the water vapor content in the air remains constant.

[0067] It should be noted that when the relative humidity data cannot meet the dew point temperature calculation conditions due to communication failure, measurement exceeding limits, or data loss, the corresponding data will be marked as invalid data according to the data validity check rules in step S1, and this data will not be used to determine the blade transpiration-condensation bilateral safety domain for the current control cycle. For relative humidity data that has passed the validity check in step S1 but whose value is close to zero, a preset minimum calculation value can be used for numerical protection to avoid the input value of the function being zero; the preset minimum calculation value is only used to ensure the continuity of numerical calculation and does not change the stored value of the original relative humidity data.

[0068] Leaf vapor pressure deficit is determined based on the difference between canopy saturated vapor pressure and actual air vapor pressure. ; In the formula, This indicates the leaf vapor pressure deficit during the current control period, expressed in kilopascals (kPa). Leaf vapor pressure deficit characterizes the vapor pressure gradient between the crop leaf side and the surrounding air. An increased leaf vapor pressure deficit indicates a stronger driving force for the release of water vapor from the leaves into the surrounding air; when the leaf vapor pressure deficit is below a suitable range, the driving force for crop transpiration weakens; when the leaf vapor pressure deficit is above a suitable range, the rate of water loss from the leaves may exceed the rate of water supply from the roots to the canopy.

[0069] Determine leaf dew margin based on the difference between crop canopy temperature and air dew point temperature: ; In the formula, This indicates the blade condensation margin for the current control cycle, expressed in degrees Celsius. When When the temperature is greater than zero, the crop canopy temperature is higher than the air dew point temperature; when When the temperature approaches zero, the blade surface approaches the conditions for water vapor condensation; when When the temperature is below zero, the crop canopy temperature is lower than the air dew point temperature, and there is a risk of condensation forming on the leaf surface.

[0070] Obtain the preset lower limit of vapor pressure deficit in the blade gas. Preset upper limit of vapor pressure deficit in leaf gas and the preset lower limit of leaf condensation margin .in, and The unit is kilopascal (kPa). The unit is Celsius. The above thresholds are predetermined based on crop type, growth stage, diurnal time, and cultivation management objectives.

[0071] This embodiment takes the daytime period of greenhouse-grown tomatoes during the fruiting stage as an example, and sets a preset lower limit for leaf vapor pressure deficit. Set to 0.5 kPa, and set the preset upper limit of vapor pressure deficit for the leaf gas. Set to 1.5 kPa, and set the preset lower limit of leaf condensation allowance. The value is set to 1.0 degrees Celsius; this value is used to illustrate the specific implementation of this embodiment. In other embodiments, a correspondence can be established between crop type, growth stage, day / night time, and various safety boundary thresholds, and the threshold corresponding to the current crop state can be read at the beginning of the current control cycle. For example, for crops with high leaf moisture sensitivity, the preset lower limit of leaf dew margin can be set to 1.5 to 2.0 degrees Celsius; when the crop's transpiration demand decreases at night, the preset lower limit of leaf vapor pressure deficit can be lowered relative to the daytime period.

[0072] The range of canopy environmental states that simultaneously meet the following constraints is defined as the leaf transpiration-condensation bilateral safety zone: ; In the formula, This indicates the bilateral safety domain of leaf transpiration and condensation; This represents the canopy environment state, determined by both leaf vapor pressure deficit and leaf condensation margin; symbol " "" indicates that the following conditions are constraints that need to be met when the canopy environment is within the leaf transpiration-condensation bilateral safety domain.

[0073] It should be noted that the term "bilateral" refers to the joint constraint on the canopy environmental state from two risk dimensions: leaf transpiration safety and leaf condensation safety. Leaf transpiration safety includes two safety boundaries: the lower limit of leaf vapor pressure deficit and the upper limit of leaf vapor pressure deficit. Leaf condensation safety includes one safety boundary: the lower limit of leaf condensation margin.

[0074] Judging the actual thermal and humidity state of crop leaves solely based on independent thresholds for air temperature or relative humidity cannot directly reflect the actual thermal and humidity state of the crop leaves. For example, when the air temperature is within a preset range, the crop canopy may still approach the air dew point temperature due to long-wave radiation heat dissipation; and under the same relative humidity, the leaf vapor pressure deficit corresponding to different crop canopy temperatures will also be different. By constructing the aforementioned leaf transpiration-dew double-sided safety domain, three types of risks can be identified simultaneously: insufficient leaf transpiration driving force, excessive leaf transpiration driving force, and insufficient leaf dew margin, reducing the possibility of overlooking actual risks to the crop canopy when controlling solely based on air environment parameters.

[0075] The second part determines the canopy environment trajectory range and the blade safety boundary crossing time window.

[0076] After determining the bilateral safety domain of leaf transpiration and condensation, the prediction sequence of outdoor meteorological data within the preset prediction time domain is obtained, and the leaf vapor pressure deficit baseline trajectory and the leaf condensation margin baseline trajectory are predicted while keeping the current control values ​​of each environmental regulation device unchanged.

[0077] Let the current control cycle be denoted as the _ ... Each control cycle, the current control decision moment is recorded as... The length of the prediction time extending into the future from the current control decision moment is denoted as the preset prediction time domain. The prediction step size is denoted as The number of prediction steps within the preset prediction time domain is denoted as ,but: ; This embodiment will preset the prediction time domain. Setting the prediction step to 30 minutes will adjust the prediction time. Set to 5 minutes, therefore predict the number of steps. It is 6. Each predicted time point is represented as: ; In the formula, Take 1 to 1 in sequence .

[0078] The preset prediction time domain is determined based on the longest estimated start-up delay and rise time of the environmental control equipment, and is set to be no shorter than the sum of the start-up delay and rise time. In this embodiment, the sum of the longest estimated start-up delay and rise time of each environmental control equipment does not exceed 25 minutes. Therefore, the preset prediction time domain is set to 30 minutes to ensure that after predicting the risk of crossing the safety boundary, it is possible to evaluate whether the environmental control equipment can promptly form an environmental correction effect within the preset prediction time domain.

[0079] The outdoor meteorological data prediction sequence within the preset prediction time domain includes the predicted outdoor temperature, outdoor relative humidity, solar irradiance, and outdoor wind speed for each prediction time. This prediction sequence can be provided by an external weather forecast service and mapped to each prediction time through linear interpolation. When the external weather forecast service is temporarily unavailable, the current outdoor temperature, current outdoor relative humidity, and current outdoor wind speed are used as the continuous prediction values ​​for the corresponding meteorological elements within the preset prediction time domain. The future solar irradiance is linearly extrapolated based on the solar irradiance variation trend of the three most recent valid sampling times, and the extrapolation result is limited to not being lower than zero.

[0080] Let the first The outdoor weather forecast vector for each forecast time is: ; In the formula, Indicates the first The predicted outdoor temperature at each predicted time, in degrees Celsius; Indicates the first Predicted outdoor relative humidity at each forecast time, in percentage; Indicates the first Predicted solar irradiance at each predicted time, in watts per square meter; Indicates the first The predicted outdoor wind speed at each forecast time is in meters per second; the superscript " " indicates vector transpose.

[0081] The first [value] is determined based on the predicted outdoor temperature and relative humidity. Actual outdoor water vapor pressure at each predicted time: ; In the formula, Indicates the first The actual outdoor water vapor pressure at the predicted time, in kilopascals.

[0082] Let the current control variables of each environmental control device constitute the device control vector: ; In the formula, Indicates the current control value of the ventilation window; This indicates the current control quantity of the circulating fan; This indicates the current control quantity of the heating equipment; This indicates the current control quantity of the atomizing humidifier; This indicates the current control value of the shading device. Each control value is converted from 0% to 100% in step S1 to a ratio between 0 and 1.

[0083] When predicting the canopy environment trajectory while maintaining the current control state of the environmental regulation equipment, the control variables of each environmental regulation equipment are kept at their current control variables throughout the entire preset prediction time domain, i.e.: ; This means that no new environmental control actions are added during the baseline trajectory prediction process; only the changes in the greenhouse environment state are predicted when the current equipment control state continues to operate.

[0084] This embodiment uses a greenhouse thermal and humidity state gray box model to predict future air temperature, actual air vapor pressure, and crop canopy temperature. The gray box model adopts a state recursive structure that can reflect the direction of greenhouse heat exchange, water vapor exchange, and canopy thermal response, and identifies the equivalent parameters within it through historical operating data.

[0085] The first The greenhouse state vector at each predicted time point is represented as: In the formula, Indicates the first The greenhouse air temperature at a predicted moment, in degrees Celsius; Indicates the first The actual water vapor pressure of greenhouse air at each predicted moment, in kilopascals; Indicates the first The crop canopy temperature at a predicted time, in degrees Celsius.

[0086] The air temperature during the current control cycle Actual water vapor pressure in air and crop canopy temperature As the initial state of the gray box model: ; The first [measurement] is determined based on the current control value of the ventilation window and the predicted outdoor wind speed. Equivalent ventilation rate at each predicted time: ; In the formula, Indicates the first The equivalent ventilation rate at each predicted time point, in hours; This represents the equivalent basic ventilation rate formed by gaps in the greenhouse enclosure structure and basic ventilation when the ventilation windows are closed, expressed in hours. This indicates the basic air exchange rate corresponding to a unit control quantity of the ventilation window when the outdoor wind speed is zero, in units of one hour. This represents the correction factor for the ventilation window's air exchange rate corresponding to a unit outdoor wind speed, and its unit is... The results are per hour.

[0087] It should be noted that the circulating fan is used to promote air circulation inside the greenhouse and the exchange of heat and moisture between the canopy and the surrounding air, but it does not directly cause the exchange of air between the inside of the greenhouse and the outside air. Therefore, the current control quantity of the circulating fan is not directly included in the equivalent ventilation rate, but is used to correct the exchange coefficient between the canopy and the air.

[0088] Determine the solar irradiance transmittance coefficient based on the current control parameters of the shading equipment: ; In the formula, This represents the solar irradiance transmittance of the shading device under the current control parameters, and is a dimensionless value. This indicates the maximum irradiance reduction achieved by the shading device from fully retracted to fully extended, and meets the following requirements: Even when the shading device is fully extended, it still retains some solar irradiance transmittance; therefore, the solar irradiance transmittance coefficient is not set to zero. In this embodiment, the solar irradiance transmittance ratio when the shading device is fully extended is 30%, corresponding to... Set it to 0.70.

[0089] No. The effective solar irradiance entering the greenhouse and participating in changes in greenhouse air and crop canopy thermal state at the predicted time is: ; In the formula, Indicates the first The effective solar irradiance at a predicted time is expressed in watts per square meter.

[0090] Determine the equivalent sensible heat exchange coefficient between the canopy and greenhouse air based on the current control parameters of the circulating fan: ; ; In the formula, The coefficient representing the equivalent effect of changes in canopy temperature on changes in air temperature, expressed in units of one hour. The equivalent effect coefficient of air temperature change on canopy temperature change, expressed in units of hourly; and These represent the basic exchange coefficients when the circulating fan is in its minimum operating state; and These represent the exchange enhancement coefficients corresponding to the unit control quantity of the circulating fan.

[0091] Based on the water vapor pressure difference between the blade side and the greenhouse air, the current control value of the circulating fan, and the effective solar irradiance, determine the first Equivalent canopy transpiration rate at each predicted time point: ; In the formula, This represents the equivalent growth rate of actual water vapor pressure in greenhouse air caused by canopy transpiration, expressed in kilopascals per hour. This represents the equivalent canopy transpiration coefficient corresponding to the current crop type and growth stage, in units of hourly. This represents the dimensionless enhancement coefficient of the exchange of water vapor in the canopy boundary layer caused by the circulating fan; symbol This indicates taking the larger value between the number inside the parentheses and zero; This represents the upper limit of the leaf vapor pressure deficit used to limit the input range of the equivalent transpiration model, in kilopascals. The dimensionless correction factor representing the effect of solar irradiance on the equivalent canopy transpiration; This represents the normalized constant of solar irradiance, expressed in watts per square meter.

[0092] This embodiment will Set as Tiles per square meter, Set as kPa. By setting an upper limit for the leaf vapor pressure deficit, it is possible to avoid predicting canopy transpiration based on an infinite linear growth relationship when the leaf vapor pressure deficit is high and the crop may be undergoing stomatal regulation.

[0093] It should also be noted that the equivalent canopy transpiration model is used to describe the combined effects of canopy transpiration on changes in air vapor pressure within the current crop species, growth stage, and prediction time domain of this embodiment, and is not used as an independent crop physiological mechanism model. The equivalent canopy transpiration coefficient... It comprehensively reflects crop canopy density, leaf area level, growth stage and stomatal response characteristics, and is calibrated using historical operating data of the current greenhouse.

[0094] Based on the heat exchange of the greenhouse enclosure structure, ventilation heat exchange, sensible heat exchange between the canopy and the air, solar radiation heat gain, the function of heating equipment, and the evaporative cooling effect of the atomizing humidification equipment, the greenhouse air temperature is recursively determined: ; In the formula, the predicted step size is used during the recursive calculation. Convert to hours; This represents the air temperature change coefficient corresponding to a unit temperature difference between the inside and outside of the greenhouse enclosure structure, expressed in hours. This represents the air temperature growth coefficient corresponding to a unit of effective solar irradiance, and its unit is... The result is in degrees Celsius per hour; This indicates the air temperature rise rate corresponding to a unit of control quantity of the heating equipment, expressed in degrees Celsius per hour. This indicates the rate at which the air cools down due to water evaporation under a unit of controlled volume, expressed in degrees Celsius per hour.

[0095] Based on ventilation and water vapor exchange, canopy transpiration, and water vapor input from atomizing humidification equipment, the actual water vapor pressure of the greenhouse air is recursively determined: ; In the formula, This indicates the actual air vapor pressure growth rate corresponding to a unit of control quantity of the atomizing humidifier, expressed in kilopascals per hour.

[0096] It should be noted that a strict greenhouse moisture mass balance can be achieved using air humidity ratio or water vapor density as state variables. In this embodiment, to reduce the number of model states and computational load, the actual air vapor pressure is used as the moisture state variable, and the aforementioned recursive relationship is taken as an equivalent dynamic relationship near the current normal operating range of the greenhouse. Historical operating data is used to identify the parameters corresponding to the ventilation rate, canopy transpiration, and atomizing humidification equipment, ensuring that these parameters comprehensively reflect the combined effects of greenhouse volume, air temperature changes, and water vapor mass changes on the actual air vapor pressure. Therefore, the aforementioned recursive relationship for actual air vapor pressure belongs to the gray-box approximation model applicable to the control range of this embodiment, rather than a moisture mass conservation formula unrestricted by operating conditions.

[0097] The crop canopy temperature is determined recursively based on the absorption of solar radiation by the canopy, the sensible heat exchange between the canopy and the air, the heat loss due to canopy transpiration, and the cooling effect of atomizing humidification equipment. ; In the formula, This represents the canopy warming coefficient corresponding to a unit of effective solar irradiance, with units of... The result is in degrees Celsius per hour; The canopy cooling coefficient corresponding to a unit canopy transpiration vapor pressure growth rate is expressed in degrees Celsius per kilopascal. This indicates the equivalent cooling rate of the canopy produced by the atomizing humidifier under a unit control quantity, expressed in degrees Celsius per hour.

[0098] The parameters of the gray box model mentioned above are divided into fixed equipment operating parameters and updatable environmental response parameters.

[0099] The fixed equipment operating parameters include the irradiance reduction ratio of the shading equipment, the air temperature rise rate per unit control quantity of the heating equipment, the air temperature drop rate per unit control quantity of the atomizing humidification equipment, the water vapor pressure rise rate per unit control quantity of the atomizing humidification equipment, and the canopy temperature drop rate per unit control quantity of the atomizing humidification equipment. These fixed equipment operating parameters are determined through a step test of the equipment before the greenhouse is put into operation and remain unchanged as long as the equipment is not replaced and its rated operating capacity does not change significantly.

[0100] The updatable environmental response parameters include the basic ventilation rate, ventilation window ventilation coefficient, outdoor wind speed ventilation correction coefficient, greenhouse envelope heat transfer coefficient, sensible heat exchange coefficient between the canopy and the air, equivalent canopy transpiration coefficient, circulating fan exchange enhancement coefficient, solar irradiation air warming coefficient, solar irradiation canopy warming coefficient, and solar irradiation transpiration correction coefficient. These updatable environmental response parameters are periodically updated based on historical actual canopy environmental trajectories and their associated stored outdoor meteorological data and environmental regulation equipment control parameters.

[0101] Let the above-mentioned fixed equipment operating parameters constitute a set of fixed parameters. The above-mentioned updatable environmental response parameters constitute the updatable parameter set. Based on the aforementioned state recursion formula, construct a single-step state prediction mapping: ; In the formula, Indicates the first The measured greenhouse state vector at each historical sampling moment; This represents the control vector of the environmental conditioning equipment at the corresponding moment; Represents the outdoor weather vector at the corresponding time; This represents a single-step state prediction mapping composed of recursive formulas for air temperature, actual air vapor pressure, and canopy temperature.

[0102] According to the preset model update cycle, using the most recent preset number of valid historical samples, the updatable environmental response parameters are updated using constrained least squares. ; In the formula, This represents the updated set of updatable environmental response parameters; This represents a set of parameter constraints used to limit each updatable environmental response parameter to be within its corresponding preset value range and to restrict the magnitude of a single change of each parameter relative to the parameter before the update. This indicates the number of valid historical samples used for this model parameter update; This represents the state error normalization matrix, used to eliminate dimensional differences between air temperature, actual air vapor pressure, and crop canopy temperature. This represents the constraint coefficient for parameter variation; This represents the set of updatable environmental response parameters before the update. This represents the L2 norm.

[0103] This embodiment sets the model parameter update cycle to 24 hours and the preset model sample quantity threshold to 120 valid historical samples. In addition to the number of valid historical samples reaching the preset model sample quantity threshold, the input variables corresponding to the parameters to be updated must have a change range no less than a preset identification change threshold. When updating the outdoor wind speed ventilation correction coefficient, the change range of the ventilation window control quantity in the historical samples must be no less than 20 percentage points, and the change range of the outdoor wind speed must be no less than 2 meters per second; when updating the circulating fan exchange enhancement coefficient, the change range of the circulating fan control quantity must be no less than 20 percentage points; when updating the solar irradiance air warming coefficient, solar irradiance canopy warming coefficient, or solar irradiance evaporation correction coefficient, the change range of solar irradiance intensity must be no less than 200 watts per square meter. Parameters that do not meet the corresponding identification change conditions retain their original values, and corresponding parameters are not forcibly updated using samples with insufficient input changes.

[0104] When the number of valid historical samples is insufficient, or when the range of change of the input variable corresponding to the parameter to be updated is lower than the preset identification change threshold, the original value of the corresponding parameter is maintained, and the parameter is not forcibly updated using data with insufficient input change. By processing the fixed equipment action parameters determined by the equipment step test separately from the environmental response parameters updated based on historical operating data, the parameter correlation and parameter drift caused by the simultaneous online identification of multiple equipment action parameters can be reduced.

[0105] The role of historical actual canopy environment trajectories in baseline trajectory prediction is to provide a basis for actual state response for gray box model parameter calibration and periodic updates; historical prediction error data is used to subsequently form the uncertainty interval corresponding to the baseline trajectory. These two factors act on the baseline model and the prediction interval respectively, without adding additional short-term trend corrections to the baseline trajectory, thus avoiding repeated compensation for the same historical deviation.

[0106] Using the updated gray box model, from the initial state Initially, the device control vector is maintained at each prediction time. And input the outdoor weather forecast vector. The air temperature at each predicted time is obtained step by step through recursion. Actual water vapor pressure in air and crop canopy temperature .

[0107] The air dew point temperature is determined directly based on the actual air vapor pressure at each predicted time. First, calculate: ; Confirm again: ; In the formula, Indicates the first Dew point calculation variables at each prediction time; Indicates the first The air dew point temperature at each predicted moment, in degrees Celsius. Since the gray box model directly outputs the actual air vapor pressure, the actual air vapor pressure is not converted to relative humidity before recalculating the air dew point temperature.

[0108] The baseline predicted value for leaf vapor pressure deficit is determined based on the crop canopy temperature and actual air vapor pressure at each prediction time: ; In the formula, Indicates the first The predicted baseline value of blade vapor pressure deficit at a given time point, assuming the current control state of the environmental regulation equipment; unit: kPa. This represents the baseline forecast.

[0109] The baseline predicted value for leaf dew margin is determined based on the crop canopy temperature and air dew point temperature at each prediction time: ; In the formula, Indicates the first The predicted blade condensation margin is the baseline prediction value for maintaining the current environmental regulation equipment control state at the predicted time, in degrees Celsius.

[0110] Arrange the predicted values ​​of the blade vapor pressure deficit baseline at each prediction time in chronological order to obtain the blade vapor pressure deficit baseline trajectory: ; Arrange the predicted leaf condensation margin baseline values ​​for each prediction time in chronological order to obtain the leaf condensation margin baseline trajectory: ; It should be noted that the gray box model described above is the baseline trajectory prediction method used in this embodiment. In other embodiments, state-space models, autoregressive models with external inputs, or lightweight neural network models trained with historical data can also be used, as long as the model can output the leaf vapor pressure deficit baseline trajectory and leaf dew margin baseline trajectory when maintaining the current environmental control state, based on the current greenhouse environment state, the current crop canopy state, the outdoor weather forecast sequence, the current control values ​​of each environmental control device, and the historical actual canopy environment trajectory.

[0111] Further, based on historical prediction error data, the prediction error intervals corresponding to each prediction time are determined. For the [previous prediction time]... For each prediction time, the prediction lead time is extracted from the historical prediction error data saved in step S1. The historical prediction errors of leaf vapor pressure deficit are used to form a sample set of leaf vapor pressure deficit errors: ; In the formula, Indicates the first In the first effective historical control cycle, the... The prediction error is obtained by subtracting the corresponding predicted value from the actual value of the vapor pressure deficit at each predicted time. Indicates the first The number of valid historical prediction errors corresponding to each prediction time.

[0112] The leaf vapor pressure deficit prediction error range is determined using the 5th and 95th percentiles of the aforementioned leaf vapor pressure deficit error sample set. ; In the formula, Indicates the first The prediction error range of leaf vapor pressure deficit at each prediction time; This represents the 5th percentile of the historical prediction error for leaf vapor pressure deficit. This represents the 95th percentile of the historical prediction error for leaf vapor pressure deficit.

[0113] Using the same method, according to the first The historical prediction error of leaf dew margin at each prediction time determines the prediction error range of leaf dew margin: ; In the formula, Indicates the first The prediction error range of leaf condensation margin at each prediction time; and These represent the 5th and 95th percentiles of the historical prediction error for leaf condensation margin, respectively.

[0114] Using the 5th and 95th percentiles allows the prediction error range to cover 90% of the middle range of historical effective prediction errors, reducing the excessive impact of individual abnormal prediction errors on the trajectory interval width. The percentile range used can be adjusted according to the risk tolerance level; when a higher level of conservatism is required, the lower percentile can be lowered and the upper percentile increased.

[0115] In this embodiment, the preset error sample size threshold is set to 60 valid historical prediction errors. When the... When the number of valid historical prediction errors corresponding to a given prediction time is less than 60, a preset initial error boundary determined by model validation data before greenhouse operation is used. In this embodiment, the initial lower and upper deviation boundaries for leaf vapor pressure deficit are set to -0.3 kPa and +0.3 kPa, respectively, and the initial lower and upper deviation boundaries for leaf dew margin are set to -0.5 degrees Celsius and +0.5 degrees Celsius, respectively. When the number of valid historical prediction errors reaches the preset error sample size threshold, the corresponding initial error boundaries are replaced by the 5th percentile and 95th percentile of the actual historical prediction errors.

[0116] Based on the baseline prediction value of the vapor pressure deficit and the corresponding prediction error range, determine the first... Lower and upper limits of leaf vapor pressure deficit at each predicted time point: ; ; In the formula, superscript Indicates the lower bound, superscript Indicates the upper bound.

[0117] Based on the baseline predicted value of leaf condensation margin and the corresponding prediction error range, determine the first... Lower and upper bounds of leaf dew margin at each predicted time point: ; ; By connecting the lower and upper limits of leaf vapor pressure deficit at each prediction time in chronological order, we obtain the lower and upper limits of leaf vapor pressure deficit trajectory. Similarly, by connecting the lower and upper limits of leaf dew condensation margin at each prediction time in chronological order, we obtain the lower and upper limits of leaf dew condensation margin trajectory. These four trajectories together form the canopy environment trajectory interval.

[0118] For example, if the baseline predicted value for the leaf vapor pressure deficit at the second prediction time is 1.05 kPa, and the 5th percentile of the historical prediction error at this prediction time is -0.12 kPa and the 95th percentile is +0.18 kPa, then the lower limit of the leaf vapor pressure deficit at this prediction time is 0.93 kPa, and the upper limit is 1.23 kPa. If the currently used preset upper limit for leaf vapor pressure deficit is 1.20 kPa, then although the baseline predicted value has not yet exceeded the preset upper limit, the upper limit of the leaf vapor pressure deficit has already exceeded the upper limit, indicating that there is a possibility that the leaf vapor pressure deficit at this prediction time exceeds the safety boundary.

[0119] Furthermore, based on the relationship between the canopy environment trajectory interval and each safety boundary in the leaf transpiration-condensation bilateral safety domain, the earliest possible crossing time and the latest possible crossing time of each safety boundary are determined respectively.

[0120] For the preset lower limit of vapor pressure deficit of the blade gas The trajectory of the lower boundary of the vapor pressure deficit in the leaf gas was lowered for the first time. The moment is determined as the earliest possible crossing time of this safety boundary, and the trajectory of the upper boundary of the leaf vapor pressure deficit is first lower than the threshold. The time is determined as the latest possible crossing time of the safety boundary.

[0121] For the preset upper limit of vapor pressure deficit of the blade gas The upper limit trajectory of the leaf vapor pressure deficit was higher than the target for the first time. The moment was determined as the earliest possible crossing time of this safety boundary, and the trajectory of the lower boundary of the leaf vapor pressure deficit was first higher than that of the boundary. The time is determined as the latest possible crossing time of the safety boundary.

[0122] For the preset lower limit of leaf condensation margin The trajectory of the lower limit of leaf dew condensation margin was lowered for the first time. The moment is determined as the earliest possible crossing time of this safety boundary, and the trajectory of the upper limit of the leaf dew margin is first lower than the threshold. The time is determined as the latest possible crossing time of the safety boundary.

[0123] The earliest possible crossing time is determined by the trajectory on the side facing the corresponding risk boundary within the prediction error interval, which is used to characterize the earliest possible time when the canopy environment state may be at risk within the historical prediction error range; the latest possible crossing time is determined by the trajectory on the side away from the corresponding risk boundary, which is used to characterize the latest possible time when the canopy environment state may be at risk within the prediction error range.

[0124] Since the canopy environment trajectory is formed by the trajectory values ​​corresponding to discrete prediction times, when the same upper or lower bound trajectory at two adjacent prediction times is located on both sides of the safety boundary, linear interpolation is used to determine the safety boundary crossing time. Let the adjacent prediction times be... and The corresponding trajectory values ​​are respectively and The safe boundary value to be crossed is Then travel through time for: In the formula, and This represents the values ​​of the same upper or lower bound trajectory at adjacent prediction times. When At that time, no linear interpolation is performed, and the discrete prediction time that first satisfies the crossing condition is determined as the crossing time.

[0125] If the trajectory used to determine the earliest possible crossing time does not cross the corresponding safety boundary within the preset prediction time domain, it is determined that there is no predicted crossing of the safety boundary.

[0126] When the trajectory used to determine the earliest possible crossing time has crossed the corresponding safety boundary, while the trajectory used to determine the latest possible crossing time has not crossed the corresponding safety boundary within the preset prediction time domain, the termination time of the preset prediction time domain is determined as the latest possible crossing time of the safety boundary, and a latest crossing time truncation marker is generated. The latest crossing time truncation marker indicates that: within the current prediction uncertainty range, there is a possibility that the corresponding safety boundary has been crossed, but it cannot yet be confirmed that the canopy environment trajectory interval has crossed the safety boundary entirely within the preset prediction time domain.

[0127] When the canopy environment trajectory range has exceeded a certain safety boundary at the current control decision time, the current control decision time is determined as the earliest possible crossing time of that safety boundary. When the canopy environment trajectory range is entirely outside the corresponding safety boundary at the current control decision time, the current control decision time is simultaneously determined as the latest possible crossing time of that safety boundary.

[0128] The earliest and latest possible crossing times corresponding to each predicted crossing safety boundary are collected. The earliest possible crossing time is determined as the start time of the blade safety boundary crossing time window, and the latest possible crossing time is determined as the end time of the blade safety boundary crossing time window. The blade safety boundary crossing time window is represented as follows: ; In the formula, Indicates the time window when the blade safety boundary crosses; Indicates the start time of the blade safety boundary crossing window; This indicates the end time of the blade safety boundary crossing window.

[0129] When none of the three safety boundaries—the preset lower limit of blade vapor pressure deficit, the preset upper limit of blade vapor pressure deficit, and the preset lower limit of blade condensation margin—are predicted to be crossed within the preset prediction time domain, the blade safety boundary crossing time window is marked as an empty time window, and a no-prediction-crossing flag is generated. Subsequent step S4 determines the environmental correction amount to zero based on the no-prediction-crossing flag and uses the action sequence that maintains the current control values ​​of each environmental regulation device as the baseline candidate environmental control action sequence.

[0130] It should also be noted that the leaf vapor pressure deficit baseline trajectory and the blade condensation margin baseline trajectory are point prediction results under the condition of maintaining the current environmental regulation equipment control state, while the canopy environmental trajectory interval further incorporates historical prediction errors corresponding to different prediction lead times. By judging the safety boundary crossing based on the canopy environmental trajectory interval rather than a single baseline trajectory, potential risks can be identified when the baseline prediction value has not yet crossed the safety boundary, but the prediction error range has already intersected with the safety boundary.

[0131] Furthermore, by constructing a blade safety boundary crossing window using the earliest and latest possible crossing times, the timeframe within which environmental risks may occur can be compared with the timeframe required for subsequent environmental control equipment to function effectively. Therefore, subsequent control processes no longer merely determine whether environmental indicators have exceeded the safety boundary, but further assess whether the environmental control equipment can provide sufficient environmental correction before the earliest possible occurrence of the risk, providing a time basis for determining control timing margin and effective regulation margin.

[0132] S3: Based on the operating data of the environmental regulation equipment and historical control feedback data, determine the timing of the action of each environmental regulation equipment on the index level of blade vapor pressure deficit and blade condensation margin under the current operating conditions, as well as the environmental regulation contribution per unit control quantity.

[0133] Based on the historical control quantity sequence of each device, determine the timing of control quantity changes, the magnitude of signed changes, and the direction of changes, and extract trajectory segments within the preset response observation period from the historical actual trajectory to form historical control response samples; For leaf vapor pressure deficit and leaf condensation margin, the starting time of the effect relative to the dynamic baseline trend before the change, the time when the effect reaches the preset effective response threshold, and the time when the effect subsides after the control quantity is restored are determined respectively. Based on the above time, the respective indicator-level action start delay, action rise time and residual action time after stopping of the two indicators are determined to form the indicator-level equipment action timing sequence. Based on the signed change amplitude and the effective change amount within the corresponding effective response statistical period, determine the environmental regulation contribution of each device to the unit control quantity of the two indicators; The time series and unit contribution of the index are saved according to the direction of change. The samples in the same direction are matched and weighted according to the difference between the current working condition and the working conditions of each sample to obtain the time series and unit contribution of the index under the current working condition.

[0134] The first part involves constructing historical control response samples; Based on the historical control quantity sequence of each environmental control device obtained in step S1, identify the control quantity change events of each environmental control device in chronological order.

[0135] Let the first The normalized control quantity sequence of each environmental control device is as follows: , among which, the The environmental control equipment includes ventilation windows, circulating fans, heating equipment, atomizing humidification equipment, or sunshade equipment. The range of values ​​is to The control variable in step S1 is When, the corresponding normalized control quantity is The control quantity is When, the corresponding normalized control quantity is .

[0136] For each sampling time in the historical control quantity sequence Calculate the median of the control variables during the preset control reference period prior to the sampling time: ; In the formula, Indicates the first An environmental control device at the sampling time Previous control baseline values; Indicates the length of the preset control reference time period; This indicates that the median is taken. This embodiment will preset a control reference time period. Set as minute.

[0137] When sampling time When the absolute difference between the corresponding environmental control equipment control quantity and the control quantity reference value is not lower than a preset candidate change threshold, and the deviation state is maintained continuously for no less than two sampling points, the first sampling time that meets the above conditions is determined as the candidate control quantity change time. In this embodiment, the preset candidate change threshold is set to 0.05, corresponding to the first 5 percentage points of normalization. Using two consecutive sampling points for confirmation can reduce erroneous identification caused by equipment feedback signal jitter and short-term fine-tuning of the controller.

[0138] Regarding the first The first environmental control equipment At what time do the candidate control variables change? The median of the control quantities within a preset control reference period prior to the change of the candidate control quantity is determined as the reference value of the control quantity before the change. ; Further, starting from the moment the candidate control variable changes, the search proceeds backward to determine the control stabilization period after the change. When the absolute value of the control variable difference between adjacent sampling points is continuously no higher than the preset control stabilization threshold, and the duration is no shorter than the preset stabilization confirmation time, the corresponding time period is determined as the control stabilization period after the change.

[0139] In this embodiment, the preset control stability threshold is set to 0.02, corresponding to the first 2 percentage points of normalization, and the preset stability confirmation time is set to 2 minutes. The search time for the control stability period after the change does not exceed the preset maximum mechanical action time of the corresponding environmental control equipment. The preset maximum mechanical action time for ventilation windows and sunshade equipment is set to 5 minutes, and the preset maximum mechanical action time for circulating fans, heating equipment, and atomizing humidification equipment is set to 2 minutes.

[0140] The median of the control quantity during the stable period after the change is determined as the level of the control quantity after the change: ; In the formula, Indicates the first The control stabilization period after the change of each candidate control variable event.

[0141] When the corresponding environmental control equipment is not configured with an actual control quantity feedback interface, the new control command value written by the controller at the moment of candidate control quantity change is used as the level of the changed control quantity, and a command substitution identifier is added to the corresponding historical control response sample. When the environmental control equipment with an actual feedback interface does not form a stable control period after the change within the preset maximum mechanical action time, the corresponding control quantity change event is marked as an incomplete execution event, and this event is not used to determine the environmental control contribution per unit control quantity.

[0142] The difference between the changed control level and the original control baseline value is defined as the control change magnitude. ; In the formula, Indicates the first The first environmental control equipment The magnitude of the control quantity change for each control quantity change event is a signed dimensionless value; a positive value indicates an increase in the control quantity, and a negative value indicates a decrease in the control quantity.

[0143] when When the corresponding candidate control variable change event is confirmed as a valid control variable change event, this embodiment will preset a minimum control variable change range. The value is set to 0.10, corresponding to the first 10 percentage points of normalization. When the change in control quantity is lower than the preset minimum change in control quantity, the corresponding event is not used as an independent historical control response sample, in order to reduce the impact of equipment mechanical clearance, controller fine-tuning, and measurement noise on the response identification results.

[0144] For environmental control equipment with only start-stop states, the control quantity change range from off to on is 1, and the control quantity change range from on to off is -1. For environmental control equipment with continuous adjustment capabilities and where the actual control quantity gradually reaches the target value, the control quantity change range formed by this action is determined by the control stabilization period after the change, avoiding splitting the same gradual execution process into multiple control quantity change events.

[0145] For each effective control variable change event, extract a segment of the canopy environment trajectory from the historical actual canopy environment trajectory, starting from the moment of control variable change and extending backward to a preset response observation period: ; In the formula, Indicates the first The first environmental control equipment Canopy environment trajectory segments corresponding to each effective control quantity change event; This indicates the actual value of the vapor pressure deficit in the leaf air; This indicates the actual value of the leaf condensation margin; This indicates the preset response observation period length.

[0146] The preset response observation period is determined based on the longest estimated start-up delay, rise time, duration of control action, and residual action time after cessation of the environmental control equipment. This embodiment uses the preset response observation period... Set to 60 minutes.

[0147] The environmental control equipment identification, the time of control quantity change, the baseline value of the control quantity before the change, the level of the control quantity after the change, the magnitude of the control quantity change, the direction of the control quantity change, the canopy environmental data before the change, the canopy environmental trajectory segment after the change, the corresponding outdoor meteorological data, and the control quantity records of other environmental control equipment are associated and stored to form a historical control response sample.

[0148] Further identifiability screening of historical control response samples is performed. Initial screening is conducted within a preset primary response identification period following the control variable change. Then, after determining the moment the action is achieved, a secondary verification is performed on the data between the control variable change and the moment the action is achieved. In this embodiment, the preset primary response identification period is set to 30 minutes.

[0149] When any of the following conditions exist between the time of control change and the time of action, the corresponding historical control response sample will be marked as a multi-factor coupled sample: 1. Except for the target environmental control equipment, the change in the control quantity of any other environmental control equipment exceeds 0.10; 2. The target environmental control equipment experiences another change in the control quantity with an absolute amplitude of not less than 0.05; 3. Outdoor temperature fluctuations exceed 3 degrees Celsius; 4. The outdoor relative humidity changes by more than 10 percentage points; 5. The variation in solar irradiance exceeds 250 watts per square meter; 6. Outdoor wind speed changes by more than 2 meters per second; 7. The proportion of invalid sampling points and interpolated sampling points to all sampling points in the corresponding data segment exceeds 10%. .

[0150] Historical control response samples not marked as multi-factor coupling samples are identified as priority identification samples. Multi-factor coupling samples are not used for the primary estimation of the environmental regulation contribution per unit control quantity. When the number of priority identification samples is insufficient, the equipment step calibration parameters before greenhouse operation are used as the initial parameters of the corresponding environmental regulation equipment, and complex environmental changes are not forcibly assigned to a single environmental regulation equipment in the absence of clear response attribution basis.

[0151] The second part is to determine the timing of equipment operation. For each priority identification sample, based on the changes in leaf vapor pressure deficit and leaf condensation margin during the preset control baseline period before the change of control quantity, a baseline change trend is constructed when the control quantity has not changed this time.

[0152] The linear baseline trend of vapor pressure deficit in the blades was determined using the least squares method: ; The linear baseline trend of leaf condensation margin was determined using the same method: ; In the formula, and These represent the leaf vapor pressure deficit baseline value and the leaf condensation margin baseline value corresponding to the time of change of the control quantity, respectively. and These represent the slopes of the changes in the two canopy environmental indicators before the change in the control quantity.

[0153] Corresponding trend identification thresholds are set for leaf vapor pressure deficit and leaf condensation margin, respectively. In this embodiment, the leaf vapor pressure deficit trend identification threshold is set as follows: ; Set the threshold for identifying leaf condensation margin trends to: ; When the statistical fluctuation of the fitted slope is large within the historical baseline period, the corresponding trend identification threshold is set to the larger of the above-mentioned engineering threshold and twice the standard deviation of the historical fitted slope.

[0154] When the absolute value of the slope of the corresponding environmental indicator before the change of the control quantity is lower than the preset trend identification threshold, the corresponding slope is set to zero, and the median of the corresponding environmental indicator before the change of the control quantity is used as a constant benchmark. When the absolute value of the slope reaches the preset trend identification threshold, a dynamic benchmark trend with an effective extrapolation duration limit is adopted.

[0155] Let the time interval after the change in the control quantity be . The effective duration of the preset trend is recorded as The duration of trend decay is denoted as This embodiment will Set to 20 minutes, Set to 10 minutes.

[0156] Taking the vapor pressure deficit of the leaf gas as an example, the trend of its dynamic baseline change is as follows: ; in .

[0157] Within the preset effective duration of the trend, the dynamic benchmark maintains the slope of change before the change in the control quantity; within the trend decay duration, the slope of change of the dynamic benchmark follows... The slope of change decreases linearly to zero; after exceeding the sum of the preset effective duration of the trend and the duration of trend decay, the dynamic benchmark remains at the value at the end of the decay. The dynamic benchmark change trend of blade condensation margin is determined in the same way.

[0158] Based on the residuals between the actual sampled values ​​before the control variable change and the corresponding baseline change trend, determine the standard deviation of the vapor pressure deficit residuals for each blade gas. and the standard deviation of the leaf condensation margin residual And determine the corresponding baseline fluctuation half-width: ; ; In the formula, and These represent the minimum fluctuation half-widths set considering both sensor overall error and normal short-time fluctuations. This embodiment will... Set to 0.05 kPa, Set to 0.2 degrees Celsius.

[0159] Therefore, the vapor pressure deficit and condensation margin of the leaves at the sampling time The corresponding preset benchmark fluctuation ranges are as follows: ; and: ; Further determine the response deviation of the canopy environment trajectory segment relative to the dynamic baseline trend: ; ; To determine the primary response direction and effective response magnitude corresponding to this control change, a reference statistical period for the effective response is established. This reference statistical period begins at the moment of the control change and terminates at the earliest of the following times: 1. The moment when the control quantity of the target environmental conditioning equipment first returns to the level of the control quantity before the change; 2. The moment when the target environmental control equipment experiences another effective change in the controlled quantity; 3. Preset the end time of the main response recognition period.

[0160] The effective response reference values ​​for each environmental indicator are determined based on the 90th and 10th percentiles of the response deviation within the effective response reference statistical period. For leaf vapor pressure deficit, the larger of the two percentiles is determined as the effective response reference value. For leaf condensation margin, the one with the larger absolute value among the two corresponding percentiles is determined as... The effective response reference value retains its positive or negative sign; a positive value indicates that the corresponding environmental indicator has increased relative to the dynamic baseline, and a negative value indicates that the corresponding environmental indicator has decreased relative to the dynamic baseline.

[0161] When satisfied At that time, the vapor pressure deficit of the leaf gas was determined as an effective response index in this sample.

[0162] When satisfied When the leaf condensation margin is determined, it is considered an effective response index for the sample.

[0163] This embodiment sets the minimum effective response threshold for leaf vapor pressure deficit. Set to 0.08 kPa, and set the minimum effective response threshold for leaf condensation margin. The temperature was set to 0.25 degrees Celsius. When neither of the two environmental indicators met the corresponding effective response conditions, it was considered that the change in the control quantity did not form a canopy environmental response that could be distinguished from normal fluctuations and stability, and this sample was not used to determine the equipment operating parameters.

[0164] The response direction of the corresponding environmental indicator is determined by the sign of the valid response reference value: ; ; In the formula, Represents a symbolic function.

[0165] The effective response index for blade vapor pressure deficit will meet the following requirements. The first sampling time of at least three consecutive sampling points is determined as the starting time of the leaf vapor pressure deficit. .

[0166] The effective response index for leaf condensation margin will meet the following requirements. The first sampling time of at least three consecutive sampling points is determined as the starting time of the leaf condensation margin effect. .

[0167] When both environmental indicators are valid response indicators, the earlier of the two action start times is determined as the universal action start time for this sample: ; When only one environmental indicator is a valid response indicator, the start time of action corresponding to that environmental indicator is determined as the general start time of action.

[0168] The adaptive effective response threshold is determined based on the minimum effective response threshold, the baseline fluctuation half-width, and the effective response reference value: ; ; In the formula, This represents the preset response ratio, which is set to 0.60 in this embodiment.

[0169] The search proceeds from the initial moment of the corresponding environmental indicator's effect to the moment when the effect is achieved. For the effective response index to leaf vapor pressure deficit, it will satisfy... The moment at which the vapor pressure deficit is achieved is determined by maintaining at least two sampling points continuously at the first sampling time. .

[0170] For the effective response index of leaf condensation margin, the same method was used to determine the time when the leaf condensation margin effect was achieved. .

[0171] When both environmental indicators are valid response indicators, the earlier of the two action achievement times is determined as the universal action achievement time for this sample: ; When only one environmental indicator is a valid response indicator, the time when the effect is achieved corresponding to that environmental indicator is determined as the general effect achievement time.

[0172] The time when the general effect is achieved is used to characterize the first stable and identifiable effect of the environmental regulation equipment on at least one canopy environmental indicator; the time when the leaf vapor pressure deficit effect is achieved and the time when the leaf dew margin effect is achieved are retained respectively, and are used in step S4 to determine the effective effect arrival time of the corresponding environmental indicator according to the specific environmental correction requirements.

[0173] Further, the control recovery time is determined from the historical control quantity sequence of the target environmental control equipment. The first sampling time at which the control quantity of the target environmental control equipment is restored to near the reference value of the control quantity before the change, and at least two sampling points are continuously maintained, is determined as the control recovery time. The control quantity recovery condition is: ; In the formula, This indicates the control quantity recovery tolerance, which is set to 0.05 in this embodiment.

[0174] After the control quantity recovery time, the first sampling time when the leaf vapor pressure deficit response deviation is restored to the corresponding preset reference fluctuation range and is continuously maintained at no less than three sampling points is determined as the time when the leaf vapor pressure deficit effect subsides; the first sampling time when the leaf condensation margin response deviation is restored to the corresponding preset reference fluctuation range and is continuously maintained at no less than three sampling points is determined as the time when the leaf condensation margin effect subsides.

[0175] When both environmental indicators are effective response indicators, the later of the two decay times is determined as the decay time for that sample. When only one environmental indicator is a valid response indicator, the time when the effect of the corresponding environmental indicator fades is determined as the time when the effect of the sample fades.

[0176] When the target environmental control equipment fails to recover to the pre-change control level within the preset response observation period, the corresponding sample is not used to determine the residual effect time after cessation. When the target environmental control equipment has recovered to the pre-change control level, but the canopy environmental trajectory has not recovered to the preset baseline fluctuation range within the preset response observation period, the corresponding sample is marked as a right-truncation sample of residual effect, and the termination time of the preset response observation period is not directly identified as the time of effect dissipation.

[0177] Based on the time of change of control quantity, the start time of general action, the time of achievement of general action, the time of recovery of control quantity, and the time of expiration of action, the following are determined respectively: ; ; ; In the formula, Indicates a delay in the activation of the function; Indicates the rise time of action; Indicates the residual effect time after cessation.

[0178] The device action timing sequence of this historical control response sample is composed of the above three time parameters: ; The timing of the device action refers to the actual time process by which changes in the control quantity of the environmental regulation device have an effect on the canopy environment, rather than the time required for the actuator of the environmental regulation device to complete the mechanical action.

[0179] In addition to the general equipment action sequence, for each environmental indicator that generates a valid response, the corresponding indicator-level equipment action sequence is retained. For environmental indicators... ,in Indicates a vapor pressure deficit in the leaves. Indicates the dew condensation margin on the leaves, determining the start-up delay of the indicator-level effect: ; Determine the rise time of indicator-level effects: ; Determine the residual duration of action after the indicator level stops: ; In the formula, , and These represent environmental indicators. The corresponding start time, effect completion time, and effect decay time; This indicates the time when the actual control quantity of the corresponding environmental regulation equipment recovers to near the control quantity before the change. The general equipment action sequence is used for the preliminary screening of candidate control actions and the description of the overall equipment action time; the index-level equipment action sequence is used in step S4 to calculate the formation process of the effect of candidate control actions on the leaf vapor pressure deficit and the leaf condensation margin.

[0180] Part Three: Determine the environmental regulation contribution of the unit control quantity; For each historical control response sample with an effective response, the environmental regulation contribution of unit control quantities for leaf vapor pressure deficit and leaf condensation margin is determined.

[0181] To address the leaf vapor pressure deficit, a preset effective response statistical period is set starting from the moment the leaf vapor pressure deficit is achieved; similarly, to address the leaf condensation margin, a preset effective response statistical period is set starting from the moment the leaf condensation margin is achieved. In this embodiment, the preset effective response statistical period is set to 5 minutes.

[0182] The actual effective response statistical period terminates at the earliest of the following times: 1. The moment when the control quantity of the corresponding environmental control equipment returns to the level before the change; 2. The moment when the environmental control equipment experiences another effective change in the controlled quantity; 3. The moment when the control quantity of other environmental control equipment changes beyond the preset disturbance threshold; 4. Preset the end time of the effective response statistics period.

[0183] When there are fewer than three valid sampling points within the truncated effective response statistical period, the unit control quantity environmental regulation contribution of the corresponding environmental indicator will not be used to calculate the environmental regulation contribution of the corresponding environmental indicator.

[0184] The median of the deviation in the blade vapor pressure deficit response during the actual effective response statistical period is determined as the effective change in blade vapor pressure deficit: ; The median of the leaf dew margin response deviation during the actual effective response statistical period is determined as the effective change in leaf dew margin: ; In the formula, and These represent the actual effective response statistical periods corresponding to leaf vapor pressure deficit and leaf condensation margin, respectively.

[0185] The environmental regulation contribution per unit of control quantity is determined based on the magnitude of the change in the control quantity and the effective change in the environmental indicators: ; ; In the formula, Indicates the first The environmental control equipment is in the first The contribution of a unit control quantity to the environmental regulation of the vapor pressure deficit of the leaf gas in a historical control response sample is expressed in kPa per unit normalized control quantity. This represents the environmental regulation contribution of the corresponding environmental regulation equipment to the blade condensation margin per unit of normalized control, expressed in degrees Celsius per unit of normalized control.

[0186] Since the magnitude of the control variable change is retained by sign, the environmental regulation contribution per unit of control variable can characterize the local incremental relationship between the direction of change of the control variable and the direction of change of the corresponding environmental indicator. A positive environmental regulation contribution per unit of control variable indicates that the direction of change of the environmental indicator is consistent with the direction of change of the control variable; a negative environmental regulation contribution per unit of control variable indicates that the direction of change of the environmental indicator is opposite to the direction of change of the control variable. The actual incremental effect corresponding to a candidate environmental control action is determined by multiplying the environmental regulation contribution per unit of control variable by the magnitude of the change of the signed control variable.

[0187] For example, when the ventilation window control value increases by 0.20 and the leaf vapor pressure deficit effectively increases by 0.10 kPa, the environmental regulation contribution of the unit control value of the leaf vapor pressure deficit is 0.50 kPa per unit normalized control value. That is, when the ventilation window opening increases by 10 percentage points, the corresponding increase in leaf vapor pressure deficit is about 0.05 kPa near this historical operating condition.

[0188] Further, according to the direction of control quantity change, the historical control response samples of each environmental control device are divided into control quantity increase samples and control quantity decrease samples, and the general equipment action time series, index-level equipment action time series and unit control quantity environmental control contribution of the historical control response samples corresponding to the two directions are saved respectively.

[0189] For the same environmental control equipment and the same direction of control change, the median of the action start-up delay, action rise time, residual action time after stopping, and the environmental control contribution per unit of control quantity for the two environmental indicators are taken from multiple valid historical control response samples to form the initial equipment action timing and initial unit control quantity environmental control contribution for the corresponding direction of control change. These initial parameters are used as the parameter baseline for the system's first operation, when the current operating condition matching sample is insufficient, and after the environmental control equipment has been replaced.

[0190] Part Four: Correcting the Current Operating Conditions; Based on the greenhouse environmental status data, crop canopy status data, outdoor meteorological data, and current control parameters of each environmental regulation device during the current control period, construct the current operating condition vector: ; In the formula, Indicates the current air temperature; This indicates the current actual water vapor pressure in the air; This indicates the current crop canopy temperature; Indicates the current outdoor temperature; This indicates the current actual outdoor water vapor pressure; Indicates the current solar irradiance; Indicates the current outdoor wind speed; to This represents the current normalized control value of each environmental control device; This indicates the number of environmental control devices.

[0191] Regarding the first The first environmental control equipment A historical control response sample is used to construct a historical sample operating condition vector based on the median of the corresponding data within a preset control baseline period before the change in the control quantity of the sample: ; To avoid the differences in dimensions and value ranges of variables under different operating conditions directly affecting the calculation of operating condition differences, each operating condition variable is normalized. Let the first variable be... The normalization scale for each operating condition variable is: The difference between the current operating condition and the historical sample operating conditions is: ; In the formula, Indicates the current operating condition and the first Dimensionless differences between the operating conditions corresponding to each historical control response sample; Indicates the number of operating condition variables; Indicates the first The difference weights of each working condition variable, and satisfying as well as .

[0192] Normalized scale The percentile is determined based on the difference between the 95th percentile and the 5th percentile of the corresponding variable within the most recent preset historical period. When the percentile difference is less than a preset minimum engineering scale, the preset minimum engineering scale is used as the normalization scale to avoid excessive amplification of variables with small long-term changes due to an excessively small denominator.

[0193] In this embodiment, the minimum engineering scale for air temperature, crop canopy temperature, and outdoor temperature is set to 10 degrees Celsius; the minimum engineering scale for actual air vapor pressure and actual outdoor vapor pressure is set to 1.5 kPa; the minimum engineering scale for solar irradiance is set to 500 W / m²; the minimum engineering scale for outdoor wind speed is set to 5 m / s; and the minimum engineering scale for the normalized control values ​​of each environmental regulation device is set to 1. The differential weights of each operating condition variable are determined based on the historical sensitivity of the corresponding variable to the device's start-up delay, rise time, and environmental regulation contribution per unit control value. When the number of historical samples is insufficient for sensitivity analysis, the total weight of the corresponding variables for indoor environmental state and crop canopy state is set to 0.4, the total weight of outdoor meteorological variables is set to 0.3, and the total weight of the current control values ​​of each environmental regulation device is set to 0.3, with the corresponding weights evenly distributed within the same category of variables.

[0194] For each environmental control device, historical control response samples matching the current operating condition are selected from both the control quantity increase and control quantity decrease samples. Historical control response samples in the same direction with operating condition differences not exceeding a preset operating condition matching threshold are determined as the current operating condition matching samples. In this embodiment, the preset operating condition matching threshold is set to 0.30.

[0195] Let the direction of change of the control quantity be... ,in Indicates the direction of increase in control quantity. Indicates the direction of decrease in the control quantity. The first [value / value] is determined based on operating condition differences and sample reliability. Weights of matching samples in the corresponding direction for each environmental control device: ; In the formula, Indicates the first Normalized weights of matching samples in the same direction; Indicates the first The environmental control equipment under current operating conditions and the direction of change of control variables The following is a set of matching samples; Indicates the reliability of the sample; This indicates the distance attenuation scale under operating conditions.

[0196] Prioritize the identification of samples based on the actual feedback control values ​​from the equipment and ensuring the data is complete. The reliability of these samples will be assessed. Set to 1; for priority identification samples formed based on control command values ​​with instruction substitution identifiers, set the sample confidence level to 0.8. This embodiment uses the operating condition distance attenuation scale. Set it to 0.20.

[0197] Based on the weights of the matched samples, determine the first... The start-up delay of the control variable in the direction of increase and decrease of control variable for an environmental control device under the current operating conditions: ; And the time it takes for the effect to rise: ; The residual action time after cessation is weighted only using matched samples with complete control recovery and action decay processes. After re-normalizing the weights in the corresponding complete sample set, the following is determined: ; The direction of change of control quantity under the current operating condition The corresponding start-up delay, rise time, and residual action time after shutdown constitute the device action timing sequence: ; Based on the matched samples with effective responses to the corresponding environmental indicators, determine the current operating condition and the direction of change of the control variables. The environmental regulation contribution of the unit control quantity: ; ; In the formula, Indicates the direction of change of control quantity The set of matched samples with effective leaf vapor pressure deficit response; Indicates the direction of change of control quantity The set of matched samples with effective leaf condensation margin response; and This represents the sample weights after renormalization in the corresponding set.

[0198] It consists of the environmental regulation contribution of the unit control quantity corresponding to two environmental indicators: ; In the formula, Indicates the first The environmental control equipment under current operating conditions and the direction of change of control variables The environmental regulation contribution vector of the unit control quantity.

[0199] Therefore, for each environmental control device, the following results were obtained: as well as The superscript "+" indicates the direction of increasing control quantity, and the superscript "−" indicates the direction of decreasing control quantity. In step S4, when generating candidate environmental control actions, the corresponding device action timing and unit control quantity environmental adjustment contribution are invoked based on the direction of control quantity change of the candidate action.

[0200] When the current control value of the environmental control equipment has reached its upper limit, the direction of increasing the control value of the corresponding equipment is marked as an unavailable direction; when the current control value of the environmental control equipment has reached its lower limit, the direction of decreasing the control value of the corresponding equipment is marked as an unavailable direction.

[0201] When the number of matching samples for a certain device under the current operating condition and a certain control variable change direction is less than 3, the preset operating condition matching threshold is gradually widened in increments of 0.05, up to a maximum of 0.45. If the number of matching samples is still less than 3 after widening, the initial device action timing and initial unit control variable environmental adjustment contribution for the corresponding device and the corresponding control variable change direction are used, and an insufficient operating condition correction data flag is generated.

[0202] The insufficient operating condition correction data indicator is used in subsequent steps to conservatively process the operating speed and adjustment capability of the corresponding equipment, avoiding overestimation of the equipment response speed and the environmental adjustment contribution per unit control quantity when historical response samples are insufficient.

[0203] Regarding environmental indicators , No. The direction of change of environmental control equipment and control quantities The current operating condition matching samples with valid responses to the corresponding environmental indicators will form an indicator-level matching sample set: ; The original current operating condition matching weights are renormalized within the set of matching samples at the index level: ; The activation delay of the current operating condition indicator level is determined based on the renormalized weights: ; Determine the rise time of the current operating condition indicator level: ; The residual effect time after cessation is determined only based on matched samples with complete control recovery and effect decay processes. These samples constitute the residual effect matched sample set: ; Within this set, the sample weights are renormalized to determine: ; When the number of matching samples at the corresponding indicator level is less than 3, the initial indicator-level equipment action timing sequence of the corresponding equipment, the direction of change of the corresponding control quantity, and the corresponding environmental indicator is adopted, and an indicator-level timing data shortage flag is generated.

[0204] Further retain the general equipment action timing, index-level equipment action timing, and sample-level unit control quantity environmental regulation contribution corresponding to each matched historical control response sample. The weighted parameters under the current operating condition are used to characterize the central response level of the environmental regulation equipment near the current operating condition; the sample-level parameters are used in step S4 to determine the effective action arrival time of the same candidate environmental control action under multiple historical response conditions, and to form an effective action arrival time window based on multiple effective action arrival times.

[0205] It should also be noted that step S3 does not directly infer the equipment's regulating capacity based on the rated power, theoretical air volume, or changes in control commands of the environmental regulating equipment. Instead, it determines the time process of the environmental regulating equipment's start-up, effective action formation, and action decay after the control quantity is restored, based on the actual deviation of the canopy environmental trajectory from the dynamic baseline after the control quantity changes. It also calculates the signed unit control quantity contribution of the two environmental indicators, leaf vapor pressure deficit and leaf condensation margin.

[0206] By determining the amplitude of control quantity changes based on the stable control period after changes, the underestimation of the amplitude of changes caused by slow-moving equipment not yet reaching the target control quantity can be reduced. By limiting the effective extrapolation duration of the dynamic baseline trend, the response deviation caused by the long-term extrapolation of short-term environmental trends can be reduced. By simultaneously using the baseline fluctuation half-width, absolute response threshold, and relative response ratio to determine the effective response, it can be ensured that the identified equipment action can be distinguished from normal environmental fluctuation stability. By separately determining the equipment action parameters in the direction of control quantity increase and decrease, the hysteresis and response asymmetry in the bidirectional control process of environmental regulation equipment can be reflected. Through current operating condition matching and distance-weighted correction, the equipment action parameters under a single historical operating condition can be avoided as fixed parameters directly used for different seasons and different operating states.

[0207] Therefore, the timing sequence of the current operating condition equipment output in step S3 is used in step S4 to determine when the candidate environmental control action begins to form a canopy environmental effect, when it forms an effective environmental effect, and the duration of the effect after it stops; the environmental regulation contribution of the unit control quantity is used to convert the change amplitude of the candidate control quantity into the leaf vapor pressure deficit correction quantity and the leaf dew margin correction quantity, providing a parameter basis for step S4 to generate the effective effect arrival window and predict the canopy environmental trajectory.

[0208] S4: Determine the environmental correction amount and generate candidate environmental control action sequences based on the deviation between the canopy environment trajectory interval and the bilateral safety domain. Construct historical response calculation scenarios based on the index-level time series and unit contribution, and determine the effective arrival time window, representative predicted canopy environment trajectory, and conservative predicted canopy environment trajectory envelope for each candidate sequence.

[0209] Based on the deviation of the canopy environmental trajectory interval relative to the bilateral safety domain during the correction demand assessment period starting from the start time of the crossing time window, the leaf vapor pressure deficit correction amount and the leaf dew margin correction amount formed no later than the start time are determined to constitute the environmental correction amount. Candidate environmental control action sequences are generated based on the current control variables and adjustable ranges of each device. Each sequence includes the action start time, signed change amplitude, and duration of each device. For each candidate sequence, sample-level index-level time series and unit contribution are selected from the matching samples of the devices participating in the action in the same direction, and combined to form a historical response calculation scenario; In each scenario, the cumulative effective correction amount of the two indicators is calculated based on the indicator-level time sequence and unit contribution of each device, and the arrival time to reach the corresponding required correction amount is determined. Based on this, the earliest and latest arrival times of the candidate sequence and the proportion of reachable scenarios are determined, forming an effective arrival window. Based on the baseline trajectory maintained under the current control state and the cumulative correction amount of each scenario, a scenario prediction trajectory is generated, and a representative predicted canopy environment trajectory is obtained by weighting according to the scenario weights. By combining the predicted trajectories for each scenario, the historical prediction error boundaries for different prediction lead times, the uncertainties of device interaction, and the extrapolation uncertainties of duration, a conservative prediction envelope for the canopy environment trajectory is formed.

[0210] In this embodiment, step S4 is used to determine the leaf vapor pressure deficit correction amount and leaf condensation margin correction amount that need to be formed before the start time of the leaf safety boundary crossing time window, based on the deviation between the canopy environment trajectory interval obtained in step S2 and the leaf transpiration-condensation bilateral safety domain; based on the current control quantity, adjustable range of the control quantity and the current operating condition equipment action parameters obtained in step S3, multiple candidate environmental control action sequences are generated; for each candidate environmental control action sequence, the cumulative effective environmental correction amount formed over time under multiple historical response conditions is calculated, the corresponding effective action arrival time window is determined, and the incremental response generated by the candidate control action relative to the current control state is superimposed on the baseline trajectory obtained in step S2 to obtain the corresponding predicted canopy environment trajectory.

[0211] Step S4 includes four parts: determining the environmental correction amount, generating candidate environmental control action sequences, determining the effective action arrival window, and generating the predicted canopy environmental trajectory.

[0212] Part One: Determine the Environmental Correction Amount; When the blade safety boundary crossing time window obtained in step S2 is empty, a no-environmental-correction-requirement flag is generated, and both the blade vapor pressure deficit correction and the blade condensation margin correction are set to zero. At this time, the action sequence that maintains the current control values ​​of each environmental regulation device is used as the baseline candidate environmental control action sequence.

[0213] When the blade safety boundary crossing time window is not empty, obtain the start time of the blade safety boundary crossing time window. And set a period for revising the requirements assessment: ; In the formula, Indicates the prediction step size; This indicates the preset prediction time domain. The revised demand assessment period is used to determine the degree of risk development within one prediction step after the earliest possible crossing of the blade safety boundary.

[0214] Set the safe retention amount for vapor pressure deficit. and leaf dew margin safety retention amount This embodiment will: ; When the start and end times of the revised demand assessment period are inconsistent with the discrete prediction time, the linear interpolation method in step S2 is used to determine the trajectory interval value at the corresponding time. In this embodiment, the maximum and minimum values ​​within the revised demand assessment period are searched according to an internal calculation time resolution of 1 minute.

[0215] The feasible lower bound of the blade vapor pressure deficit correction amount is determined based on the lower bound trajectory of the blade vapor pressure deficit during the correction demand assessment period: ; The feasible upper bound of the blade vapor pressure deficit correction amount is determined based on the upper bound trajectory of the blade vapor pressure deficit during the correction demand assessment period: ; when, When there exists a uniform incremental correction amount that ensures the entire blade vapor pressure deficit trajectory range within the assessment period of the correction requirement remains within a safe range after maintaining a safe distance, the value with the smallest absolute value within the feasible correction range is determined as the blade vapor pressure deficit correction amount. ; when This indicates that, during the correction demand assessment period, it is impossible to bring the entire vapor pressure deficit trajectory range within the preservative safe range by applying a uniform incremental correction. In this case, the following will be done: ; The compromise leaf vapor pressure deficit correction amount is determined, and an indicator is generated for the leaf vapor pressure deficit range that cannot be fully corrected.

[0216] The leaf dew margin correction amount is determined based on the lower bound trajectory of the leaf dew margin during the correction demand assessment period: ; The environmental correction for the current control cycle consists of the leaf vapor pressure deficit correction and the leaf condensation margin correction. ; The value of the environmental correction amount is determined based on the most unfavorable deviation of the canopy environmental trajectory interval relative to the safety boundary within the correction requirement assessment period, and represents the minimum incremental correction required for candidate environmental control actions no later than the start time of the blade safety boundary crossing time window. The correction requirement assessment period is only used to determine the magnitude of the environmental correction amount and does not change the processing method in step S5, which uses the start time of the blade safety boundary crossing time window as the evaluation benchmark for control timeliness and effective adjustment. Whether the candidate environmental control action sequence meets the safety constraints within the entire preset prediction time domain is determined based on its conservatively predicted canopy environmental trajectory envelope.

[0217] The second part involves generating candidate environmental control action sequences. Based on the current control values ​​of each environmental control device and the adjustable range of the control values ​​of each environmental control device obtained in step S1, multiple candidate environmental control action sequences are generated.

[0218] Let the first The current normalized control quantity of each environmental control device is: The lower limit and upper limit of the control quantity are as follows: ; In this embodiment, the normalized control quantity range of the continuously regulating device is typically 0 to 1; for devices with only start-stop states, the control quantity can be 0 or 1. When the device is under maintenance, undergoing fault derating, or subject to cultivation management constraints, the corresponding lower and upper limits of the control quantity are adjusted according to the device's current permissible operating range. The candidate environmental control action sequence is represented as follows: ; No. The environmental control equipment is in the first The action records in the candidate environmental control action sequence are as follows: ; In the formula, Indicates the moment the action begins; Indicates the magnitude of change in the signed control quantity; Indicates the duration of the control variable after a change. When When this occurs, it indicates that the corresponding device maintains its current control variable unchanged. The change range of the candidate control variable should satisfy: ; when When the control quantity obtained in step S3 is called, the timing of the directional device's action and the environmental regulation contribution of the unit control quantity are increased; when At that time, the device action parameters corresponding to the direction of decrease of the control quantity are invoked.

[0219] When the current control value of the environmental control equipment has reached the upper limit of the control value, no control value increase action is generated; when the current control value of the environmental control equipment has reached the lower limit of the control value, no control value decrease action is generated.

[0220] In this embodiment, the candidate control quantity change amplitude of the continuously adjusting device is generated in steps of 0.10, the action start time is generated with a scheduling resolution of 1 minute, and the control duration is generated as an integer multiple of 5 minutes. The action start time is located between the current control decision time and the preset prediction time domain termination time, and the control duration does not exceed the remaining time between the action start time and the preset prediction time domain termination time.

[0221] For equipment with only start-stop states, the control quantity change amplitude is either 1 or -1 depending on the current state. To avoid extrapolating the local unit control quantity environmental adjustment contribution obtained in step S3 to a control range lacking historical sample support, this embodiment preferentially limits the absolute value of the single candidate control quantity change amplitude of continuously adjusted equipment to no more than 0.30. When the equipment step calibration data or the current operating condition matching sample covers a larger control quantity change amplitude, the range of single control quantity change can be expanded to the corresponding calibration range.

[0222] When only one environmental indicator requires correction, the time of effect formation corresponding to that indicator is used; when both environmental indicators require correction, the larger of the two environmental indicator time of effect formation is used. The environmental control equipment changes in the direction of the control quantity. The latest reference start time is: ; In the formula, This represents the set of environmental indicators that require correction during the current control cycle. This indicates a vapor pressure deficit in the leaves. Indicates the condensation allowance on the leaves; It indicates the direction in which the control quantity increases or decreases.

[0223] The latest start time is used only to narrow down the search range of the initial candidate sequences and is not used as the final basis for judging whether the candidate environmental control action sequences can form environmental correction quantities in a timely manner.

[0224] Candidate environment control action sequences are generated in the following way: First, generate a baseline candidate environmental control action sequence that maintains the current control values ​​of each environmental control device unchanged.

[0225] Second, based on the direction of the environmental correction, select the equipment and control quantity change direction that contribute in the same direction to the required environmental correction from the unit control quantity environmental adjustment contribution obtained in step S3, and generate a directional candidate environmental control action sequence.

[0226] Third, generate representative candidate environmental control action sequences formed by single device actions, coordinated actions of two devices, and joint actions of the main correction device and the side effect compensation device.

[0227] Fourth, random perturbations are applied near the start time, control variable change amplitude, and control duration of the directional candidate environment control action sequence to form diverse candidate environment control action sequences.

[0228] We do not use a method that exhaustively enumerates all devices, all startup times, and all control variable changes, in order to avoid the number of candidate sequences increasing in combination with the number of devices and prediction steps.

[0229] Based on the environmental regulation contribution of the current operating condition unit control quantity obtained in step S3, a preliminary estimate is made of the potential stable incremental effect of the candidate environmental control action sequence: ; In the formula, Indicates the first The environmental control equipment is in the first The direction of change of control quantity in each candidate environmental control action sequence; when At that time, the gain effect of the device is not calculated.

[0230] The preliminary correction residuals are determined based on the difference between the estimated effect of the steady-state increment and the environmental correction amount: ; In the formula, This represents the normalized weight matrix of environmental indicators; This represents the norm. The normalized scale for the vapor pressure deficit dimension of leaf gas can be expressed as... The normalized scale for the dew margin dimension of the blades can be a preset dew margin engineering scale.

[0231] Prioritize retaining candidate environmental control action sequences that can reduce the initial correction residual, while also retaining some candidate environmental control action sequences with different side effects and different action costs, in order to avoid the initial candidate set being concentrated on a single device combination too early.

[0232] If the control duration of a candidate action is shorter than the start-up delay of the corresponding device under current operating conditions, and the device step calibration data does not indicate that a short-time pulse can form an identifiable response, the corresponding candidate action is deleted.

[0233] The initial candidate sequence set consists of multiple generated candidate environment control action sequences: ; In the formula, This represents the number of initial candidate environment control action sequences. Step S5 further performs swarm intelligence iterative optimization based on the initial candidate sequence set.

[0234] Part Three: Determine the effective action arrival time window; For each candidate environmental control action sequence, based on the current operating condition matching historical control response sample retained in step S3, the cumulative effective environmental correction amount generated by the candidate environmental control action sequence over time under different historical response conditions is calculated.

[0235] For each environmental control device participating in the action in the candidate environmental control action sequence, representative historical control response samples are selected from the current operating condition matching samples corresponding to the direction of change of the control quantity.

[0236] Regarding the first The candidate environmental control action sequence will be the first one. The first environmental control equipment The formation time of the conservative effect of a historical control response sample under the current correction requirement is determined as follows: ; In the formula, This represents the set of environmental indicators that require correction during the current control cycle; when only one environmental indicator requires correction, the time of action corresponding to that environmental indicator is used.

[0237] For each environmental control device involved in the action, priority is given to selecting historical control response samples with the highest sample weight, samples near the weighted 10th percentile, weighted 50th percentile, and weighted 90th percentile corresponding to the conservative action formation time. When multiple selection criteria point to the same historical control response sample, only one is retained; when there are fewer than 4 matching samples for the current operating condition, all matching samples for the current operating condition are retained.

[0238] The sample-level parameters selected from each participating device are combined to form multiple historical response calculation scenarios. Priority is given to combining historical control response samples corresponding to the same quantile level from each environmental control device into basic scenarios. Then, cross-combined scenarios are added according to their initial weights from largest to smallest until a preset upper limit for the number of scenarios is reached. In this embodiment, the upper limit for the number of historical response calculation scenarios is set to 20.

[0239] Let the first In the historical response calculation scenario, the first The historical control response sample number used by the participating action device is: The initial weights for this scenario are: ; In the formula, Indicates the first The set of environmental control devices that execute new control actions from a candidate sequence of environmental control actions; This represents the current operating condition matching weight of the corresponding environmental control equipment, the direction of change of the corresponding control quantity, and the corresponding historical control response sample; (symbol) This involves multiplying the sample weights corresponding to each environmental control device involved in the action.

[0240] Normalize the initial weights for each scenario: ; Obtain the scenario weights for each historical response calculation scenario.

[0241] For environmental indicators ,in This indicates a vapor pressure deficit in the leaves. Indicates the dew condensation margin of the leaves, let the first... The first candidate environmental control action sequence In the historical response calculation scenario, the first The start-up delay, rise time, and residual duration of the indicator-level effect for each environmental control device are as follows: ; Environmental indicators The corresponding start time of the action is determined as follows: ; No. The stopping time of each environmental control device in the candidate environmental control action sequence is:

[0242] The percentage of the index level action that has been formed when the equipment control action stops is: ; No. Environmental control equipment affects environmental indicators The normalized action coefficient is: ; In the formula, This indicates taking the larger value between the number inside the parentheses and zero; when The vapor pressure deficit coefficient of the blades was obtained at that time. ;when The action coefficient of leaf condensation margin was obtained at that time. The start-up delay, rise time, and residual effect time after shutdown of the same environmental control equipment are allowed to differ for two environmental indicators.

[0243] Based on the environmental regulation contribution of the unit control quantity for the vapor pressure deficit, the variation range of the candidate control quantity, and the vapor pressure deficit effect coefficient, the following is determined: ; Based on the environmental regulation contribution of the unit control quantity of leaf dew margin, the variation range of candidate control quantities, and the effect coefficient of leaf dew margin, the following is determined: ; The cumulative effective environmental correction amount represents the sum of the incremental environmental effects formed by each environmental control device participating in the action at the same time, and does not represent the time integration of the environmental correction amount.

[0244] When the vapor pressure deficit correction amount is not zero, the first moment that satisfies the following formula is determined as the [missing value]. Arrival time of blade vapor pressure deficit correction under a historical response calculation scenario: ; The arrival time is denoted as .

[0245] When the leaf dew margin correction is greater than zero, the first moment that satisfies the following formula is determined as the leaf dew margin correction arrival time: ; The arrival time is denoted as .

[0246] When both environmental indicators require correction, the later of the correction arrival times for the two environmental indicators is determined as the first. The arrival time of forming a complete environmental correction in a historical response calculation scenario: ; When only one environmental indicator requires correction, the time when the correction for that environmental indicator is reached is determined as the time when the complete environmental correction is reached.

[0247] The arrival time is searched according to a preset internal calculation time resolution. In this embodiment, the internal calculation time resolution is set to 1 minute. When the cumulative effective environmental correction amount first reaches the corresponding correction amount between two adjacent internal calculation times, linear interpolation is used to determine the arrival time.

[0248] When an environmental indicator that requires correction fails to reach the corresponding correction amount within the preset prediction time domain, the historical response calculation scenario is marked as an unreachable scenario.

[0249] Let the first The number of historical response calculation scenarios with complete arrival times for each candidate environmental control action sequence is: The total number of historical response calculation scenarios is The corresponding reachability ratio is: ; When all historical response calculation scenarios have complete arrival times, the earliest and latest arrival times for each scenario are determined as follows: ; ; And form an effective arrival window: ; When some historical response calculation scenarios are reachable while others are not, the earliest arrival time in the reachable scenarios is determined as the earliest arrival time, the termination time of the preset prediction time domain is determined as the latest arrival time, and an arrival window truncation flag is generated. ; In the formula, This represents the set of reachable historical response calculation scenarios. When all historical response calculation scenarios are unreachable, the corresponding candidate environmental control action sequence is marked as an unreachable candidate sequence for environmental correction, and no effective arrival window is generated for it.

[0250] The reachability ratio, arrival window cutoff flag, and unreachable flag for environmental correction are all used as inputs for the feasibility evaluation of candidate sequences in step S5. This process avoids ignoring historical scenarios with slow responses or small unit adjustment contributions by only counting reachable samples.

[0251] When the environmental correction is zero, the effective action arrival time window is not calculated, and the baseline candidate environmental control action sequence that maintains the current environmental regulation equipment control quantity is taken as the priority candidate sequence.

[0252] Part Four: Generating Predicted Canopy Environmental Trajectories; For each candidate environmental control action sequence, based on the leaf vapor pressure deficit baseline trajectory and leaf condensation margin baseline trajectory obtained in step S2 to maintain the current environmental regulation equipment control state, as well as the cumulative effective environmental correction amount formed by the candidate environmental control action sequence under each historical response calculation scenario, a corresponding predicted canopy environmental trajectory is generated.

[0253] Let the first The candidate environmental control action sequence in the first In the historical response calculation scenario, the first The predicted value of the vapor pressure deficit at each prediction time is: ; No. The predicted leaf condensation margin for each prediction time is: ; In the formula, and These represent the results obtained in step S2 while maintaining the current environmental regulation equipment control state. Baseline predicted values ​​for leaf vapor pressure deficit and leaf condensation margin at each prediction time; and They represent the first The candidate environmental control action sequence in the first In a historical response calculation scenario, the cumulative effective environmental correction amount for leaf vapor pressure deficit and the cumulative effective environmental correction amount for leaf condensation margin are relative to the conditions for maintaining the current control state.

[0254] The baseline trajectory obtained in step S2 already includes the environmental impact caused by the continuous action of the current control quantities of each environmental control device. Therefore, this part only superimposes the incremental effect of the candidate control action relative to the current control quantity on the baseline trajectory, and does not superimpose the environmental control contribution corresponding to the current control quantity of each environmental control device again.

[0255] Calculate the scenario weights based on the historical responses determined in the third part of step S4, and determine the first... The candidate environmental control action sequence in the first The vapor pressure deficit at each predicted time point represents the predicted value: ; Determine the leaf condensation margin to represent the predicted value: ; In the formula, Indicates the first The number of historical response calculation scenarios corresponding to each candidate environmental control action sequence; Indicates the first The normalized scenario weights for each historical response scenario are calculated, and satisfy the following conditions: ; as well as: ; Scenario weights for each historical response calculation scenario The initial weights are determined according to the scenario initial weight normalization method described in Part 3 of this step.

[0256] 1. Determine the equipment interaction uncertainty: When the effective action periods of two or more environmental control devices overlap in the candidate environmental control action sequence, determine the equipment interaction uncertainty based on the joint control response samples in the historical control feedback data.

[0257] The joint control response samples that have the same equipment combination, the same direction of change of control quantities of each environmental control device, and a difference of no more than 0.20 in the magnitude of change of control quantities of each environmental control device, and whose difference in pre-action operating conditions determined by the method in step S3 is no greater than 0.45, are identified as joint control response samples that are similar to the current candidate environmental control action sequence.

[0258] For the Based on a sample of similar joint control responses, environmental indicators were determined. Joint response residuals: ; In the formula, Indicates a vapor pressure deficit in the leaves. Indicates the condensation allowance on the leaves; Indicates the first Environmental indicators in a joint control response sample The actual incremental response relative to the corresponding dynamic baseline trend; Indicates according to the first The incremental response of environmental indicators is determined by the historical control response parameters of each individual environmental control device. This refers to the collection of environmental control equipment that participates in joint control.

[0259] When the number of similar joint control response samples is no less than 5, the 90th percentile of the maximum absolute value of the joint response residuals of each similar joint control response sample during the effective overlap period is determined as the environmental indicator. Corresponding device interaction uncertainty boundary: ; In the formula, Indicates the first The time period during which the effective effects of two or more environmental control devices overlap in a joint control response sample.

[0260] No. The candidate environmental control action sequence in the first The device interaction uncertainty in a historical response calculation scenario is: ; In the formula, Indicates the first The candidate environmental control action sequence in the first In a historical response calculation scenario, at time... An effective set of environmental control equipment has been formed; symbol When used with sets, it indicates the number of elements in the set.

[0261] When the number of similar joint control response samples is less than 5, the equipment interaction uncertainty is determined as 20% of the sum of the absolute values ​​of the incremental environmental actions corresponding to the environmental control equipment with overlapping effective actions: ; It also generates an uncertain identifier for device interaction.

[0262] At that moment When there is no overlap in the effective functions of two or more environmental control devices, the corresponding device interaction uncertainty is set to zero.

[0263] 2. Determine the duration extrapolation uncertainty: Based on the actual control duration in the historical control response sample matched with the current operating conditions, determine the coverage range of the historical duration of each environmental control device under the corresponding control variable change direction.

[0264] When the The environmental control equipment changes in the direction of the control quantity. When the number of historical control response samples matching the current operating condition is no less than 5, the 10th percentile and 90th percentile of the actual control duration corresponding to the historical control response samples are respectively determined as the lower limit and upper limit of historical duration coverage. ; ; In the formula, Indicates the first The actual control duration of each environmental control device corresponds to a historical control response sample.

[0265] When the number of historical control response samples matching the current operating condition is less than 5, the control duration range covered by the corresponding equipment step settling data is adopted; when there is no corresponding equipment step settling data, the minimum and maximum values ​​of the actual control duration in the existing valid historical control response samples are used to form the historical duration coverage range.

[0266] When the In the candidate environmental control action sequence, the first... The control duration of each environmental control device meets the following requirements: ; or: ; When this occurs, the environmental control device is marked as a duration extrapolation device, and a duration extrapolation identifier is generated.

[0267] No. The candidate environmental control action sequence in the first In a historical response calculation scenario, environmental indicators The extrapolation uncertainty of the duration is: ; In the formula, Indicates the first There is a set of environmental regulation devices with duration extrapolation among the candidate environmental control action sequences.

[0268] When there is continuous operation calibration data with corresponding environmental control equipment and corresponding control variable change direction, the duration extrapolation uncertainty is determined according to the 90th percentile of the absolute value of the error between the actual environmental response and the predicted environmental response in the calibration results, in order to replace the above-mentioned 15% preset extrapolation ratio.

[0269] When there is no duration extrapolation device in the candidate environmental control action sequence, the duration extrapolation uncertainty is set to zero.

[0270] 3. To form a conservative prediction canopy environment trajectory envelope, the conservative prediction canopy environment trajectory envelope corresponding to the candidate environmental control action sequence is determined by simultaneously considering the baseline prediction error, the differences in equipment action between historical response calculation scenarios, the uncertainty of equipment interaction, and the uncertainty of duration extrapolation.

[0271] No. The candidate environmental control action sequence in the first The lower bound for the predicted vapor pressure deficit at each predicted time point is: ; The upper limit for predicting vapor pressure deficit in leaf gas is: ; The lower bound for predicting leaf condensation margin is: ; The upper limit for predicting leaf condensation margin is: ; In the formula, and They respectively represent the first step in step S2. The lower quantile and upper quantile of the historical prediction error for the leaf vapor pressure deficit corresponding to each prediction lead time; and These represent the lower quantile and upper quantile of the historical prediction error for leaf condensation margin, respectively. and These represent the equipment interaction uncertainties corresponding to the blade vapor pressure deficit and the blade condensation margin, respectively. and These represent the extrapolated uncertainties in the duration corresponding to the leaf vapor pressure deficit and the leaf condensation margin, respectively.

[0272] When there is no device interaction uncertainty or duration extrapolation in the candidate environmental control action sequence, the corresponding uncertainty is set to zero.

[0273] The predicted values ​​of leaf vapor pressure deficit and leaf condensation margin at each prediction time are arranged in chronological order to form the [number]. The representative predicted canopy environment trajectory corresponding to each candidate environmental control action sequence; the lower bound, upper bound, lower bound, and upper bound of leaf vapor pressure deficit prediction, leaf dew caking margin prediction, and leaf dew caking margin prediction are connected in chronological order to form the first... Conservatively predicted canopy environment trajectory envelopes corresponding to candidate environmental control action sequences.

[0274] The representative predicted canopy environment trajectory is used to characterize the weighted center prediction results of the candidate environmental control action sequence under multiple historical response calculation scenarios; the conservative predicted canopy environment trajectory envelope comprehensively reflects the differences in equipment action between historical response calculation scenarios, the historical prediction error of the corresponding prediction lead time, the uncertainty of equipment interaction and the uncertainty of duration extrapolation, and serves as the basis for step S5 to determine whether the candidate environmental control action sequence satisfies the blade transpiration-condensation bilateral safety domain constraint.

[0275] 4. Insufficient processing equipment operating parameter data and extrapolation of control quantity range.

[0276] When a candidate environmental control action sequence contains an environmental regulation device with an insufficient generation condition correction data flag or an insufficient index-level time series data flag, when forming the conservative prediction canopy environmental trajectory envelope, the larger of the current matching sample parameter and the initial calibration parameter is used for the index-level action start delay and index-level action rise time corresponding to that environmental regulation device; for unit control quantity environmental regulation contributions that are beneficial to meeting the corresponding safety boundary, the smaller absolute value of the current matching sample parameter and the initial calibration parameter is used; for unit control quantity environmental regulation contributions that are not beneficial to meeting the corresponding safety boundary, the larger absolute value is used.

[0277] When the change in the control quantity of a candidate environmental control action sequence exceeds the local change range covered by the historical control response sample matching the current operating condition, the corresponding candidate environmental control action sequence is marked as an amplitude extrapolation candidate sequence.

[0278] For candidate sequences for amplitude extrapolation, the segmented unit control quantity environmental regulation contribution obtained from the equipment step calibration is used first for calculation. When the control quantity change amplitude of the corresponding environmental regulation equipment spans multiple calibration intervals, the incremental effect of the corresponding interval is calculated according to the unit control quantity environmental regulation contribution within each calibration interval, and the incremental effects of each interval are summed.

[0279] When no equipment step calibration data covering the corresponding control quantity change amplitude is available, the amplitude extrapolation identifier is retained, but the amplitude extrapolation candidate sequence is not considered as the final feasible candidate environmental control action sequence that satisfies all safety constraints. The amplitude extrapolation candidate sequence can still retain its representative predicted canopy environmental trajectory and conservative predicted canopy environmental trajectory envelope, which is used in step S5 to determine whether it is necessary to adjust the combination of environmental conditioning equipment, reduce the control quantity change amplitude of a single device, or expand the candidate action search range.

[0280] Therefore, step S4 outputs the environmental correction amount of the current control cycle, multiple candidate environmental control action sequences, the effective arrival time window corresponding to each candidate environmental control action sequence, the proportion of reachable scenarios, the representative predicted canopy environment trajectory, the conservative predicted canopy environment trajectory envelope, and the corresponding uncertainty identifier, providing a data basis for step S5 to calculate the control timing margin, effective adjustment margin, minimum normalized safety distance, and control action cost.

[0281] S5: The candidate sequences are iteratively updated using a swarm intelligence algorithm. In each iteration, the effective arrival time window, the representative predicted canopy environment trajectory, and the conservative predicted canopy environment trajectory envelope of each candidate environmental control action sequence are redefined. The control timing margin is determined based on the crossing time window and the arrival time window. The effective adjustment margin is determined based on the cumulative effective correction and environmental correction before the start time of the crossing time window. The strategy is evaluated by combining the conservative predicted envelope, timing margin, and adjustment margin. After the iteration terminates, the best candidate sequence is selected from those whose conservative envelope satisfies the safety domain constraint and whose timing margin and adjustment margin are not lower than the corresponding threshold. If there is no candidate sequence that satisfies the constraints, the emergency sequence is determined based on the degree of constraint violation.

[0282] Determining the control timing margin and effective regulation margin includes: The control timing margin is obtained by subtracting the latest arrival time of the arrival time window from the start time of the travel time window. When there is a need for correction of leaf vapor pressure deficit, the cumulative correction amount of leaf vapor pressure deficit at the beginning of each scenario is projected in the direction of the correction amount, and the minimum value of the projection result of each scenario is taken as the conservative effective correction amount of leaf vapor pressure deficit. When there is a need to correct the condensation margin, the minimum cumulative correction amount of the condensation margin at the beginning of each scenario is taken as the conservative effective correction amount of the condensation margin. The difference between each conservative effective correction and the corresponding required correction is divided by the larger of the absolute value of the required correction and the preset normalized protection value to obtain the normalized adjustment margin of the two indicators. The minimum value among the margins of indicators with correction needs is taken as the effective adjustment margin. When the environmental correction is zero, the timing margin and adjustment margin are not considered as feasibility constraints.

[0283] Iterative updates and policy evaluation using swarm intelligence algorithms include: Each candidate sequence is treated as an individual, and the start time, change range, and duration of each device's action are used as control variables. Each iteration updates variables based on information interaction between individuals, limiting the magnitude of change to an adjustable range and limiting the timing and duration of actions to the prediction time domain; The scenario is reconstructed for each updated sequence, and the arrival window, reachability ratio, representative trajectory and conservative envelope, as well as the temporal margin and adjustment margin are redefined. The minimum normalized safety distance is determined based on the distance between the conservative envelope and each safety boundary. After normalizing the temporal margin, adjustment margin and control action cost, the policy evaluation result is determined, where the control action cost is determined based on the change magnitude, duration and number of switching. After the iteration terminates, the best strategy evaluation sequence is selected from the candidate sequences that satisfy the safety domain, time margin and adjustment margin not lower than the threshold and meet the accessibility constraint and equipment operation constraint. If no sequence satisfies all constraints, an emergency sequence is determined based on the degree of violation of safety, timing, adjustment, and reachability constraints, and an identifier for no fully feasible control strategy and a result of constraint violation are generated.

[0284] In this embodiment, step S5 is used to treat the multiple candidate environmental control action sequences generated in step S4 as candidate individuals in the swarm, and to use a swarm intelligence algorithm to iteratively update the action start time, control quantity change amplitude, and control duration corresponding to each candidate individual. In each iteration, for each candidate environmental control action sequence after iterative update, the effective action arrival window, reachable scene ratio, predicted canopy environment trajectory, and conservative predicted canopy environment trajectory envelope are re-determined according to the method in step S4. The control timing margin is determined based on the blade safety boundary crossing time window and the re-determined effective action arrival time window. The effective regulation margin is determined based on the cumulative effective environmental correction amount and environmental correction amount formed by the candidate environmental control action sequence before the start time of the blade safety boundary crossing time window. The strategy evaluation result is formed based on the conservative predicted canopy environment trajectory envelope, control timing margin, effective regulation margin, and control action cost. After satisfying the preset iteration termination condition, the preferred environmental control action sequence is determined from the candidate environmental control action sequences that satisfy the blade transpiration-condensation bilateral safety domain constraint, control timing margin constraint, and effective regulation margin constraint, based on the strategy evaluation result.

[0285] Step S5 includes four parts: candidate individual encoding and group iterative update, determination of control time margin and effective adjustment margin, generation of policy evaluation results, and determination of iterative termination and optimal environmental control action sequence.

[0286] The first part involves candidate individual coding and group iterative updates. Each candidate environmental control action sequence generated in step S4 is treated as a candidate individual in the population.

[0287] Let the number of candidate individuals in the population be... , No. The candidate individuals in the th The control variable vector in the round of iteration is: ; In the formula, Indicates the number of environmental control equipment; Indicates the first Normalized variables for the start-up time of each environmental control device; Indicates the first The amplitude of the signed control quantity change of each environmental control device; Indicates the first The normalized variable for the duration of control of an environmental regulation device.

[0288] The normalized variable at the action initiation time is: ; The normalized variable for controlling duration is: ; in ,and .

[0289] when When, it indicates the first The environmental control device does not perform any new control actions and sets the corresponding control duration to zero.

[0290] In this embodiment, the initial candidate environmental control action sequences obtained in step S4 are used as the initial population. When the number of initial candidate environmental control action sequences is greater than the preset population size, stratified selection is performed according to the preliminary corrected residual, equipment combination type, and control action cost; when the number of initial candidate environmental control action sequences is less than the preset population size, random perturbations are applied to the action start time, control quantity change amplitude, and control duration near the directional candidate environmental control action sequences to supplement and form a preset number of candidate individuals. In this embodiment, the population size... Set it to 30.

[0291] This embodiment uses the particle swarm optimization algorithm as a specific implementation of the swarm intelligence algorithm. A velocity vector with the same dimension as the control variable vector is set for each candidate individual. In the first In each iteration, the velocity of a candidate individual is updated based on its own historical best position, the group's historical best position, and its current velocity. ; In the formula, Indicates as of the date During the first iteration The historical optimal control variable vector of each candidate individual; Indicates as of the date The vector of historically optimal control variables in the population during round iteration; Indicates the first The inertia weight of each iteration; and These represent the individual learning coefficient and the group learning coefficient, respectively. and Indicates that each component is taken independently to A random vector of random numbers between; symbol This indicates that corresponding components are multiplied.

[0292] The inertia weight decreases linearly with the number of iterations as follows: ; In the formula, and These represent the upper and lower limits of the inertia weight, respectively; This indicates the maximum number of iteration rounds. This embodiment will... Set to 0.9, Setting it to 0.4 will adjust the individual learning coefficient. Group learning coefficient All are set to 1.5.

[0293] To avoid excessively large updates in a single cycle, the components of the velocity vector are limited to their corresponding preset velocity ranges. The absolute values ​​of the velocities corresponding to the normalized variables at the start of the action and the normalized variables for the control duration do not exceed 0.20, and the absolute values ​​of the velocities corresponding to the changes in the control quantity do not exceed 0.30.

[0294] The temporary control variable vector for candidate individuals is: ; Since the candidate environment control action sequence contains both continuously updated variables and discrete execution variables, the control range projection, discretization, and equipment constraint correction are performed on the temporary control variable vector.

[0295] Regarding the first For a continuously regulating device, a set of allowable control variable variation ranges is constructed based on the device's current control variable and allowable control range: ; In the formula, Indicates the first The set of control quantity variation ranges that a continuous control device is allowed to use in the current control cycle; This indicates the distance between the change in the control quantity and the step size; in this embodiment, it is set to 0.10.

[0296] Project the temporary control variable change range to the nearest value in the set of permissible control variable change ranges: ; When the range of change of the temporary control quantity is the same as the distance between zero and a certain non-zero discrete value, zero should be selected first to avoid unnecessary equipment actions when the evaluation benefits are the same.

[0297] For environmental control equipment that only has start-stop states, the changed control quantity is projected to 0 or 1, and the change range of the control quantity is determined based on the difference between the projected changed control quantity and the current control quantity.

[0298] The action initiation time is projected onto the interval between the current control decision time and the preset prediction time domain termination time, and discretized according to a scheduling resolution of 1 minute: ; In the formula, This indicates the action scheduling resolution, which is set to 1 minute in this embodiment; This indicates taking the nearest integer.

[0299] The control duration is limited to the remaining time between the action start time and the preset prediction time domain termination time, and discretized according to a duration resolution of 5 minutes: ; In the formula, This indicates the control of the duration resolution; in this embodiment, it is set to 5 minutes.

[0300] When the change amplitude of the control quantity is not zero, the control duration is not shorter than the preset minimum control duration, and the following conditions are met: ; In this embodiment, the preset minimum control duration is set to 5 minutes.

[0301] When the change in the control quantity is zero, the action initiation time is set to the current control decision time, and the control duration is set to zero.

[0302] When the duration of the discretized control is shorter than the start-up delay of the corresponding device in the current control direction, and there is no historical control response sample or device calibration data that can prove that the short-term control action can form an identifiable canopy environment response, the change amplitude of the corresponding control quantity will be corrected to zero.

[0303] Each candidate environmental control action sequence is further subjected to equipment safety operation constraint checks, including equipment availability status, maximum control variable change rate, minimum continuous running time, minimum downtime, and preset mutually exclusive operation constraints. When a candidate action violates the equipment safety operation constraints, the corresponding variable is preferentially projected to the nearest feasible value; if it cannot be repaired by projection, the corresponding candidate individual is marked as an individual that violates the equipment constraints.

[0304] When the change in the control quantity after iterative update exceeds the coverage range of the historical samples or equipment step calibration data determined in step S4, an amplitude extrapolation identifier is generated; when the control duration exceeds the coverage range of the duration of the historical samples matched with the current operating condition, a duration extrapolation identifier is generated; when the effective operating periods of two or more environmental control devices overlap and the number of similar joint control response samples is less than 5, an equipment interaction uncertainty identifier is generated.

[0305] After the above-described constraint projection, discretization, and device constraint correction, we obtain the first... Candidate environment control action sequence in round iteration.

[0306] To maintain search stability, the two candidate individuals with the best comprehensive strategy evaluation results are retained in each iteration, and no random replacement is performed on these candidate individuals. When the evaluation result of the group's historical best strategy does not show significant improvement for a preset number of consecutive iterations, some candidate individuals with poor evaluation results are selected from the non-retained candidate individuals, and random perturbations are applied near the group's historical best control variable vector to improve search diversity.

[0307] For each candidate environment control action sequence updated in each iteration, the following processing is re-executed according to step S4: 1. Based on the updated direction of change in the control quantity, reselect the corresponding sample-level device action parameters; 2. Reconstruct the historical response calculation scenario; 3. Recalculate the cumulative effective environmental correction for leaf vapor pressure deficit and leaf condensation margin at the index level; 4. Redetermine the reachability of scenes and the effective arrival window; 5. Regenerate the predicted canopy environment trajectory and the conservative predicted canopy environment trajectory envelope; 6. Redefine the amplitude extrapolation flag, duration extrapolation flag, device interaction uncertainty flag, and device constraint violation flag.

[0308] This ensures that each round of strategy evaluation is based on the actual candidate environment control action sequence after the current iteration update.

[0309] The second part determines the control timing margin and the effective regulation margin; 1. Determine the control timing margin; When there is an environmental correction requirement in the current control cycle, and the first When a candidate environmental control action sequence can form a complete environmental correction value in all historical response calculation scenarios, the corresponding effective action arrival window is obtained: ; The time difference obtained by subtracting the latest arrival time of the effective action arrival time window from the start time of the blade safety boundary crossing time window is determined as the control timing margin of the candidate environmental control action sequence: ; In the formula, Indicates the first The control timing margin of each candidate environmental control action sequence, in minutes.

[0310] when When, it means that the candidate environmental control action sequence can form a complete environmental correction before the earliest possible crossing of the blade safety boundary, and has a corresponding time buffer.

[0311] when The time indicates that the latest arrival time of the candidate environmental control action sequence coincides with the start time of the blade safety boundary crossing time window.

[0312] when When this occurs, it indicates that the candidate environmental control action sequence cannot form a complete environmental correction before the start of the blade safety boundary crossing time window.

[0313] When a candidate environmental control action sequence is only reachable in some historical response calculation scenarios, its effective arrival time window is a right-truncated time window. The preset prediction time domain termination time is only a right-truncated representation and does not indicate that the unreachable scenario can form a complete environmental correction at that time. Therefore, the actual control timing margin of this candidate environmental control action sequence is marked as undetermined.

[0314] To enable swarm intelligence algorithms to numerically sort control action sequences for some reachable candidate environments, a surrogate value for control timing margin evaluation is set: ; In the formula, This represents the proportion of reachable scenarios obtained in step S4. The evaluation substitution value is only used for penalty calculation in policy evaluation and does not represent the actual control time margin.

[0315] For candidate environment control action sequences that are unreachable in all historical response calculation scenarios, set For candidate environment control action sequences that are reachable across all historical response calculation scenarios, set... .

[0316] When the blade safety boundary crossing window obtained in step S2 is empty and the environmental correction is zero, the control timing margin is not used as a feasibility constraint for the candidate environmental control action sequence, and its normalized control timing margin is set to 1.

[0317] 2. Determine the effective adjustment margin; According to the correction direction corresponding to the environmental correction amount, determine the first The cumulative effective environmental correction for leaf vapor pressure deficit and the cumulative effective environmental correction for leaf condensation margin formed before the start of the blade safety boundary crossing time window for each candidate environmental control action sequence.

[0318] When the starting time of the blade safety boundary crossing the time window is between two adjacent calculation times in the internal calculation time grid of step S4, linear interpolation is used to determine the conservative effective environmental correction amount corresponding to the starting time.

[0319] For blade vapor pressure deficit requiring correction, the first The effective environmental correction amount after projection according to the required correction direction in a historical response calculation scenario is determined as follows: ; For the dew condensation margin of blades requiring correction, determine: ; In the formula, and Values ​​are selected in a direction that is conducive to meeting the corresponding safety boundaries; when the value is negative, it indicates that the candidate environmental control action sequence has an adverse effect on the correction requirement in the corresponding historical response calculation scenario.

[0320] For blade vapor pressure deficit requiring correction, the following is determined: ; In the formula, In step S4, the first Conservative effective correction amount determined according to the correction direction required for leaf vapor pressure deficit in a historical response calculation scenario; This indicates the conservative cumulative effective environmental correction amount for the leaf vapor pressure deficit that the candidate environmental control action sequence can guarantee before the start of the time window.

[0321] For the dew condensation margin of blades requiring correction, the following is determined: ; When there is a need to correct the vapor pressure deficit, determine the normalized adjustment margin of the vapor pressure deficit: ; In the formula, This represents the normalized protection value for the vapor pressure deficit of the leaf gas; in this embodiment, it is set to 0.05 kPa.

[0322] When there is a need to correct the leaf dew condensation margin, determine the normalized adjustment margin of the leaf dew condensation margin: ; In the formula, This represents the normalized protection value for leaf condensation margin, which is set to 0.2 degrees Celsius in this embodiment.

[0323] When both environmental indicators require correction, the smaller of the two normalized adjustment margins is determined as the effective adjustment margin of the candidate environmental control action sequence. ; When only one environmental indicator has a need for correction, the normalized adjustment margin corresponding to that environmental indicator is determined as the effective adjustment margin.

[0324] When the environmental correction is zero, the effective adjustment margin is not used as a feasibility constraint for the candidate environmental control action sequence, and its normalized effective adjustment margin is set to 1.

[0325] An effective adjustment margin greater than zero indicates that the conservative cumulative effective environmental correction amount formed by the candidate environmental control action sequence before the start time of the time window exceeds the required environmental correction amount; an effective adjustment margin equal to zero indicates that the conservative cumulative effective environmental correction amount just reaches the required environmental correction amount; an effective adjustment margin less than zero indicates that the candidate environmental control action sequence cannot form a sufficient environmental correction amount before the start time.

[0326] When step S4 generates an indicator that the leaf vapor pressure deficit cannot be fully corrected, the effective adjustment margin only indicates whether the candidate environmental control action sequence has reached the compromise environmental correction amount, and does not mean that it necessarily meets the leaf vapor pressure deficit safety constraint. It is still necessary to make a safety judgment based on the complete conservative prediction of the canopy environmental trajectory envelope.

[0327] The third part generates strategy evaluation results; 1. Determine the minimum normalized safety distance; Regarding the first Given a candidate environmental control action sequence, obtain the conservatively predicted canopy environment trajectory envelope redefined in step S4: ; Based on the distance between the predicted lower limit of leaf vapor pressure deficit and the preset lower limit of leaf vapor pressure deficit, the first... Normalized safety distance for the lower limit of leaf vapor pressure deficit at each predicted time: ; Based on the distance between the preset upper limit of blade vapor pressure deficit and the predicted upper limit of blade vapor pressure deficit, determine the first... Normalized safety distance for the upper limit of the vapor pressure deficit at each predicted time: ; The normalized safety distance for leaf condensation margin is determined based on the distance between the predicted lower limit of leaf condensation margin and the preset lower limit of leaf condensation margin. ; In the formula, This represents the normalized engineering scale for leaf condensation margin; in this embodiment, it is set to... Celsius.

[0328] The minimum normalized safety distance among all predicted times and all safety boundaries within the preset prediction time domain is determined as the minimum normalized safety distance of the candidate environment control action sequence: ; when When, it means that the conservative predicted canopy environment trajectory envelope corresponding to the candidate environmental control action sequence satisfies the leaf transpiration-condensation bilateral safety domain constraint throughout the entire preset prediction time domain.

[0329] when When, it indicates that the conservative predicted canopy environment trajectory envelope at at least one prediction time crosses the corresponding safety boundary.

[0330] 2. Determine the cost of control actions; The control action cost of the candidate environmental control action sequence is determined based on the control variable change amplitude, control duration, and number of control variable switching of each environmental control device. ; In the formula, Indicates the first The cost of control actions for each candidate environmental control action sequence; Indicates the first The continuous operating cost coefficient of an environmental control device; This represents the cost coefficient for a single control quantity switching of the corresponding environmental control equipment; This indicates the number of control quantity switching events that occur in the device within the preset prediction time domain.

[0331] when hour, When the device initiates a control action within the preset prediction time domain and reverts to the current control value after the control duration ends, When the device control action continues until the preset prediction time domain termination time, and the recovery action is outside the current prediction time domain, .

[0332] The continuous operation cost coefficient is determined based on the rated energy consumption of the environmental control equipment, the energy consumption per unit of control quantity, and the water or heat energy consumption; the control quantity switching cost coefficient is determined based on the wear and tear of the actuator, start-stop impact, and allowable operating frequency.

[0333] 3. Normalize the evaluation indicators; The minimum normalized safety distance is limited to a preset evaluation range to obtain the normalized safety distance: ; In the formula, Indicates will Limited to the range Inside.

[0334] For a candidate environment control action sequence with a defined control timing margin, determine the normalized control timing margin: ; For candidate environment control action sequences with uncertain or completely unreachable control time margins, a control time margin evaluation is used as an alternative for normalization. ; Determine the normalized effective adjustment margin: ; Let the preset control action reference cost be Determine the normalized control action cost: ; The preset control action reference cost is determined based on the sum of the costs of all environmental control devices performing a typical effective control action once within the preset prediction time domain.

[0335] 4. Determine the degree of constraint violation and the penalties for uncertainty; Determine the amount of safety constraint violation based on the minimum normalized safety distance: ; Based on the preset time margin threshold Determine the amount of timing constraint violation: ; In the formula, Let: This represents the control timing margin used for policy evaluation; for a candidate environment control action sequence reachable across all historical response calculation scenarios, let: ; Based on the preset adjustment margin threshold Determine the amount of regulation constraint violation: ; Determine the accessibility violation rate based on the proportion of accessible scenarios: ; When the environmental correction is zero, , and All are set to zero.

[0336] Let the number of environmental control devices executing the new control action in the candidate environmental control action sequence be . In the formula Let the characteristic function be represented. Suppose that the number of devices with amplitude extrapolation and lacking corresponding step calibration data is . Determine the range of extrapolation penalty: ; Let the number of devices with duration extrapolation be . Determine the duration extrapolation penalty: ; Let the number of device pairs in the candidate environmental control action sequence that have overlapping effective action periods be . Among them, the number of device pairs with fewer than 5 similar joint control response samples is [number missing]. Determine the penalty for uncertainty in device interaction: ; When a candidate environmental control action sequence violates equipment safety operation constraints and cannot be repaired by projecting control variables, the equipment constraint violation amount is set to... ,otherwise .

[0337] 5. Formulate strategy evaluation results; The basic policy score for determining candidate environment control action sequences is based on normalized safety distance, normalized control timing margin, normalized effective regulation margin, and normalized control action cost. ; In the formula, , , and These represent the evaluation weights for safety distance, control timing margin, effective regulation margin, and control action cost, respectively, and satisfy the following conditions: This embodiment will , , , .

[0338] The total penalty value is determined based on the number of violations of various constraints and the uncertainty penalty: ; In this embodiment, the penalty coefficients corresponding to safety constraints, timing constraints, and adjustment constraints are greater than the uncertainty penalty coefficient. For example, it can be set to... , , , , , , , .

[0339] Subtracting the total penalty value from the base policy score yields the comprehensive policy score for the candidate environment control action sequence: ; The higher the overall strategy score, the better the candidate environmental control action sequence performs in terms of safety, control timeliness, environmental adjustment capability, equipment response reliability, and control action cost.

[0340] The strategy evaluation results for candidate environment control action sequences are composed of comprehensive strategy score, minimum normalized safety distance, control timing margin, effective adjustment margin, control action cost, reachability ratio, and various identifiers. ; In the formula, These indicate safety constraint violation, arrival window right truncation, amplitude extrapolation, duration extrapolation, device interaction uncertainty, and device constraint violation.

[0341] In the process of updating the historical best position of candidate individuals and the historical best position of the group, a constraint-first comparison rule is adopted: 1. A candidate environment control action sequence that satisfies all preset feasibility constraints is better than a candidate environment control action sequence that violates constraints; 2. When both candidate environmental control action sequences satisfy all preset feasibility constraints, the candidate environmental control action sequence with the larger comprehensive strategy score is better; 3. When both candidate environmental control action sequences have constraint violations, the candidate environmental control action sequence with the smaller total penalty value is better; 4. When the total penalty value is the same, the candidate environment control action sequence with the larger minimum normalized safety distance is better; 5. When the minimum normalized safety distance is still the same, the candidate environment control action sequence with lower control action cost is better.

[0342] Both the historical best position of the candidate individual and the historical best position of the group are stored as vectors of actual executable control variables after control range projection, discretization, and device constraint correction.

[0343] The fourth part involves satisfying the iteration termination condition and determining the optimal environmental control action sequence; The preset iteration termination conditions include the maximum number of iterations and the stability of the strategy evaluation results.

[0344] This embodiment will specify the maximum number of iteration rounds. Set to 50 iterations. The maximum number of iterations is satisfied when the number of iterations reaches the maximum number of iterations.

[0345] When the improvement in the comprehensive strategy score of the group's historical best candidate environment control action sequence is lower than the preset score improvement threshold in 10 consecutive iterations, the strategy evaluation result is deemed stable. In this embodiment, the preset score improvement threshold is set to 0.001.

[0346] The swarm intelligence algorithm stops iterating when either the maximum number of iterations condition or the stability condition of the strategy evaluation result is met.

[0347] After the iteration terminates, feasible candidate environmental control action sequences are selected based on the strategy evaluation results redefined for each candidate environmental control action sequence.

[0348] When there is an environmental correction requirement in the current control cycle, the candidate environmental control action sequence must simultaneously meet the following conditions: First, the conservatively predicted canopy environment trajectory envelope corresponding to the candidate environmental control action sequence satisfies the following within the preset prediction time domain: ; ; as well as: ; in, Take 1 to 1 in sequence .

[0349] Second, the candidate environmental control action sequence can form a complete environmental correction quantity in all historical response calculation scenarios, does not have a right truncation marker for the arrival time window, and the control timing margin is satisfied. .

[0350] Third, the effective adjustment margin meets the requirements. .

[0351] Fourth, there are no irreparable violations of equipment safety operation constraints in the candidate environmental control action sequence.

[0352] Fifth, the candidate environmental control action sequence does not have amplitude extrapolation that lacks historical samples or equipment step calibration data support; or its amplitude extrapolation part has been verified by segmented step calibration data.

[0353] Sixth, the candidate environmental control action sequence does not have a duration extrapolation that lacks historical control response samples or equipment continuous operation calibration data support; or the duration extrapolation has already formed a duration extrapolation uncertainty in step S4, and the expanded conservative prediction canopy environmental trajectory envelope still satisfies the leaf transpiration-condensation bilateral safety domain constraint.

[0354] Seventh, when the effective operating periods of two or more environmental control devices overlap, the device interaction uncertainty has been formed according to step S4 based on similar joint control response samples or preset proportions, and the expanded conservative prediction of the canopy environmental trajectory envelope still satisfies the leaf transpiration-condensation bilateral safety domain constraint.

[0355] In this embodiment, the preset timing margin threshold is set to... Minutes, set the preset adjustment margin threshold to An effective adjustment margin threshold of 0.05 indicates that the conservative cumulative effective environmental correction formed by the candidate environmental control action sequence before the start of the blade safety boundary crossing time window is at least 5% higher than the required environmental correction.

[0356] When there is no need for environmental correction in the current control cycle, the candidate environmental control action sequence is not required to have an effective arrival time window, nor are control timing margin thresholds and effective adjustment margin thresholds imposed on it. It is only required that its conservatively predicted canopy environmental trajectory envelope satisfies the leaf transpiration-condensation bilateral safety domain constraint and that there is no violation of equipment safety operation constraints.

[0357] The candidate environmental control action sequences that meet the corresponding conditions are used to form a feasible candidate set: ; When the feasible candidate set is not empty, the candidate environmental control action sequence with the highest comprehensive strategy score is determined as the preferred environmental control action sequence: ; When the difference in the comprehensive strategy scores of multiple feasible candidate environmental control action sequences is less than the preset ties threshold, the preferred environmental control action sequence is determined in the order of lower control action cost, larger minimum normalized safety distance, and larger control timing margin.

[0358] When the blade safety boundary crossing window is empty and the environmental correction is zero, if the conservatively predicted canopy environmental trajectory envelope corresponding to the baseline candidate environmental control action sequence that maintains the current control state satisfies the blade transpiration-condensation bilateral safety domain constraint, then the baseline candidate environmental control action sequence is preferentially determined as the preferred environmental control action sequence to avoid unnecessary equipment actions when there is no safety correction requirement.

[0359] When the feasible candidate set is empty after the iteration terminates, an extended search is performed on the population within the current allowable control range and equipment safety operation constraints of each environmental control device. The extended search is achieved by increasing the number of candidate individuals, expanding the range of random disturbances near the directional candidate actions, and increasing the cooperative combinations of different devices, but must not exceed the adjustable range of equipment control quantities and equipment safety operation constraints.

[0360] If no candidate environmental control action sequence that satisfies all preset constraints is found after the extended search, an identifier indicating no fully feasible control strategy is generated, and the candidate environmental control action sequence with the smallest total penalty value is selected from all candidate environmental control action sequences. When multiple candidate environmental control action sequences have the same total penalty value, the candidate environmental control action sequence with the larger minimum normalized safe distance, the higher proportion of reachable scenarios, and the smaller control action cost is selected in sequence as the emergency environmental control action sequence.

[0361] The emergency environmental control action sequence is not a conventional preferred environmental control action sequence that satisfies all safety constraints. Instead, it is used to minimize the degree of canopy environment boundary crossing when the environmental regulation equipment is insufficient, the blade safety boundary crossing time is too early, or the equipment availability range is limited, and to trigger the abnormal control feedback and manual intervention prompt in step S6.

[0362] It should also be noted that step S5 does not select solely based on the canopy environment state at the end of the prediction time domain of the candidate environmental control action sequence, but simultaneously evaluates the latest time at which the candidate action forms a complete environmental correction, the conservative effective adjustment amount before the blade safety boundary crosses the time window, the minimum normalized safety distance within the entire preset prediction time domain, the reliability of the equipment response, and the cost of the control action.

[0363] By recalculating the effective action arrival window and predicting the canopy environment trajectory after each round of group iteration, the changes in action initiation time, control variable amplitude, and control duration can be accurately reflected in the strategy evaluation results. By using a conservative prediction of the canopy environment trajectory envelope to calculate the minimum normalized safety distance, it is possible to avoid masking slower equipment responses, equipment interaction effects, and prediction errors based solely on representative predicted trajectories. By setting control timing margin and effective regulation margin separately, it is possible to distinguish between two infeasible situations: control action arrival time being too late and control action intensity being insufficient. By adopting a constraint-priority candidate individual comparison method, it is possible to avoid candidate environmental control action sequences with low control action costs but violating blade safety constraints from incorrectly guiding the group iteration direction.

[0364] Therefore, step S5 outputs the preferred environmental control action sequence and its strategy evaluation results; when there is no completely feasible candidate environmental control action sequence, it outputs the identifier of no completely feasible control strategy, the emergency environmental control action sequence and its constraint violation results, providing a basis for step S6 to execute environmental regulation equipment control, collect actual canopy environmental response and update equipment operation parameters.

[0365] S6: Select the preferred or emergency sequence as the sequence to be executed and output the equipment control command. After obtaining the actual canopy environment trajectory, update the prediction error and boundary according to the deviation between the actual control path and the predicted trajectory. Update the historical samples according to the identifiable actual response segments, correct the index-level time series and unit contribution, and use the updated results for the next cycle.

[0366] The preferred environmental control action sequence, or the emergency environmental control action sequence when no fully feasible control strategy identifier is generated, is determined as the environmental control action sequence to be executed in the current control cycle; The control command corresponding to the current cycle is sent to each device for execution, and the actual feedback control quantity, air temperature, air relative humidity and crop canopy temperature are collected to form the actual canopy environment trajectory; The path difference between the actual control process and the planned process is determined based on the actual feedback control quantity. When the difference does not exceed the preset threshold, the representative trajectory corresponding to the sequence to be executed is used as the predicted trajectory corresponding to the actual control process. Otherwise, the actual action sequence is generated based on the actual feedback control quantity and the predicted trajectory is regenerated. The actual trajectory is compared with the predicted trajectory corresponding to the actual control process according to the prediction lead time to obtain the prediction error of the two indicators. The historical prediction error set and its upper and lower deviation boundaries of each prediction lead time are updated for the formation of the trajectory interval in the next cycle. Establish equipment response tracking records based on the actual feedback control quantity change process, extract actual control response segments from closed records, add segments that meet the single device identification conditions to the historical sample set according to device, change direction and effective response index, and add segments with overlapping effective effects of multiple devices that cannot be separated to the joint sample set. Based on the segments that meet the identifiable conditions, determine the actual indicator-level effect start-up delay, rise time, residual effect time after stopping, and actual unit contribution of the two indicators, and perform confidence weighting and single change limit update on the directional initial parameters in the same direction; The actual trajectory, historical prediction error set, deviation boundary, updated historical sample set, joint sample set, and directional initial parameters are used as historical control feedback data for the next cycle. In the next cycle, the operating condition matching is re-executed to determine the index-level time series and unit contribution.

[0367] In this embodiment, step S6 is used to execute the expected environmental control action sequence for the current control cycle determined in step S5, obtain the actual execution process of the environmental conditioning equipment and the actual canopy environmental trajectory; update the historical prediction error data and its deviation boundary according to the deviation between the actual canopy environmental trajectory and the predicted canopy environmental trajectory corresponding to the actual control process; update the historical control response sample according to the completed actual control response segments that meet the identifiable conditions, and correct the equipment action timing and the environmental conditioning contribution of the unit control quantity; and use the updated results for the next control cycle.

[0368] The environmental control action sequence to be executed in the current control cycle is the preferred environmental control action sequence when there are fully feasible candidate environmental control action sequences, and is the emergency environmental control action sequence when no fully feasible control strategy identifier is generated in step S5.

[0369] Step S6 includes four parts: execution of control commands and formation of actual canopy environment trajectory, update of prediction error boundary, update of equipment operating parameters, and preparation of data for the next control cycle.

[0370] The first part executes control commands and forms the actual canopy environment trajectory; The sequence of environmental control actions to be executed in the current control cycle is as follows: ; In the formula, This indicates the preferred sequence of environmental control actions; This represents the sequence of emergency environmental control actions. The sequence of environmental control actions to be executed during the current control cycle is represented as follows: ; Among them, the first The pending actions of each environmental control device are recorded as follows: ; In the formula, Indicates the start time of the action to be performed; Indicates the magnitude of change in the signed control quantity; Indicates the duration of control.

[0371] The rolling time-domain control method only selects control actions from the expected environmental control sequence that fall within the current control cycle. The actions of initiating, maintaining, and restoring control quantities within the system are converted into control commands for the environmental control equipment and sent to the corresponding environmental control equipment for execution.

[0372] Planned actions in subsequent control cycles are not locked in advance. After entering the next control cycle, steps S1 to S5 are re-executed based on newly acquired greenhouse environmental status data, crop canopy status data, outdoor meteorological data, environmental regulation equipment operation data, and historical control feedback data.

[0373] No. The stopping time of each environmental control device is: ; When the action is initiated within the current control cycle, the first... The target control quantity for each environmental control device is determined as follows: ; In the formula, Indicates the number of seconds before the start of the action. The actual control quantity of each environmental control device; and These represent the lower and upper limits of the currently permissible control parameters for the environmental control equipment, respectively. This indicates that the target control quantity is limited between the lower limit and the upper limit of the control quantity.

[0374] when At the same time, the current control quantity of the corresponding environmental control equipment remains unchanged.

[0375] When the action stops within the current control cycle, the corresponding environmental control equipment is restored to the actual control value before the start of this action. When the action stops in a subsequent control cycle, the recovery operation is not locked in advance in the current control cycle; the rolling optimization result of the next control cycle determines whether the corresponding equipment should continue to maintain, readjust, or recover.

[0376] While sending control commands to the environmental control equipment, the actual feedback control quantities of each environmental control equipment are collected, and the first... The actual feedback control quantity of each environmental control device is denoted as: When the environmental control equipment is not configured with an actual control quantity feedback interface, the control command value sent to the environmental control equipment is used as the feedback control quantity substitute value, and a command substitution identifier is attached.

[0377] For continuous control equipment configured with an actual control quantity feedback interface, when the following conditions are met... Furthermore, when the number of continuously confirmed sampling points is not less than the preset number, an abnormal device control execution flag is generated.

[0378] In the formula, Indicates the first The execution deviation threshold for each environmental control device. In this embodiment, the execution deviation threshold for continuous control devices is set to 0.05, and the number of preset confirmation sampling points is set to 3. For environmental control devices that only have start-stop states, the control execution result is determined based on whether the actual feedback state is consistent with the target state.

[0379] When the environmental control equipment is not equipped with an actual control quantity feedback interface, but has action completion status, position arrival status, or operation status feedback, the control command is determined based on the status feedback. When there is neither actual control quantity feedback nor status feedback that can verify action completion, an execution status unverifiable flag is added to the corresponding control response data.

[0380] Feedback data from environmental control equipment and canopy environment data with abnormal control execution indicators are saved as abnormal operation data and are not used to update normal historical prediction error boundaries and single equipment action parameters. Data with unverifiable execution status indicators are used for updating equipment action parameters with lower sample confidence only when there are no other abnormalities and the control response meets the identifiability condition of step S3.

[0381] Following the sampling interval determined in step S1, continue to collect air temperature, relative humidity, and crop canopy temperature. Using the method in step S2, calculate the actual air vapor pressure, air dew point temperature, leaf vapor pressure deficit, and leaf dew margin corresponding to each sampling time.

[0382] Let the first The actual sampling time is The corresponding actual values ​​of leaf vapor pressure deficit and leaf condensation margin are respectively and Arrange the above actual values ​​according to the sampling time to form the actual canopy environment trajectory: ; In the formula, This indicates the number of valid sampling points within the corresponding data collection period.

[0383] The actual canopy environment trajectory must cover at least the current control period. For environmental control equipment that experiences changes in effective control quantities within the current control period, equipment response tracking records must be established. Each equipment response tracking record must include at least: environmental control equipment identification, time of actual control quantity change, direction of actual control quantity change, magnitude of actual control quantity change, response observation start time, control quantity recovery time, and response tracking status.

[0384] The response tracking status includes an unclosed state, a closed state, and an invalid state.

[0385] The device response tracking record in the unclosed state allows for the continued supplementation of actual canopy environmental data across subsequent control cycles until any of the following conditions are met: the environmental effect of the corresponding environmental control device has subsided; the preset response observation period set in step S3 has been reached; a new control action that cannot be separated from the current control response has occurred; or the feedback data from the environmental control device or sensor data is continuously abnormal.

[0386] When the preset response observation period ends but the effect decay process is not fully completed, the equipment response tracking record is marked as a residual effect right-truncation record. When a new control action that cannot be separated or continuous abnormal data occurs, the equipment response tracking record is marked as an invalid record.

[0387] Closed device response tracking records are written to the historical control response sample set only once, avoiding the repeated recording and updating of the same actual control response in multiple control cycles.

[0388] When the sampled data experiences communication interruptions, sensor malfunctions, or exceeds a preset physical reasonable range, it is marked as invalid and short-term interpolated according to the data quality processing rules in step S1. Data segments where the ratio of invalid sampling points to interpolated sampling points exceeds a preset ratio are not used to update the normal historical prediction error boundary and equipment operating parameters.

[0389] Part Two: Update the prediction error boundary; Obtain the representative predicted canopy environment trajectory corresponding to the expected sequence of environmental control actions to be executed in the current control cycle, and record it as... and The predicted canopy environment trajectory is the central predicted trajectory formed by calculating the scenario weighted average of multiple historical responses, based on the sequence of environmental control actions to be executed in the current control cycle in step S4.

[0390] The prediction error calculation uses the predicted canopy environment trajectory consistent with the actual control process to avoid incorrectly including subsequent rolling optimization, deviations in the execution of environmental conditioning equipment, or changes in the control path caused by manual operation in the model prediction error.

[0391] Suppose that from the current control decision time to the nth Predicted time Between these points, the planned control quantity trajectory corresponding to the expected sequence of environmental control actions to be executed in the current control cycle is: The actual feedback control trajectory of each environmental control device is as follows: Determine the maximum control path difference between the planned control trajectory and the actual feedback control trajectory: ; In the formula, Indicates the first The control quantity of an environmental control device in the planned control quantity trajectory; This represents the corresponding actual feedback control quantity.

[0392] when At that time, it is believed that from the current control decision moment to the [missing information] moment... The actual control process at each predicted time point is consistent with the planned control process, and makes... , .

[0393] In the formula, This represents the control path consistency threshold. In this embodiment, the control path consistency threshold is set to 0.05.

[0394] when This indicates that rolling optimization, equipment execution deviation, or manual operation has altered the original planned control process.

[0395] Based on the time from the current control decision moment to the prediction moment The actual feedback control quantity trajectory between the components decomposes the actual control process of each environmental control device into one or more actual control quantity change segments. Each actual control quantity change segment includes the time of actual control quantity change, the amplitude of the signed actual control quantity change, and the actual duration.

[0396] For the prediction time The control variable change segment that has not yet ended will predict the time. The segment termination time used in this control consistency prediction is determined. It is composed of segments representing changes in each actual control variable up to the prediction time. The actual control action sequence.

[0397] Using the initial greenhouse state frozen at the time of the original prediction, gray box model parameters, historical response calculation scenario parameters, and outdoor weather prediction sequence generated during the current control cycle, and taking the actual control action sequence as the incremental control input, the control consistency prediction value is recalculated according to the prediction methods in steps S2 and S4. and The control consistency prediction value is only used to evaluate the model prediction error under actual control path conditions and does not change the previously completed strategy evaluation results.

[0398] When the actual feedback control quantity of the environmental control equipment is missing, and the actual control path cannot be reliably restored based on the information, action status feedback, or equipment operating status confirmed by the control command, the normal historical prediction error data corresponding to the prediction lead time will not be updated, and the corresponding data will be saved as an unverifiable record of the control path.

[0399] Regarding the first The prediction lead time is used to determine the prediction error of the vapor pressure deficit: ; Determine the prediction error of leaf condensation margin: ; When multiple valid actual sampling points exist near the prediction time, the median of the actual sampling values ​​within the prediction's immediate vicinity, centered on the prediction time, is used as the corresponding actual value. When no valid actual sampling points exist near the prediction time, the historical prediction error data corresponding to that prediction lead time is not updated.

[0400] If there are abnormalities in device control execution, continuous sensor malfunctions, or unverifiable control paths within the corresponding prediction period, the corresponding prediction error will not be added to the normal historical prediction error set.

[0401] The prediction error corresponding to each control decision moment and each prediction lead time is written into the historical prediction error set only once.

[0402] Let the first The historical prediction error sets for leaf vapor pressure deficit and leaf condensation margin corresponding to the prediction advance times are respectively: and The currently obtained effective prediction error is added to the corresponding historical prediction error set, and the most recent ones are retained in chronological order. One effective prediction error: ; In the formula, This indicates a vapor pressure deficit in the leaves. Indicates the condensation allowance on the leaves; This indicates that the most recent items should be retained in chronological order. One effective prediction error. This embodiment will Set it to 200.

[0403] When the number of valid historical prediction errors corresponding to a certain prediction lead time is not less than 60, calculate the candidate lower bias boundary and the candidate upper bias boundary: ; In the formula, and These represent the 5th percentile and the 95th percentile, respectively.

[0404] To avoid excessive changes in the deviation boundary between adjacent control cycles, a single-cycle variation limit is applied to the candidate deviation boundary: ; ; In the formula, Indicators of environmental indicators The maximum permissible variation of the deviation boundary within a single control cycle. In this embodiment, the maximum permissible variation of the leaf vapor pressure deficit deviation boundary in a single cycle is set to 0.08 kPa, and the maximum permissible variation of the leaf condensation margin deviation boundary in a single cycle is set to 0.25 degrees Celsius.

[0405] Perform sequential correction on the deviation boundaries after amplitude limiting: ; ; This ensures that the lower deviation boundary is not higher than the upper deviation boundary.

[0406] When the number of valid historical prediction errors corresponding to a certain prediction lead time is less than 60, the historical prediction error boundary corresponding to the previous control cycle is continued to be used; when the system has not yet formed a historical prediction error boundary, the preset initial error boundary set in step S2 is used.

[0407] Based on the updated lower and upper deviation boundaries, the predicted canopy environment trajectory corresponding to the actual control process is adjusted within a range.

[0408] The lower and upper bounds for the correction of vapor pressure deficit in leaf gas are as follows: ; ; The lower and upper bounds for the leaf condensation margin correction are as follows: ; ; The control-consistent canopy environment trajectory interval for the current control cycle is formed by the lower and upper bounds of the correction at each prediction time, and is used to verify the prediction results for the current control cycle.

[0409] The updated historical prediction error boundary is used for canopy environment trajectory interval formation in step S2 of the next control cycle. (Next control cycle...) The boundary of the vapor pressure deficit interval at each predicted time point is: ; ; Next control cycle The boundary of the leaf dew margin interval for each predicted time is: ; ; In the formula, and These represent the leaf vapor pressure deficit baseline prediction value and the leaf condensation margin baseline prediction value, respectively, regenerated in step S2 based on the actual initial state of the next control cycle.

[0410] When the median of the most recent preset number of valid prediction errors continuously exceeds the corresponding preset deviation threshold in the same direction, a persistent prediction deviation flag is generated. The persistent prediction deviation flag is used to indicate that the gray box model parameters, outdoor meteorological forecast data, or sensor status may drift, and does not directly replace the normal historical prediction error boundary update.

[0411] Part Three: Update the equipment operating parameters; Based on the actual feedback control quantities of each environmental control device, and following the control quantity change event identification method in step S3, determine the time of actual control quantity change, the magnitude of actual control quantity change, and the direction of control quantity change.

[0412] For environmental control equipment that is not configured with an actual control quantity feedback interface, the control quantity change process is determined based on the control command, action completion status and equipment operating status, and the command substitution identifier or execution status unverifiable identifier is retained in the corresponding equipment response tracking record.

[0413] Once the device response tracking record is closed, the corresponding actual control response segment is extracted from the actual canopy environment trajectory: ; In the formula, Indicates the first The first environmental control equipment The actual control quantity change time of an actual control quantity change event; This indicates the preset response observation period set in step S3.

[0414] The actual control response segments are filtered according to the identifiability criteria in step S3. The single-device action parameters are not updated using the corresponding actual control response segment when any of the following conditions exist: the target environmental control equipment has an abnormal control execution flag; other environmental control equipment experiences control quantity changes exceeding a preset interference threshold during the main response identification period; outdoor weather conditions change beyond the identifiability range set in step S3; the ratio of invalid sampling points to interpolated sampling points exceeds a preset ratio; or the target environmental control equipment experiences another effective control quantity change before the main response is formed.

[0415] When the effective operating periods of multiple environmental control devices overlap and the independent operation of each environmental control device cannot be reliably separated, the corresponding actual control response segment is marked as a joint control response sample.

[0416] The joint control response sample is added to the joint sample set of the corresponding equipment combination. It is used in the next control cycle to determine whether the candidate environmental control action sequences of multiple equipment have similar joint response basis and to determine the uncertainty boundary of equipment interaction. The environmental response in the joint control response sample is not forcibly assigned to any individual environmental control equipment.

[0417] For a single device's actual control response segment that meets the identifiable conditions, the actual action start time, actual action achievement time, and actual action dissipation time corresponding to the blade vapor pressure deficit and blade condensation margin are determined according to the dynamic benchmark trend, benchmark fluctuation range, adaptive effective response threshold, and continuous sampling point confirmation rules in step S3.

[0418] For environmental indicators ,in This indicates a vapor pressure deficit in the leaves. Indicates the dew condensation margin on the leaves, determining the actual indicator-level effect start-up delay: ; Determine the rise time of actual indicator-level effects: ; Determine the residual effect time after the actual indicator level stops: ; In the formula, Indicators of environmental indicators The corresponding actual start time of action; Indicates the moment when the actual effect is achieved; Indicates the time when the actual effect diminishes; This indicates the time when the actual control quantity of the corresponding environmental control equipment returns to near the control quantity before the change.

[0419] When an actual control response segment does not contain a complete control quantity recovery process and action decay process, it is marked as a residual action right-truncated sample. This residual action right-truncated sample can be used to update the action initiation delay and action rise time, but not to update the residual action time after stopping.

[0420] Based on the actual effective response statistical period determined in step S3, calculate the environmental indicators. The actual effective change: ; In the formula, Indicators of environmental indicators The actual trajectory deviates from the response of the corresponding dynamic baseline trend; Indicators of environmental indicators The corresponding period for actual effective response statistics.

[0421] The environmental regulation contribution per unit of control quantity is determined based on the actual change range of the control quantity. ; In the formula, It represents the magnitude of the change in the signed actual control quantity determined based on the actual feedback control quantity of the environmental control equipment.

[0422] When a control command is used to replace the actual feedback control quantity, the change amplitude of the signed control quantity corresponding to the control command is used instead. This reduces the credibility of the corresponding historical control response samples.

[0423] The sign of the change in the control quantity is retained. When the environmental regulation contribution of the actual unit control quantity is positive, it indicates that the direction of change of the environmental indicator is the same as the direction of change of the control quantity; when the environmental regulation contribution of the actual unit control quantity is negative, it indicates that the direction of change of the environmental indicator is opposite to the direction of change of the control quantity.

[0424] Actual control response segments that pass the identifiability screening will be added to the historical control response sample set according to the environmental control equipment, the direction of control quantity change, and the environmental indicators that form an effective response.

[0425] To correct for system cold start, insufficient current operating condition matching samples, and directional initial parameters used after replacement of environmental control equipment, a confidence weighting and single change limit update are performed on each closed and valid actual control response sample.

[0426] Let the environmental regulation contribution of any index-level action time parameter or unit control variable to be updated be... The corresponding actual observed value is Then the directionality initial parameters are updated as follows: ; In the formula, Indicates the direction of increasing or decreasing the control quantity; Indicates the correction coefficient for the basic parameters; Indicates the credibility of the actual control response sample; This indicates the maximum allowable change in a single update of the corresponding parameter.

[0427] This embodiment uses the basic parameter correction coefficient. Set it to 0.20.

[0428] When the actual feedback from the environmental control equipment is complete, the actual canopy environmental data is complete, and there is no obvious external interference, the confidence level of the actual control response sample is set to 1; when the actual canopy environmental data contains a small amount of interpolated data but still meets the identifiable conditions, the confidence level of the sample is set to 0.8; when control commands are used to replace the actual control quantity feedback but there is action completion status or operation status verification, the confidence level of the sample is set to 0.7.

[0429] The maximum permissible single-time variation is set for the action start-up delay, action rise time, residual action time after stopping, and environmental regulation contribution per unit control quantity, to avoid abrupt changes in the directional initial parameters caused by a single actual control response sample.

[0430] When the deviation between the actual observed value and the initial directional parameter before correction exceeds a preset anomaly multiple of the historical discrete scale of the corresponding parameter, the corresponding sample is marked as a parameter anomaly sample, and the initial directional parameter is not immediately corrected using this sample. The parameter anomaly sample is retained in the sample set to be reviewed; subsequently, when multiple actual control response samples with the same direction of change of control variables, similar operating conditions, and consistent change trends appear, they are then converted into valid update samples.

[0431] When the actual control response segment only forms an effective response to one of the environmental indicators, namely leaf vapor pressure deficit or leaf condensation margin, only the action sequence and unit control quantity environmental regulation contribution of the corresponding environmental indicator are updated, and the response parameters of the other environmental indicator are not presumed.

[0432] When executing step S3 in the next control cycle, the current operating condition is re-matched and the samples are weighted based on the updated historical control response sample set. Combined with the corrected directional initial parameters, the equipment action timing and unit control quantity environmental regulation contribution used in the next control cycle are obtained.

[0433] Therefore, this step does not use a single actual control response sample to directly cover the current operating condition equipment parameters. Instead, it completes the equipment parameter correction by updating the historical control response sample set, correcting the directional initial parameters, and re-executing the current operating condition matching.

[0434] Part Four: Preparing data for the next control cycle; Write the following data generated during the current control cycle and subsequent equipment response tracking process into the historical control feedback data: 1. The current control cycle is expected to execute the environmental control action sequence, environmental conditioning equipment control commands, target control variables, and action sending times; 2. Actual feedback control quantities, command substitution indicators, unverifiable execution status indicators, and abnormal equipment control execution indicators of environmental control equipment; 3. Actual canopy environment trajectory; 4. Prediction errors of leaf vapor pressure deficit and leaf condensation margin corresponding to each prediction advance time; 5. The updated set of historical prediction errors and its lower and upper bias boundaries; 6. Samples of actual control responses for single devices and samples of combined control responses that have been closed; 7. Residual effect right truncation marker, parameter anomaly sample marker, persistent prediction deviation marker, and control path unverifiable record; 8. There is no fully feasible control strategy identifier and its constraint violation results.

[0435] The updated set of historical prediction errors, lower deviation boundary, and upper deviation boundary are used to form the canopy environment trajectory interval in the next control cycle step S2.

[0436] The updated historical control response sample set and the corrected directional initial parameters are used for current condition matching in step S3 of the next control cycle to determine the timing of the action of index-level equipment, the timing of the action of general equipment, and the environmental regulation contribution of the unit control quantity in the next control cycle.

[0437] The joint control response sample is used in the next control cycle step S4 to retrieve similar joint control response bases and determine the uncertainty boundary of device interaction.

[0438] Steps S4 and S5 regenerate and evaluate candidate environmental control action sequences based on the canopy environment trajectory range and equipment action parameters redefined for the next control cycle, without directly using subsequent actions that have not yet been executed in the environmental control action sequence to be executed in the current control cycle.

[0439] When the expected environmental control action sequence for the current control cycle is an emergency environmental control action sequence, the identifier of no fully feasible control strategy and the corresponding constraint violation result are retained, and a prompt indicating insufficient equipment adjustment capability or manual intervention is generated. The actual control response formed by the emergency environmental control action is only used for updating equipment operating parameters if the identifiability condition in step S3 is met.

[0440] When equipment control execution malfunctions, sensors exhibit continuous abnormalities, prediction deviations persist, or equipment parameters continuously reach the maximum permissible change in a single instance, a control feedback malfunction flag is generated.

[0441] The next control cycle will perform at least one of the following actions based on the anomaly type corresponding to the control feedback anomaly identifier: limit the allowable control range of the abnormal environment control equipment; mark the abnormal environment control equipment as temporarily unavailable; expand the corresponding canopy environment trajectory interval or conservatively predict the canopy environment trajectory envelope; increase the uncertainty of equipment interaction, parameter extrapolation, or strategy evaluation penalty corresponding to the control execution anomaly; output an environmental control equipment check or manual intervention prompt.

[0442] Control feedback anomaly identification does not directly lead to the deletion of existing historical control response samples. A historical control response sample is only marked as invalid when it is confirmed that the error was caused by a sensor malfunction, environmental control equipment malfunction, or incorrect control variable recording.

[0443] Step S6 operates in a rolling manner. Each control cycle executes only the control actions already determined for the current control cycle, and uses the latest actual canopy environment status, historical prediction error boundaries, and historical control response samples to re-complete subsequent predictions and optimize environmental control actions, forming a closed-loop environmental control process from the output of control commands from environmental regulation equipment, acquisition of actual canopy environment trajectories, updating of prediction error boundaries, correction of equipment operating parameters, to re-prediction and optimization in the next control cycle.

[0444] Example 2 is the second embodiment of the present invention, which differs from the previous embodiment in that: If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art or the current technical solution, can be embodied in the form of a software product. This current computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0445] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0446] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0447] Example 3, referring to Figure 2 As an embodiment of the present invention, a greenhouse environment control system based on swarm intelligence is provided, which includes a data acquisition module, a safety risk prediction module, an equipment operation parameter determination module, a candidate control action generation module, a control strategy optimization module, and a control execution and feedback correction module.

[0448] Data acquisition module: Acquires greenhouse environmental status data, crop canopy status data, outdoor meteorological data, environmental regulation equipment operation data, and historical control feedback data for the current control cycle; Safety risk prediction module: Based on the greenhouse environmental status data and crop canopy status data, determine the leaf transpiration-condensation bilateral safety domain; based on the greenhouse environmental status data, crop canopy status data, outdoor meteorological data and historical control feedback data, predict the canopy environment trajectory range to maintain the current equipment control state, and determine the leaf safety boundary crossing time window; Equipment Operation Parameter Determination Module: Based on the operating data of the environmental regulation equipment and historical control feedback data, determine the timing of the operation of each environmental regulation equipment and the environmental regulation contribution per unit control quantity to the indicator-level equipment for the current operating conditions of the blade vapor pressure deficit and blade condensation margin. Candidate control action generation module: Determines the environmental correction amount and generates candidate environmental control action sequences based on the deviation between the canopy environment trajectory interval and the bilateral safety domain; Constructs a historical response calculation scenario based on the index-level time series and unit contribution; Determines the effective arrival time window, representative predicted canopy environment trajectory, and conservative predicted canopy environment trajectory envelope for each candidate sequence. The control strategy optimization module utilizes a swarm intelligence algorithm to iteratively update candidate sequences. In each iteration, the effective arrival window, the representative predicted canopy environment trajectory, and the conservative predicted canopy environment trajectory envelope of each candidate environmental control action sequence are redefined. The control timing margin is determined based on the crossing and arrival windows, and the effective regulation margin is determined based on the cumulative effective correction and environmental correction before the start of the crossing window. The strategy is evaluated by combining the conservative predicted envelope, timing margin, and regulation margin. After the iteration terminates, the best candidate sequence is selected from those whose conservative envelope satisfies the safety domain constraint and whose timing margin and regulation margin are not lower than the corresponding thresholds. If no candidate sequence satisfies the constraints, an emergency sequence is determined based on the degree of constraint violation. Control execution and feedback correction module: It takes the preferred or emergency sequence as the sequence to be executed and outputs the equipment control command. After obtaining the actual canopy environment trajectory, it updates the prediction error and boundary according to the deviation between the actual control path and the predicted trajectory. It updates the historical samples according to the identifiable actual response segments, corrects the index-level time series and unit contribution, and uses the updated results for the next cycle.

[0449] Example 4 is an embodiment of the present invention, which provides a greenhouse environment control method based on swarm intelligence. To verify the technical effects of the present invention in terms of maintaining canopy environment safety, control timeliness, adjustment adequacy, equipment operation economy and rolling feedback correction, a paired control experiment was carried out in multiple independent control areas of greenhouses.

[0450] This embodiment uses a multi-span plastic greenhouse with an east-west orientation and a total effective area of ​​approximately 480 square meters as the experimental site. The greenhouse interior is divided into eight independent control zones, each approximately 60 square meters in size, using light-transmitting and heat-insulating partitions. Each control zone has its own independent air intake and exhaust channels, environmental control equipment control interfaces, and sensor acquisition channels to reduce air exchange and control interference between adjacent control zones. The same tomato variety in its fruiting stage is planted in each control zone, using the same planting density, substrate formula, irrigation system, and fertilization system. The eight control zones are divided into four paired control zones according to the longitudinal position of the greenhouse. Two control zones in each group are located in close proximity for lighting and ventilation. One control zone uses the swarm intelligent rolling control method of this invention, designated as Invention Groups F1 to F4, while the other control zone uses the traditional threshold linkage control method, designated as Control Groups C1 to C4. The experiment runs continuously for 14 days. The first two days serve as the cold start phase for the method of this invention. For the remaining 12 days, effective high-humidity risk events are statistically analyzed, with each control zone experiencing a total of 12 effective high-humidity risk events.

[0451] Three air temperature and humidity sampling points were set up in each control zone at a height of 0.3 meters above the crop canopy, and two infrared canopy temperature sampling points were set up at a height of 1.1 to 1.3 meters above the canopy. Outdoor temperature, outdoor relative humidity, solar irradiance, and outdoor wind speed were obtained by meteorological data acquisition equipment located outside the greenhouse. Each control zone was equipped with adjustable ventilation windows, variable-speed circulating fans, hot water heating valves, atomizing humidification equipment, and internal shading equipment. Each device provided feedback on its actual position, rotation speed, valve opening, or operating status. The sensor sampling interval was set to 30 seconds, the environmental control decision cycle was set to 5 minutes, and the prediction time domain was set to 30 minutes, forming 6 prediction time points according to a prediction step size of 5 minutes. Before the experiment began, step calibration was performed on each environmental regulation device with different control directions and control amplitudes to obtain the initial action start-up delay, initial action rise time, residual action time after initial stop, and environmental regulation contribution of the initial unit control quantity used in the cold start phase; at the same time, the initial prediction error boundary was determined based on 24-hour baseline operating data.

[0452] The invention team first acquires air temperature, relative humidity, crop canopy temperature, outdoor meteorological data, actual control values ​​of each environmental regulation device, and historical control feedback data for each control cycle. The range of leaf vapor pressure deficit from 0.50 kPa to 1.50 kPa and leaf dew margin not less than 1.00 °C is defined as the leaf transpiration-dew bilateral safety domain. Subsequently, the canopy environmental trajectory range within the next 30 minutes is predicted while maintaining the current environmental regulation device control state, and the time range within which the lower boundary of leaf vapor pressure deficit, the upper boundary of leaf vapor pressure deficit, or the lower boundary of leaf dew margin may cross the corresponding safety boundary is determined. Based on historical control response samples matched to the current operating conditions, the index-level action sequence and unit control value environmental regulation contribution of each environmental regulation device to leaf vapor pressure deficit and leaf dew margin are determined. Based on the deviation between the predicted trajectory and the bilateral safety domain, the environmental correction value that needs to be formed before the start time of the leaf safety boundary crossing time window is determined, and candidate environmental control action sequences containing different device combinations, action start times, control value change amplitudes, and control durations are generated.

[0453] For each candidate environmental control action sequence, up to 20 historical response calculation scenarios are constructed. The cumulative effective environmental correction, earliest arrival time, latest arrival time, and scenario accessibility under different combinations of device response samples are evaluated. A conservatively predicted canopy environment trajectory envelope is formed by combining historical prediction errors, device interaction uncertainties, and duration extrapolation uncertainties. Swarm intelligence optimization uses 30 candidate individuals, with a maximum of 25 iterations. A control timing margin of at least 2 minutes, an effective regulation margin of at least 0.05, and the conservatively predicted canopy environment trajectory envelope satisfying bilateral safety domain constraints are considered fully feasible conditions. Simultaneously, the magnitude of device control quantity changes, runtime, and number of switching operations are evaluated. When no fully feasible candidate sequence exists, the emergency environmental control action sequence is determined based on the degree of violation of safety constraints, timing constraints, and regulation constraints of different candidate sequences. The invention group adopts a rolling execution method, executing only the device operations corresponding to the current 5-minute control cycle. Upon entering the next control cycle, the state is reacquired, and prediction and optimization are completed.

[0454] The control group used traditional air temperature and humidity threshold control. When the relative humidity was higher than 90%, the ventilation windows and circulating fans were activated at a preset fixed opening; when the air temperature was lower than 18.0 degrees Celsius, the heating equipment was activated at a preset fixed opening; and the equipment was stopped when the corresponding environmental parameters recovered to the threshold hysteresis range. The control group did not calculate the blade safety boundary crossing time window, did not identify the index-level action sequence of the environmental control equipment, and did not optimize the actions based on conservative environmental trajectories and historical response scenarios of multiple devices. To make the evaluation indicators of the two methods comparable, after the experiment, the control timing margin, effective regulation margin, and achievable proportion of historical response calculation scenarios of the control group were evaluated offline based on the actual equipment control trajectory and the actual canopy environment trajectory. The offline evaluation results were not used in the actual control of the control group.

[0455] During the experiment, the following parameters were recorded for each risk event: the start time of the safety boundary crossing window, the latest arrival time of the equipment's effective action, control timing margin, effective regulation margin, satisfaction of conservative envelope safety constraints, actual duration of leaf vapor pressure deficit exceeding the limit, duration of insufficient leaf condensation margin, equipment operating time, energy consumption per unit area, and number of control quantity switching. In each control zone, 100 functional leaves with fixed positions were selected, and leaf surface images were acquired every 2 minutes. Leaves forming continuous water films or identifiable water droplets were identified as leaves with visible condensation, and the proportion of visible condensation on the leaf surface was calculated. After each control cycle, the invention team identified single-equipment response segments and joint control response segments that met the identifiable conditions based on the actual feedback control quantity of the equipment and the actual canopy environment trajectory. The historical prediction error boundaries, historical control response samples, and directional initial equipment action parameters corresponding to different prediction lead times were updated. The prediction error during the cold start phase was calculated using data from the first two days of the experiment, and the prediction error during the stable phase was calculated using data from days 10 to 14. The control group used the same initial gray box model as the invention group for offline prediction, but did not update the prediction error boundary and equipment operating parameters to evaluate the impact of rolling feedback correction. The experimental results are shown in Table 1.

[0456] Table 1: Results of the Greenhouse Zone Pairing Control Experiment

[0457] The experimental results show that the invention group and the control group maintained good pairing consistency in terms of initial air temperature, relative humidity, crop canopy temperature, outdoor temperature, solar irradiance, and outdoor wind speed. For example, the initial relative humidity of the invention group F3 and the control group C3 were 92.3% and 92.1%, respectively; the initial canopy temperatures were 16.9°C and 17.0°C, respectively; and the outdoor temperatures during the event were 11.8°C and 11.9°C, respectively. Therefore, the differences in subsequent control results mainly reflect the differences between the two control methods in risk prediction, equipment dynamic response handling, and multi-equipment action coordination, rather than being caused by significant differences in initial operating conditions. Regarding the timeliness of environmental control actions, the average control time margin of the invention group was 3.58 minutes, and the control time margins of all four invention groups were positive, indicating that the candidate equipment actions had already created the required environmental effect before the earliest possible crossing of the leaf safety boundary. The control group had an average control timing margin of -2.95 minutes obtained from offline playback, indicating that traditional threshold control typically activates the equipment only after the air temperature and humidity reach the trigger threshold. However, opening ventilation windows, accelerating circulating fans, and raising the temperature of heating equipment all require equipment startup and environmental response transmission processes, resulting in the actual effective action occurring later than the occurrence of canopy safety risks. The invention group achieved an average historical response calculation scenario reachability of 0.893, significantly higher than the control group's 0.525. This demonstrates that by selecting candidate actions based on the timing of indicator-level equipment action, the environmental adjustment contribution per unit control quantity, and multiple historical response scenarios, it is possible to reduce control sequences that are effective only under ideal equipment response conditions but fail to achieve the correction target under slower response conditions.

[0458] Regarding the sufficiency of adjustment intensity and the maintenance of safety constraints, the average effective adjustment margin of the invention group was 0.188, with all four invention groups maintaining a positive value; the average of the control group was -0.160, indicating that the cumulative effective environmental correction amount formed by traditional threshold actions before the occurrence of risks is usually insufficient. The invention group conservatively predicted that the canopy environmental trajectory envelope met the safety constraints in 94.73% of cases, while the control group only achieved 61.05%. It should be noted that the safety constraint satisfaction rate of the invention group was not set to 100%, and the utilization rate of emergency environmental control action sequences was 3.3% to 8.3%. This reflects that under conditions of low outdoor temperature, high humidity, or limited equipment adjustment capabilities, there may still be no candidate actions that fully meet all constraints, which is consistent with the objective situation of limited equipment capabilities and the inability to completely eliminate environmental disturbances in actual greenhouse operation.

[0459] The average duration of leaf vapor pressure deficit exceeding the limit in a single risk event for the invention group was 3.20 minutes, a decrease of approximately 81.5% compared to the control group's 17.28 minutes; the duration of insufficient leaf dew margin decreased from 13.40 minutes in the control group to 2.08 minutes, a reduction of approximately 84.5%; and the proportion of visible dew on the leaf surface decreased from 16.38% to 2.15%, a reduction of approximately 86.9%. Among these, the operating conditions corresponding to F3 and C3 were the most unfavorable. The control timing margin for F3 was only 2.6 minutes, the effective regulation margin was 0.12, and the conservative envelope safety constraint satisfaction rate was 92.1%; however, its visible dew proportion on the leaf surface was still controlled at 3.4%. The corresponding control timing margin for C3 was -4.3 minutes, the effective regulation margin was -0.28, and the visible dew proportion on the leaf surface reached 24.8%. The results indicate that using both leaf vapor pressure deficit and leaf condensation margin as safety constraints at the leaf level can identify the risks of canopy temperature approaching the dew point and insufficient leaf transpiration drive earlier than relying solely on air temperature and relative humidity to trigger device operation.

[0460] In terms of equipment operating economy, the average environmental regulation energy consumption per unit area for 30 minutes was 14.15 Wh / m² in the invention group, compared to 18.85 Wh / m² in the control group, a reduction of approximately 24.9%. The average number of control quantity switching times per risk event in the invention group was 3.53, compared to 7.10 in the control group, a reduction of approximately 50.4%. The cumulative operating time of heating equipment and the cumulative action time of ventilation windows in the invention group were reduced by approximately 35.4% and 44.9% respectively compared to the control group. The control action cost index decreased from an average of 0.843 in the control group to an average of 0.600 in the invention group, a reduction of approximately 28.8%. This indicates that the present invention does not achieve a safety effect by maintaining high heating power or large ventilation opening for a long time, but rather selects a combination of equipment that can form a target correction effect at an appropriate time based on the direction, response time, and adjustment contribution of different equipment to the leaf vapor pressure deficit and leaf condensation margin, thereby reducing over-adjustment, mutual cancellation of equipment, and frequent start-stop near the threshold.

[0461] The invention group's average time from risk identification to stable satisfaction of bilateral safety constraints was 8.35 minutes, compared to 18.43 minutes for the control group, a reduction of approximately 54.7%. This indicator does not represent a recovery after a boundary breach, but rather the time elapsed from the prediction and identification of potential risks until both the conservative envelope and the actual trajectory reached a stable and safe state. The invention group was able to pre-plan equipment actions before actual boundary breaches, thus significantly shortening the time to achieve a stable and safe state.

[0462] Regarding rolling feedback correction, the invention group's average absolute error in predicting leaf vapor pressure deficit during the cold start phase was 0.107 kPa, decreasing to 0.060 kPa during the stable phase, a reduction of approximately 44.0%; the average absolute error in predicting leaf condensation margin decreased from 0.330 degrees Celsius to 0.168 degrees Celsius, a reduction of approximately 49.2%. The median relative error of the environmental regulation contribution per unit control quantity decreased from 17.23% during the cold start phase to 7.95% during the stable phase, a reduction of approximately 53.8%. In contrast, the control group, which did not update the prediction error boundary and equipment operating parameters, only saw its average absolute error in predicting leaf vapor pressure deficit change from 0.115 kPa to 0.112 kPa, and its average absolute error in predicting leaf condensation margin change from 0.355 degrees Celsius to 0.345 degrees Celsius, showing no significant continuous improvement. During the experiment, the invention group added 61 to 75 valid single-device response samples in each control area and 14 to 18 joint control response samples, providing practical basis for condition matching, determination of equipment interaction uncertainty boundaries and parameter correction in subsequent control cycles.

[0463] In summary, this embodiment demonstrates that the present invention identifies canopy risks in advance through the leaf transpiration-condensation bilateral safety domain and safety boundary crossing time window; evaluates the arrival time and intensity of equipment action by using the timing of index-level equipment action and the environmental regulation contribution per unit control quantity; avoids selecting control sequences solely based on ideal responses or central prediction values ​​by calculating scenarios from historical responses and conservatively predicting canopy environmental trajectory envelopes; and achieves collaborative optimization among environmental safety, control timeliness, regulation adequacy, and action costs through swarm intelligence algorithms. Simultaneously, actual control feedback continuously corrects prediction error boundaries and equipment response parameters. Therefore, the present invention can significantly shorten the duration of leaf vapor pressure deficit exceeding limits and insufficient leaf condensation margin under conditions of lower equipment switching frequency and lower energy consumption per unit area, and improves the adaptability of greenhouse environmental control to differences in equipment performance, historical response dispersion, and changes in operating conditions.

[0464] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for controlling the environment of a greenhouse based on swarm intelligence, characterized in that, include: S1: Acquire greenhouse environmental status data, crop canopy status data, outdoor meteorological data, environmental regulation equipment operation data, and historical control feedback data for the current control cycle; S2: Determine the leaf transpiration-condensation bilateral safety domain based on the greenhouse environmental status data and crop canopy status data. Based on the greenhouse environmental status data, crop canopy status data, outdoor meteorological data, and historical control feedback data, predict the canopy environment trajectory range to maintain the current equipment control state, and determine the leaf safety boundary crossing time window. S3: Based on the operating data of the environmental control equipment and the historical control feedback data, determine the timing of the action of each environmental control equipment on the index-level equipment and the environmental control contribution per unit control quantity for the current operating conditions of the blade vapor pressure deficit and blade condensation margin. S4: Determine the environmental correction amount and generate candidate environmental control action sequences based on the deviation between the canopy environment trajectory interval and the bilateral safety domain. Construct a historical response calculation scenario based on the index-level time series and unit contribution, and determine the effective arrival time window, representative predicted canopy environment trajectory, and conservative predicted canopy environment trajectory envelope for each candidate sequence. S5: The candidate sequences are iteratively updated using a swarm intelligence algorithm. In each iteration, the effective arrival window, the representative predicted canopy environment trajectory, and the conservative predicted canopy environment trajectory envelope of each candidate environmental control action sequence are redefined. The control timing margin is determined based on the crossing and arrival windows, and the effective adjustment margin is determined based on the cumulative effective correction and environmental correction before the start time of the crossing window. The strategy is evaluated by combining the conservative predicted envelope, timing margin, and adjustment margin. After the iteration terminates, the best candidate sequence is selected from those whose conservative envelope satisfies the safety domain constraint and whose timing margin and adjustment margin are not lower than the corresponding threshold. If there is no candidate sequence that satisfies the constraints, the emergency sequence is determined based on the degree of constraint violation. S6: Select the preferred or emergency sequence as the sequence to be executed and output the equipment control command. After obtaining the actual canopy environment trajectory, update the prediction error and boundary according to the deviation between the actual control path and the predicted trajectory. Update the historical samples according to the identifiable actual response segments, correct the index-level time series and unit contribution, and use the updated results for the next cycle.

2. The greenhouse environment control method based on swarm intelligence as described in claim 1, characterized in that, In step S1: The greenhouse environmental status data includes air temperature and relative humidity. Crop canopy status data includes crop canopy temperature; Outdoor meteorological data includes outdoor temperature, outdoor relative humidity, solar irradiance, and outdoor wind speed; Environmental control equipment operation data includes the current control variables and historical control variable sequences for each device; Historical control feedback data includes historical actual canopy environment trajectories and historical prediction error data.

3. The greenhouse environment control method based on swarm intelligence as described in claim 2, characterized in that, In step S2, determining the bilateral security domain includes: The actual vapor pressure and dew point temperature of the air are determined based on the air temperature and relative humidity, and the saturated vapor pressure of the canopy is determined based on the crop canopy temperature. The difference between the saturated vapor pressure of the canopy and the actual vapor pressure of the air is taken as the leaf vapor pressure deficit, and the difference between the crop canopy temperature and the air dew point temperature is taken as the leaf dew margin. The range of conditions in which the vapor pressure deficit of the blades is not lower than a preset lower limit and not higher than a preset upper limit, and the condensation margin of the blades is not lower than a preset lower limit, is defined as the bilateral safety domain.

4. The greenhouse environment control method based on swarm intelligence as described in claim 3, characterized in that, In step S2, predicting the canopy environment trajectory interval and determining the crossing time window includes: Acquire the prediction sequence of outdoor meteorological data within the preset prediction time domain, and combine it with the greenhouse status, crop canopy status, current control values ​​of each device and historical actual trajectory to predict the baseline trajectory of leaf vapor pressure deficit and leaf dew margin to maintain the current control state. Based on historical prediction errors, the error range for each prediction time is determined, forming the upper and lower bound trajectories of the two indicators, thus obtaining the canopy environment trajectory range. The earliest possible crossing time of each safety boundary is defined as the moment when the lower boundary of the leaf vapor pressure deficit first falls below its lower limit, the upper boundary first rises above its upper limit, and the lower boundary of the leaf condensation margin first falls below its lower limit. The moment when the upper limit of leaf vapor pressure deficit first falls below its lower limit, the moment when the lower limit first rises above its upper limit, and the moment when the upper limit of leaf condensation margin first falls below its lower limit are taken as the latest possible crossing time of each safety boundary. The earliest time among the earliest possible crossing times corresponding to each safety boundary with predicted crossing is taken as the start time of the blade safety boundary crossing time window, and the latest time among the latest possible crossing times is taken as the end time of the blade safety boundary crossing time window.

5. The greenhouse environment control method based on swarm intelligence as described in claim 4, characterized in that, Step S3 specifically includes: Based on the historical control quantity sequence of each device, determine the timing of control quantity changes, the magnitude of signed changes, and the direction of changes, and extract trajectory segments within the preset response observation period from the historical actual trajectory to form historical control response samples; For leaf vapor pressure deficit and leaf condensation margin, the starting time of the effect relative to the dynamic baseline trend before the change, the time when the effect reaches the preset effective response threshold, and the time when the effect subsides after the control quantity is restored are determined respectively. Based on the above time, the respective indicator-level action start delay, action rise time and residual action time after stopping of the two indicators are determined to form the indicator-level equipment action timing sequence. Based on the signed change amplitude and the effective change amount within the corresponding effective response statistical period, determine the environmental regulation contribution of each device to the unit control quantity of the two indicators; The time series and unit contribution of the index are saved according to the direction of change. The samples in the same direction are matched and weighted according to the difference between the current working condition and the working conditions of each sample to obtain the time series and unit contribution of the index under the current working condition.

6. The greenhouse environment control method based on swarm intelligence as described in claim 5, characterized in that, Step S4 specifically includes: Based on the deviation of the canopy environmental trajectory interval relative to the bilateral safety domain during the correction requirement assessment period starting from the start time of the crossing time window, the leaf vapor pressure deficit correction amount and the leaf dew margin correction amount formed no later than the start time are determined to constitute the environmental correction amount. Candidate environmental control action sequences are generated based on the current control variables and adjustable ranges of each device. Each sequence includes the action start time, signed change amplitude, and duration of each device. For each candidate sequence, sample-level index-level time series and unit contribution are selected from the matching samples of the devices participating in the action in the same direction, and combined to form a historical response calculation scenario; In each scenario, the cumulative effective correction amount of the two indicators is calculated based on the indicator-level time sequence and unit contribution of each device, and the arrival time to reach the corresponding required correction amount is determined. Based on this, the earliest and latest arrival times of the candidate sequence and the proportion of reachable scenarios are determined, forming an effective arrival window. Based on the baseline trajectory maintained under the current control state and the cumulative correction amount of each scenario, a scenario prediction trajectory is generated, and a representative predicted canopy environment trajectory is obtained by weighting according to the scenario weights. By combining the predicted trajectories for each scenario, the historical prediction error boundaries for different prediction lead times, the uncertainties of device interaction, and the extrapolation uncertainties of duration, a conservative prediction envelope for the canopy environment trajectory is formed.

7. The greenhouse environment control method based on swarm intelligence as described in claim 6, characterized in that, In step S5, determining the control timing margin and effective regulation margin includes: The control timing margin is obtained by subtracting the latest arrival time of the arrival time window from the start time of the travel time window. When there is a need for correction of leaf vapor pressure deficit, the cumulative correction amount of leaf vapor pressure deficit at the beginning of each scenario is projected in the direction of the correction amount, and the minimum value of the projection result of each scenario is taken as the conservative effective correction amount of leaf vapor pressure deficit. When there is a need to correct the condensation margin, the minimum cumulative correction amount of the condensation margin at the beginning of each scenario is taken as the conservative effective correction amount of the condensation margin. The difference between each conservative effective correction and the corresponding required correction is divided by the larger of the absolute value of the required correction and the preset normalized protection value to obtain the normalized adjustment margin of the two indicators. The minimum value among the margins of indicators with correction needs is taken as the effective adjustment margin. When the environmental correction is zero, the timing margin and adjustment margin are not considered as feasibility constraints.

8. The greenhouse environment control method based on swarm intelligence as described in claim 7, characterized in that, In step S5, the iterative update and policy evaluation using swarm intelligence algorithms include: Each candidate sequence is treated as an individual, and the start time, change range, and duration of each device's action are used as control variables. Each iteration updates variables based on information interaction between individuals, limiting the magnitude of change to an adjustable range and limiting the timing and duration of actions to the prediction time domain; The scenario is reconstructed for each updated sequence, and the arrival window, reachability ratio, representative trajectory and conservative envelope, as well as the temporal margin and adjustment margin are redefined. The minimum normalized safety distance is determined based on the distance between the conservative envelope and each safety boundary. After normalizing the timing margin, adjustment margin, and control action cost, the strategy evaluation result is determined, where the control action cost is determined based on the change magnitude, duration, and number of switching operations. After the iteration terminates, the best strategy evaluation sequence is selected from the candidate sequences that satisfy the safety domain, time margin and adjustment margin not lower than the threshold and meet the accessibility constraint and equipment operation constraint. If no sequence satisfies all constraints, an emergency sequence is determined based on the degree of violation of safety, timing, adjustment, and reachability constraints, and an identifier for no fully feasible control strategy and a result of constraint violation are generated.

9. The greenhouse environment control method based on swarm intelligence as described in claim 8, characterized in that, Step S6 specifically includes: The preferred environmental control action sequence, or the emergency environmental control action sequence when no fully feasible control strategy identifier is generated, is determined as the environmental control action sequence to be executed in the current control cycle; The control command corresponding to the current cycle is sent to each device for execution, and the actual feedback control quantity, air temperature, air relative humidity and crop canopy temperature are collected to form the actual canopy environment trajectory; The path difference between the actual control process and the planned process is determined based on the actual feedback control quantity. When the difference does not exceed the preset threshold, the representative trajectory corresponding to the sequence to be executed is used as the predicted trajectory corresponding to the actual control process. Otherwise, the actual action sequence is generated based on the actual feedback control quantity and the predicted trajectory is regenerated. The actual trajectory is compared with the predicted trajectory corresponding to the actual control process according to the prediction lead time to obtain the prediction error of the two indicators. The historical prediction error set and its upper and lower deviation boundaries of each prediction lead time are updated for the formation of the trajectory interval in the next cycle. Establish equipment response tracking records based on the actual feedback control quantity change process, extract actual control response segments from closed records, add segments that meet the single device identification conditions to the historical sample set according to device, change direction and effective response index, and add segments with overlapping effective effects of multiple devices that cannot be separated to the joint sample set. Based on the segments that meet the identifiable conditions, determine the actual indicator-level effect start-up delay, rise time, residual effect time after stopping, and actual unit contribution of the two indicators, and perform confidence weighting and single change limit update on the directional initial parameters in the same direction; The actual trajectory, historical prediction error set, deviation boundary, updated historical sample set, joint sample set, and directional initial parameters are used as historical control feedback data for the next cycle. In the next cycle, the operating condition matching is re-executed to determine the index-level time series and unit contribution.

10. A greenhouse environment control system based on swarm intelligence, used to implement the greenhouse environment control method based on swarm intelligence as described in any one of claims 1 to 9, characterized in that, include: Data acquisition module: Acquires greenhouse environmental status data, crop canopy status data, outdoor meteorological data, environmental regulation equipment operation data, and historical control feedback data for the current control cycle; Safety risk prediction module: Based on the greenhouse environmental status data and crop canopy status data, determine the leaf transpiration-condensation bilateral safety domain; based on the greenhouse environmental status data, crop canopy status data, outdoor meteorological data and historical control feedback data, predict the canopy environment trajectory range to maintain the current equipment control state, and determine the leaf safety boundary crossing time window; Equipment Operation Parameter Determination Module: Based on the operating data of the environmental regulation equipment and historical control feedback data, determine the timing of the operation of each environmental regulation equipment and the environmental regulation contribution per unit control quantity to the indicator-level equipment for the current operating conditions of the blade vapor pressure deficit and blade condensation margin. Candidate control action generation module: Determines the environmental correction amount and generates candidate environmental control action sequences based on the deviation between the canopy environment trajectory interval and the bilateral safety domain; Constructs a historical response calculation scenario based on the index-level time series and unit contribution; Determines the effective arrival time window, representative predicted canopy environment trajectory, and conservative predicted canopy environment trajectory envelope for each candidate sequence. The control strategy optimization module utilizes a swarm intelligence algorithm to iteratively update candidate sequences. In each iteration, the effective arrival window, the representative predicted canopy environment trajectory, and the conservative predicted canopy environment trajectory envelope of each candidate environmental control action sequence are redefined. The control timing margin is determined based on the crossing and arrival windows, and the effective regulation margin is determined based on the cumulative effective correction and environmental correction before the start of the crossing window. The strategy is evaluated by combining the conservative predicted envelope, timing margin, and regulation margin. After the iteration terminates, the best candidate sequence is selected from those whose conservative envelope satisfies the safety domain constraint and whose timing margin and regulation margin are not lower than the corresponding thresholds. If no candidate sequence satisfies the constraints, an emergency sequence is determined based on the degree of constraint violation. Control execution and feedback correction module: It takes the preferred or emergency sequence as the sequence to be executed and outputs the equipment control command. After obtaining the actual canopy environment trajectory, it updates the prediction error and boundary according to the deviation between the actual control path and the predicted trajectory. It updates the historical samples according to the identifiable actual response segments, corrects the index-level time series and unit contribution, and uses the updated results for the next cycle.

Images