Self-adaptive regulation and control method and system for greenhouse environment

Abstract

The invention provides a greenhouse environment adaptive regulation and control method and system, and relates to the technical field of environment control, and the method comprises the steps: collecting multi-dimensional environment parameters in a greenhouse; based on the environmental parameters, a preset crop growth period database and weather prediction data, taking minimization of a preset cost function as a target, and adopting a model prediction control algorithm to generate an equipment linkage instruction set in a future preset time period, the preset cost function fusing environmental regulation and control deviation and operation cost; issuing the equipment linkage instruction set to each execution equipment, and executing linkage regulation and control; wherein when the equipment linkage instruction set is generated, a conflict resolution mechanism based on a dynamic priority is adopted to determine an execution sequence of a plurality of equipment instructions; the adaptive regulation and control method integrating multi-source data verification, multi-scale prediction, dynamic priority conflict resolution and multi-target cost optimization improves the accuracy, economy and crop suitability of greenhouse environment regulation and control.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of environmental control, and in particular to a greenhouse environment self-adaptive regulation method and system. BACKGROUND

[0002] In the field of facility agriculture, the greenhouse is an artificial controllable environment carrier for crop growth, and the precise regulation of its environmental parameters (such as temperature, humidity, light, and CO2 concentration) directly determines the yield and quality of crops. Traditional greenhouse regulation mostly adopts a "threshold trigger" mode, that is, through a single sensor to monitor the environmental parameters, when the parameters exceed the preset fixed threshold, the start-stop operation of heating, ventilation, light supplementing and other devices is triggered. However, this mode has significant technical limitations: on the one hand, a single sensor is easily disturbed by local environment, leading to data distortion, and it is difficult to accurately reflect the overall environmental state of the greenhouse; the effectiveness of the regulation instruction is affected; on the other hand, fixed threshold control can only cope with the current environmental deviation, and cannot predict the influence of future weather changes (such as cold wave, rain) on the greenhouse environment, leading to lag of device operation, which easily causes stress reaction of crops; the current control easily causes energy waste and environmental parameter oscillation. SUMMARY

[0003] Based on the above problems, the present application provides a greenhouse environment self-adaptive regulation method and system, which integrates multi-source data verification, multi-scale prediction, dynamic priority conflict resolution, and multi-objective cost optimization self-adaptive regulation method, to improve the precision, economy and crop adaptability of greenhouse environment regulation.

[0004] In a first aspect, the present application provides a greenhouse environment self-adaptive regulation method, which comprises:

[0005] Collecting multi-dimensional environmental parameters in the greenhouse;

[0006] Based on the environmental parameters, a preset crop growth period database and weather prediction data, a model predictive control algorithm is used to generate a device linkage instruction set for a future preset time period, with the objective of minimizing a preset cost function, wherein the preset cost function integrates environmental regulation deviation and operating cost;

[0007] Downlinking the device linkage instruction set to each execution device to perform linkage regulation;

[0008] Wherein, when generating the device linkage instruction set, a conflict resolution mechanism based on dynamic priority is used to determine the execution order of multiple device instructions.

[0009] Preferably, the model predictive control algorithm synchronously generates an environmental parameter reference trajectory for a future preset time period in the process of generating the device linkage instruction set; the trajectory is an environmental target expected to be achieved by the device linkage instruction set, and the generation of the environmental parameter reference trajectory is premised on meeting the physiological needs of crops, with the core objective being to minimize the deviation of the comprehensive physiological suitability index in the preset cost function.

[0010] Preferably, the comprehensive physiological suitability index is obtained by:

[0011] By means of a crop growth period database, the ideal range and stress threshold of each environmental parameter under the current crop growth stage are obtained; based on the ideal range and stress threshold, a single-factor suitability function is constructed for each environmental parameter, which maps the measured or predicted value of the environmental parameter into a suitability score between 0 and 1;

[0012] According to the current crop growth stage, a dynamic first weight is assigned to the suitability score of each environmental parameter;

[0013] The suitability scores of the environmental parameters are weighted and summed with their corresponding dynamic weights to obtain the comprehensive physiological suitability index.

[0014] Preferably, the specific steps of the model predictive control algorithm include:

[0015] Receive weather forecast data for a future preset time period as a feedforward input;

[0016] Establish a dynamic prediction model of the greenhouse environment for simulating changes in the environmental parameters inside the greenhouse after different operation combinations of the execution devices are performed;

[0017] Construct a multi-objective cost function, which includes a first term representing the deviation of physiological benefits and a second term representing economic costs; wherein the first term is the minimization of the deviation between the cumulative comprehensive physiological suitability index and the ideal value within the future preset time period, and the second term is the minimization of the total energy consumption of the execution devices;

[0018] Take the device operation sequence to be solved within the future preset time period as a decision variable, and based on the dynamic prediction model, solve the multi-objective cost function under the condition of meeting the physiological constraints of crops to obtain the optimal device operation sequence that minimizes the function value, which is the device linkage instruction set.

[0019] Preferably, the dynamic prediction model of the greenhouse environment is constructed and trained by:

[0020] The multi-time series composed of a historical device operation sequence, a historical environment parameter sequence, and historical external weather data is taken as input, and the prediction value of future environment parameters and its confidence interval are output, and the model is trained by minimizing a loss function;

[0021] The dynamic prediction model adopts a multi-scale time series fusion architecture for extracting multi-scale features of environment parameter instantaneous fluctuations, short-term trends, and long-term regularities, and dynamically fusing the multi-scale features through an attention mechanism;

[0022] The loss function includes a first loss term and a second loss term, the first loss term is the mean square error of the prediction value and the measured value, and the second loss term is a coupling loss term;

[0023] The second loss term is calculated by the following formula:

[0024]

[0025] The coupling loss term is L_coupling; is a model predicted environment parameter change matrix; C is a predefined coupling coefficient matrix; and |||2 is an L2 norm.

[0026] Preferably, after collecting the environment parameters, the method further includes a data fusion step, including:

[0027] For multiple sensors monitoring the same environment parameter, a dynamic trust evaluation mechanism is used to assign a second weight to each sensor reading;

[0028] The second weight is used to weight average the readings of the multiple sensors to obtain a comprehensive estimate of the environment parameter;

[0029] The comprehensive estimates of all environment parameters are combined into a comprehensive environment state vector;

[0030] Physical rules based on energy conservation and / or transpiration models are used to verify the comprehensive environment state vector.

[0031] Preferably, the second weight is obtained by:

[0032] The comprehensive trust score of the sensor is obtained; the comprehensive trust score is determined by the instantaneous abnormal score, the time series consistency score, the spatial consistency score, and the external goodness of fit score;

[0033] According to the comprehensive trust scores of all sensors monitoring the same environment parameter in the greenhouse, the second weight of each sensor for the environment parameter is determined.

[0034] Preferably, in the conflict resolution mechanism based on dynamic priority, the priority rules include: the priority of instructions for ensuring the survival of crops is the highest, the priority of instructions for adapting the growth of crops is second, and the priority of instructions for optimizing resource consumption is the lowest.

[0035] Preferably, the environmental parameters include temperature, humidity, light, and CO2 concentration, and in the priority rules, the priority of temperature regulation is higher than that of humidity regulation, the priority of humidity regulation is higher than that of light and CO2 regulation; and the priority weight of each environmental parameter is dynamically adjusted according to the current crop growth stage and real-time environmental state.

[0036] In a second aspect, the present application provides a greenhouse environment adaptive control system for implementing the greenhouse environment adaptive control method described in the embodiments of the present application, and the system comprises:

[0037] a data acquisition module for acquiring multi-dimensional environmental parameters in the greenhouse;

[0038] an instruction generation module for generating a device linkage instruction set for a future preset time period based on the environmental parameters, a preset crop growth period database, and weather prediction data, with a model predictive control algorithm, and with the goal of minimizing a preset cost function that combines environmental regulation deviation and operating cost;

[0039] an execution module for issuing the device linkage instruction set to each execution device for linkage control;

[0040] In the generation of the device linkage instruction set, a conflict resolution mechanism based on dynamic priority is used to determine the execution order of multiple device instructions.

[0041] Compared with the prior art, the present application has at least the following beneficial effects: a multi-scale time sequence fusion architecture and an attention mechanism are used to capture multi-scale features of environmental parameters, a coupling loss optimization model is combined to fully consider the parameter coupling effect, and reliable basis is provided for accurate control. Through a dynamic priority conflict resolution mechanism, the priority of instructions is determined according to crop survival, growth, and resource optimization, and the weight is adjusted according to the growth stage to reduce energy waste and parameter oscillation, balance crop demand and operating cost; the trust score of multi-dimensional evaluation sensors is evaluated to improve the quality of multi-source data fusion; logical contradictory information is excluded by combining physical rule verification data to ensure the accuracy of control decisions; a multi-objective cost function is combined with physiological benefits and economic costs, and the control target and strategy are dynamically adjusted according to the crop growth period to meet the growth needs of crops in the whole cycle, and the control effect and cost control are considered. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 is a schematic diagram of the greenhouse environment adaptive control method of the embodiments of the present application. DETAILED DESCRIPTION

[0043] In view of the deficiencies in the prior art, the applicant has long-term research and a large number of practices, and finally proposes the technical scheme of the present application. The technical scheme, its implementation process and principles will be further explained in the following combined with the drawings in the embodiments of the present application and specific implementation cases.

[0044] It should be noted that the embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as a limitation on the present application. The described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, any alternative, modification, equivalent method and scheme defined by the claims are covered by the spirit, principles and scope of the present application, and all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0045] Further, in order to make the public have a better understanding of the present application, in the following detailed description of the present application, some specific details are described in detail. The present application can also be completely understood without the description of these details by those skilled in the art.

[0046] Embodiment 1: refer to the accompanying drawings Figure 1 The present application provides a greenhouse environment adaptive control method, which comprises:

[0047] Collecting multi-dimensional environmental parameters in the greenhouse through a distributed sensor network;

[0048] The central control unit generates a set of device linkage instructions for a future preset time period based on the environmental parameters, a preset crop growth period database and weather prediction data, using a model predictive control algorithm, with the goal of minimizing a preset cost function, wherein the preset cost function integrates environmental control bias and operating cost;

[0049] The set of device linkage instructions is issued to each execution device to perform linkage control; the execution device includes a roller blind, a ventilation device, a heating device, a light supplementing device, a dehumidifying device, etc.

[0050] Wherein, when generating the set of device linkage instructions, a conflict resolution mechanism based on dynamic priority is used to determine the execution order of multiple device instructions.

[0051] The working principle of the above technical scheme is:

[0052] In order to capture the complex changes of the internal environmental factors of the greenhouse, the key parameters such as temperature, humidity, light, carbon dioxide (CO2) concentration are spatially layered and grid deployed; the environmental state is comprehensively perceived from the vertical and horizontal two dimensions.

[0053] On the vertical profile of the greenhouse, the environmental conditions at different heights are significantly different. For example, the air temperature and humidity near the plant canopy (the top of the leaves) can be quite different from the soil moisture at the ground. Therefore, multiple layers of sensors are deployed, one above the canopy to monitor the air temperature and humidity that directly affect plant physiological activities, one at the ground or in the soil to monitor the soil temperature and humidity required for root growth, and one at the top of the greenhouse to monitor the incoming light intensity and air parameters in the top space, ensuring accurate understanding of the environmental conditions in the key areas for crop growth.

[0054] On the horizontal plane of the greenhouse, due to structural shading, vent location, and different crop planting densities, local hot or cold spots can be formed; for example, shading nets can cause a sudden drop in light intensity in local areas, while the air outlets of the fans can form a local low-temperature and high-humidity area. In order to eliminate local deviations, a grid layout is used to evenly distribute sensor nodes within the greenhouse, forming a dense monitoring network. This can generate a thermal map of the greenhouse environment, accurately identify and handle local anomalies, and avoid affecting overall yield and quality due to environmental stress in individual areas.

[0055] After the central control unit receives the collected environmental parameters, it combines the pre-set crop growth period database (e.g., different environmental requirement data for tomato seedling and fruiting periods) and external weather prediction data (e.g., temperature and light change predictions for the next 24 hours), and uses a model predictive control algorithm to simulate the impact of different combinations of operating devices (roller blinds, ventilation equipment, etc.) on the environment, generating a set of device linkage instructions for a pre-set future time period (e.g., the next 6 hours). At the same time, during the generation of the instruction set, a conflict resolution mechanism based on dynamic priority is used to determine the execution order of each device instruction. The central control unit then issues the generated device linkage instruction set to the corresponding execution devices, such as roller blinds, ventilation equipment, heating equipment, etc., and each device works collaboratively according to the instruction sequence and requirements, achieving linkage control of the greenhouse environment.

[0056] The effects of the above technical solutions are:

[0057] The distributed sensor network can collect multi-dimensional environmental parameters in different areas of the greenhouse, avoiding the one-sidedness of single sensor monitoring and providing more comprehensive and accurate basic data for subsequent regulation. For example, in a large tomato greenhouse, by deploying temperature and humidity sensors in the east, west, south, and north regions and the central position, the problem of local temperature and humidity differences being ignored due to uneven air flow within the greenhouse can be effectively avoided.

[0058] The instruction set generation combines the crop growth period demand, weather prediction data, and minimizes the environmental regulation deviation and operation cost, to ensure that the greenhouse environment adapts to the crop growth and controls the operation cost. For example, when it is predicted that the temperature will drop in the next 12 hours, and the current crop is in the tomato fruiting stage (which requires higher temperature), the central control unit will generate instructions to start the heating device and close the roller shutter in advance, to avoid excessive environmental fluctuations caused by temporary regulation after the temperature drops, and to reduce unnecessary energy consumption by reasonably planning the heating device start time and power.

[0059] The conflict resolution mechanism based on dynamic priority can clearly define the execution order of device instructions, to avoid regulation failure or resource waste caused by device operation conflict. For example, when the greenhouse needs to be heated (start the heating device) and the humidity needs to be reduced (start the ventilation device) at the same time, if the crop is in the seedling stage (survival priority, more sensitive to temperature), the heating device start instruction is executed first, and then the ventilation device start instruction is executed after the temperature reaches the appropriate range, to prevent the temperature from not meeting the standard due to simultaneous device start, which affects the growth of seedlings, and to avoid the waste of energy.

[0060] The scientific instruction set for the cooperative work of each execution device can effectively reduce the fluctuation of greenhouse environmental parameters, to provide a stable environment for crop growth. For example, during the high temperature period in summer, the central control unit generates a linkage instruction to start the ventilation device to ventilate and cool down first, and then start the sunshade roller shutter to block part of the light if the temperature is still higher than the appropriate range, to avoid the problems of insufficient light caused by single dependence on the sunshade roller shutter, or poor cooling effect caused by single dependence on the ventilation device, to stabilize the temperature in the greenhouse within the range suitable for crop growth.

[0061] In one possible implementation, the model predictive control algorithm synchronously generates an environmental parameter reference trajectory for a future preset time period during the generation of the device linkage instruction set; the trajectory is the environmental target expected to be reached by the device linkage instruction set, and the generation of the environmental parameter reference trajectory is based on the premise of meeting the physiological needs of crops, to minimize the deviation of the comprehensive physiological suitability index in the preset cost function as the core target.

[0062] The preset crop growth period database divides the entire growth cycle of the crop into multiple key stages, and defines the specific needs of each stage for environmental parameters such as temperature, humidity, and light. Taking tomatoes and cucumbers as examples, their environmental preferences are completely different in different growth periods, and accurate identification of these preferences is a prerequisite for high quality and high yield.

[0063] For example, tomato seeds require high temperature (25-30°C) and moderate humidity (60-65%) during the germination stage to promote rapid germination. However, after entering the flowering stage, excessively high temperature (>30°C for 3 hours) can cause pollen vitality to decrease, with more than 60% of inactivation rate, thereby causing serious flower and fruit drop problems. Similarly, cucumber seedlings require temperature between 20-25°C and air humidity above 60% to promote healthy growth; but during the fruiting stage, air humidity needs to be strictly controlled below 60% to prevent diseases such as downy mildew.

[0064] Based on the preset crop growth period database, the physiological requirements of the current crop growth stage (e.g. tomato fruiting stage, cucumber seedling stage) are first determined, and the ideal range and stress threshold of environmental parameters such as temperature, humidity, light, and CO2 concentration are determined to ensure that the trajectory generation is always centered around meeting the physiological requirements of the crops.

[0065] The deviation of the minimum comprehensive physiological suitability index is embedded as the core target in the preset cost function, where the comprehensive physiological suitability index is scored by each environmental parameter.

[0066] Combined with the collected real-time environmental parameters and future weather forecast data, the influence of different device operation combinations on the environment is simulated, and while solving the minimum environmental regulation deviation and operation cost cost function, the environmental parameter reference trajectory for the future preset time period (e.g. 6 hours) is planned simultaneously. This trajectory clearly defines the target values that each environmental parameter in the greenhouse should reach at each time node (e.g. temperature maintained at 22-25°C and humidity at 60-70% per hour during the tomato fruiting stage), and the trajectory needs to ensure that the comprehensive physiological suitability index is as close to the ideal value as possible (with the smallest deviation). The final device linkage instruction set needs to match this reference trajectory, i.e. the execution effect of the instruction set needs to be able to push the actual environmental parameters towards the trajectory target values.

[0067] The environmental parameter reference trajectory is based on the physiological requirements of the crops and the minimum deviation of the comprehensive physiological suitability index, avoiding the limitations of traditional regulation where only fixed threshold values are set as targets, ensuring that the regulation direction is highly consistent with the growth requirements of the crops. The reference trajectory clearly defines the environmental target values for each period in the future, making the expected effect of the device linkage instruction set visualized, which facilitates the staff to predict whether the regulation direction is reasonable, and also allows for comparison of the deviation between the actual environmental parameters and the trajectory in subsequent regulation, tracing the optimization space of the instruction set.

[0068] By embedding the core objective of minimizing the deviation of the comprehensive physiological suitability index, the trajectory can balance the coordinated optimization of multiple environmental parameters (e.g., avoiding neglecting the influence of humidity on crop transpiration in pursuit of temperature compliance), and the instruction set can push the overall environmental parameters towards a high suitability direction, reducing the situation of single parameter compliance but overall unsuitability. For example, during the flowering period of cucumbers, the trajectory will simultaneously optimize temperature (25-28°C), humidity (70-75%), and CO2 concentration (800-1000 ppm), and the instruction set will control heating, humidification, and CO2 generator devices to maintain the comprehensive physiological suitability index above 0.9 (close to the ideal value), thereby improving the fruit setting rate. The generation of the reference trajectory makes the calculation of environmental regulation deviation more accurate (based on the trajectory target value, rather than the broad ideal range), and the multi-objective optimization of operating cost and comprehensive physiological suitability deviation can ensure crop benefits while avoiding energy waste caused by excessive regulation.

[0069] In one possible implementation, the comprehensive physiological suitability index (CPI) is obtained by including:

[0070] Obtaining the ideal range and stress threshold of each environmental parameter at the current crop growth stage;

[0071] Based on the ideal range and stress threshold, a single-factor suitability function is constructed for each environmental parameter, which maps the measured or predicted value of the environmental parameter to a suitability score between 0 and 1;

[0072] According to the current crop growth stage, a first weight is assigned to the suitability score of each environmental parameter;

[0073] The suitability scores of each environmental parameter are weighted and summed with their corresponding dynamic weights to obtain the comprehensive physiological suitability index.

[0074] The single-factor suitability function, such as the S-shaped function, cubic spline interpolation function, or parabolic function, is used to represent the change rule that the score approaches 1 when the environmental parameter value is in the ideal range, and smoothly decreases to 0 when it deviates from the ideal range; expressing the biological law of too low, moderate best, and too high.

[0075] For example, the steps for obtaining the temperature suitability function ST(T) for the flowering period of tomatoes include:

[0076] Obtaining the ideal range from the database: querying the crop growth period database to obtain the optimal temperature range for tomato flowering period as 20-25°C.

[0077] ST(T) = 1 when temperature is in the optimal range; ST(T) = 0 when temperature is below the lower stress threshold (e.g. 10°C) or above the upper stress threshold (e.g. 35°C), representing fatal stress.

[0078] Between the optimal range and the stress threshold range, the function value smoothly decreases from 1 to 0.

[0079] The first weight is preset in the crop growth period database and can be fine-tuned according to real-time environmental conditions; for example, during the flowering stage of tomatoes, the weights of temperature and humidity are set to the highest; during the fruiting stage of tomatoes, the weights of light and CO2 concentration are increased.

[0080] The single-factor suitability function maps the relationship between parameters and suitability through a smooth curve, avoiding the traditional threshold judgment (e.g. only setting the standard / meeting the standard), and accurately restoring the tolerance, suitability, and stress gradient response of crops to environmental parameters. For example, when the temperature during the flowering stage of tomatoes is 26°C (slightly higher than the ideal upper limit), the score will not drop sharply, but will smoothly decrease to about 0.9 according to the function, which is more in line with the adaptability of crops to slight parameter deviations in reality, reducing misjudgments.

[0081] The first weight is adjusted according to the growth stage of the crop, ensuring that the core environmental parameters have a more prominent impact on CPI, and avoiding one-size-fits-all weight settings. For example, during the fruiting stage of tomatoes, the weight of light is increased, and when the light is insufficient (e.g. a score of 0.5), even if other parameters have high scores (e.g. temperature 0.9, humidity 0.8), CPI will decrease due to the high weight of light, prompting the priority control of light equipment, which is in line with the needs of crops during the fruiting stage. By integrating multi-parameter scores through weighted summation, CPI can quantify the overall suitability of the greenhouse environment to crops, providing a clear target for regulation. The calculation logic of CPI is deeply bound to the physiological needs of crops, and the subsequent model predictive control algorithm generates an instruction set with the core goal of minimizing CPI deviation, ensuring that regulation is always centered around improving crop suitability.

[0082] In one possible implementation, the specific steps of the model predictive control algorithm include:

[0083] Receiving weather forecast data for a future preset time period as a feedforward input;

[0084] Establishing a dynamic prediction model of the greenhouse environment for simulating changes in environmental parameters inside the greenhouse after different operation combinations of the execution equipment;

[0085] Constructing a multi-objective cost function, which includes a first term representing physiological benefit deviation and a second term representing economic cost; wherein the first term is the minimization of the deviation between the cumulative comprehensive physiological suitability index and the ideal value in the future preset time period, and the second term is the minimization of the total energy consumption of the execution equipment.

[0086] The device operation sequence to be solved in the future preset time period is taken as a decision variable, and the multi-objective cost function is solved based on the dynamic prediction model under the condition of meeting the crop physiological constraint, so as to obtain an optimal device operation sequence that minimizes the function value, which is the device linkage instruction set.

[0087] The multi-objective cost function is solved by using a chance-constrained stochastic MPC algorithm to process the uncertainty in the weather forecast and the greenhouse model, and to ensure that the system meets all environmental constraints under a preset probability confidence.

[0088] The working principle of the above technical solution is as follows:

[0089] First, the weather forecast data (including temperature, light, wind speed, etc.) in the future preset time period (for example, the next 12 hours) is received as a feedforward input to provide an external environment basis for subsequent environment change prediction, avoiding the lag of regulation and control due to ignoring external weather influence.

[0090] A greenhouse environment dynamic prediction model is constructed, which is trained based on historical device operation sequences, historical environment parameters, and historical weather data, and can simulate the change trend of temperature, humidity, light, and other parameters in the greenhouse under different combinations of executed devices (such as heating, ventilation, and light supplement devices) (such as heating device on for 30 minutes, ventilation device on for 15 minutes, and only light supplement device on for 1 hour, etc.), providing simulation basis for subsequent optimization of device operation.

[0091] A multi-objective cost function including two core objectives is constructed:

[0092] The first term (physiological benefit bias term): the cumulative comprehensive physiological suitability index (CPI) in the future preset time period is minimized to ensure that the regulation meets the physiological needs of the crops;

[0093] The second term (economic cost term): the total energy consumption of the executed devices in the time period is minimized to control the operating cost and achieve the optimization of physiological benefit and economic cost.

[0094] For example, the economic cost can be calculated as the sum of the cost of a single action of all intermittent actuators (such as skylights and roller shutters) and the cost of continuous operation of continuous operation devices (such as heaters).

[0095] Rolling solution based on chance-constrained stochastic MPC: Set the device operation sequence (e.g., the start-stop state of the device every hour, the running power) in the future preset time period as the decision variable, while explicitly setting the crop physiological constraints (e.g., the temperature needs to be between 10℃ and 35℃, the humidity needs to be between 40% and 80%, etc.);

[0096] Considering the uncertainty of weather forecast data (e.g., actual light may be lower than the predicted value), greenhouse model (e.g., device operation efficiency fluctuation), a multi-objective cost function is solved by using a chance-constrained stochastic MPC algorithm; by setting a preset probability confidence (e.g., 95%), it is ensured that under the influence of uncertainty, all crop physiological constraints can still be met with this confidence;

[0097] The optimal operation sequence for the first sub-time period (e.g., the next 1 hour) is solved at the current time and executed, the updated weather forecast data is received at the next time, and the solving process is repeated to dynamically adjust the subsequent operation sequence, and finally the optimal device operation sequence that minimizes the cost function value is obtained, i.e., the device linkage instruction set.

[0098] The effects of the above technical solutions are:

[0099] Taking weather forecast data as a feedforward input, combining a dynamic prediction model to simulate environmental changes, the influence of external weather on the internal environment of the greenhouse can be predicted in advance, and passive response can be avoided. For example, if it is predicted that it will rain (light will weaken) in the next 6 hours, the algorithm can generate an operation sequence for starting the light supplementing device in advance to prevent the light in the greenhouse from dropping sharply after the rain, causing insufficient photosynthesis of the crops, and ensuring the stability of the cumulative CPI. The multi-objective cost function simultaneously considers the minimum deviation of the cumulative CPI and the minimum total energy consumption, avoiding the situation where only physiological efficiency is pursued, resulting in high energy consumption, or only energy saving is emphasized, affecting crop growth. For example, during the fruiting period of tomatoes, if the light is predicted to be sufficient (no need for light supplementation) in the future period, the algorithm will preferentially select an operation sequence in which the ventilation device is intermittently operated rather than the heating device is continuously operated, which not only maintains the temperature within the appropriate range (ensuring the CPI), but also reduces the high energy consumption of the heating device, thereby reducing operating costs. The chance-constrained stochastic MPC algorithm can cope with the uncertainty of weather forecasts and greenhouse models, and by setting a probability confidence, it can ensure that the environmental parameters meet the crop physiological constraints in most cases. For example, if the weather forecast shows that the humidity may drop to 35% (lower than the 40% constraint) in the future, the algorithm will generate an operation sequence that starts the humidification device in advance and appropriately increases the humidification time, with a probability of more than 95% to ensure that the humidity is not lower than 40%, avoiding the situation where the humidity is too low to cause stress to the crops and ensuring the stability of the system.

[0100] The rolling horizon method is adopted, and the operation sequence is adjusted in combination with the updated weather data and real-time environmental state in each period, so that the long-term fixed instruction set cannot adapt to the dynamic change of the environment. For example, if the actual weather is hotter (the temperature is higher than the predicted value) than the prediction, when the next rolling period is solved, the algorithm will timely adjust the operation sequence, increase the running time of the ventilation equipment, and start the sunshade roller in advance, so as to ensure that the temperature in the greenhouse is maintained in the appropriate range, and the environment parameter exceeds the standard due to the fixed instruction.

[0101] In a possible implementation, the dynamic prediction model of the greenhouse environment is constructed and trained in the following manner, comprising:

[0102] A multivariate time series composed of historical device operation sequences, historical environmental parameter sequences and historical external weather data is taken as input, and the output includes the predicted value and its confidence interval of the future environmental parameter, and the model is trained by minimizing the loss function;

[0103] The dynamic prediction model adopts a multi-scale time series fusion architecture for extracting multi-scale features of environmental parameter instantaneous fluctuations, short-term trends and long-term rules, and dynamically fusing the multi-scale features through an attention mechanism; three layers of feature extraction are realized through a nested LSTM architecture, wherein the bottom layer learns high-frequency coupling relationships, the middle layer fuses hour-level trends, and the top layer dynamically allocates weights of each scale through an attention mechanism;

[0104] The loss function includes a first loss term and a second loss term, the first loss term is the mean square error of the predicted value and the measured value, and the second loss term is a coupling loss term; the final loss is obtained by weighting the first loss term and the second loss term;

[0105] The second loss term is calculated by the following formula:

[0106]

[0107] Wherein, L_coupling is the coupling loss term; is the environmental parameter change matrix predicted by the model; C is a pre-defined coupling coefficient matrix, and an element C{ij} of the coupling coefficient matrix represents a theoretical influence coefficient of a unit parameter j on parameter i; |||2 is an L2 norm (Euclidean distance).

[0108] The coupling coefficient matrix C is obtained and updated through the following adaptive process:

[0109] Based on the greenhouse structure, the physical properties of the covering material and the basic physiological model of the crops, a basic coupling coefficient matrix C0 is initialized;

[0110] Continuously collect actual device operation sequences and environmental parameter response sequences; use the device operation and environmental state at the previous time as the cause and the environmental parameter change at the next time as the effect to obtain a real-time, data-driven coupling coefficient matrix C_data by using multivariate linear regression with regularization or Bayesian inference method;

[0111] The basic coupling coefficient matrix C0 and the data-driven coupling coefficient matrix C_data are weighted and fused to obtain the coupling coefficient matrix C used at the current time;

[0112] The weight is dynamically adjusted according to the statistical confidence of C_data; when the data quantity is sufficient and the quality is high, C_data is given a higher weight; otherwise, the basic model C0 is trusted more.

[0113] The working principle of the above technical solution is:

[0114] A multivariate time series composed of historical device operation sequences (such as heating, start and stop time of ventilation equipment, power), historical environmental parameter sequences (such as timing data of temperature, humidity, and light in the greenhouse), and historical external weather data (such as outdoor temperature, wind speed, and precipitation) is used as the model input, and the predicted value (such as the temperature prediction value every hour in the future 6 hours) and its confidence interval of the future environmental parameter are set as the model output to clearly define the data basis and target for subsequent training.

[0115] Multi-scale time series fusion architecture design and feature extraction: the model uses a nested LSTM architecture to realize multi-scale time series fusion and extracts features in three layers:

[0116] Focus on high-frequency data to learn the instantaneous fluctuation characteristics of environmental parameters (such as sudden changes in humidity caused by the start and stop of ventilation equipment) and the high-frequency coupling relationship between parameters (such as the rapid influence of temperature rise on humidity); high-frequency data refers to the original observation data collected by the sensor at a frequency higher than the macroscopic change rate of the environment to capture the instantaneous dynamic changes of environmental parameters and the rapid interaction relationship between parameters.

[0117] Integrate the bottom layer features to capture the short-term trends of environmental parameters (such as the change rule of temperature from low to high and then to low during the day, and the hourly humidity fluctuation trend); introduce an attention mechanism at the top layer to dynamically allocate weights to the bottom layer high-frequency features and the middle layer short-term trends according to the crop demand at different growth stages and the real-time environmental state, and simultaneously fuse long-term rules (such as the change of light duration caused by seasonal alternation), and finally form multi-scale fusion features to lay the foundation for accurate prediction.

[0118] A composite loss function containing two losses is constructed, and the model training is completed by minimizing the function:

[0119] The first loss term (mean square error) calculates the mean square error between the predicted value of the environmental parameter output by the model and the actual monitoring value, measures the overall deviation between the predicted value and the true value, and ensures the basic accuracy of the prediction.

[0120] This loss term reduces the deviation in the prediction of the coupling relationship between parameters by constraining the consistency of the predicted parameter change and the theoretical coupling relationship. The final loss is obtained by weighted sum of the two losses, and the model parameters are iteratively optimized using algorithms such as gradient descent until the loss function value converges to a preset threshold, completing the model training.

[0121] Adaptive update of coupling coefficient matrix C: using the process of basic initialization, data-driven correction, and weighted fusion to dynamically update

[0122] Based on the greenhouse structure (such as greenhouse height, number of vents), the physical properties of the covering material (such as the light transmittance and thermal insulation of glass), and the basic physiological model of crops (such as the influence of crop transpiration on humidity), the basic coupling coefficient matrix is determined.

[0123] Continuous collection of real-time device operation sequences and environmental parameter response sequences, with the device operation and environmental state at the previous time as input and the environmental parameter change at the next time as output. Using multivariate linear regression with regularization or Bayesian inference method, the data-driven coupling coefficient matrix is calculated.

[0124] According to the statistical confidence of the data-driven coupling coefficient matrix (such as whether the data volume is sufficient and whether the data fluctuation is within a reasonable range), the weight is dynamically allocated to ensure that the coupling coefficient fits the actual scene.

[0125] The effect of the above technical solution is:

[0126] Through nested LSTM and attention mechanism, the model can capture the instantaneous fluctuation, short-term trend and long-term regularity of environmental parameters at the same time, avoiding the prediction bias caused by single-scale analysis. For example, a sudden short-term strong wind (instantaneous fluctuation) occurs in summer at noon. The bottom LSTM can quickly capture the impact of this change on the greenhouse ventilation efficiency, the middle LSTM combines the daytime temperature rising trend, and the top attention mechanism focuses on the weight of short-term fluctuations, finally accurately predicting the short-term decrease of greenhouse temperature, avoiding the prediction deviation caused by ignoring instantaneous changes.

[0127] The coupling loss term restricts the consistency of the environmental parameter changes and the theoretical coupling relationship, and reduces the error of the correlation prediction between parameters. For example, there is a coupling relationship between temperature and humidity that temperature rises and humidity decreases. If the model predicts that the temperature rises but the humidity does not decrease, the coupling loss term will increase, prompting the model to correct the prediction result, ensuring that the predicted changes of temperature and humidity conform to the actual coupling law, and improving the reliability of multi-parameter collaborative prediction. The matrix is updated dynamically by fusing the basic model and real-time data, which can adapt to different greenhouse structures, crop types and seasonal change scenes, so that the model can adapt to new scenes without retraining, reducing maintenance costs.

[0128] The model outputs the predicted value and confidence interval of the future environmental parameters, and the precise optimization of the compound loss function can quantify the prediction uncertainty and ensure the prediction accuracy. For example, when predicting the temperature in the greenhouse in the next 2 hours, not only is the predicted value of 22°C given, but also the confidence interval of 21-23°C is output. At the same time, through the dual control of mean square error and coupling loss, it is ensured that the predicted value and interval are highly consistent with the actual situation, providing reliable environmental prediction basis for the subsequent model prediction control algorithm to generate instruction set, and reducing the control failure caused by prediction error.

[0129] In one possible implementation, after collecting the environmental parameters, the method further includes a data fusion step, comprising:

[0130] For multiple sensors monitoring the same environmental parameter, a dynamic trust evaluation mechanism is used to assign a second weight to each sensor reading;

[0131] The second weight is used to weight and average the readings of the multiple sensors to obtain a comprehensive estimate of the environmental parameter;

[0132] The comprehensive estimate values of all environmental parameters are combined into a comprehensive environmental state vector;

[0133] Physical rules based on energy conservation and / or transpiration models are used to verify the comprehensive environmental state vector; when the verification result shows that the state vector violates the physical law, a substitute data strategy is used to obtain a reliable environmental state estimate;

[0134] The substitute data strategy includes using the reliable environmental state estimate value at the previous time or the estimate value based on the prediction model to replace the comprehensive environmental state vector at the current time which is judged to be unreliable.

[0135] The second weight is obtained by the following method:

[0136] The comprehensive trust score of the sensor is obtained; the comprehensive trust score is determined by the instantaneous abnormal score, the time sequence consistency score, the spatial consistency score, and the external goodness of fit score; wherein,

[0137] Instantaneous abnormality score: to evaluate the statistical abnormality degree of sensor reading within its short-term time window, including:

[0138] The instantaneous abnormality score is calculated by the following dynamic rules:

[0139] Obtain the current sensor reading and its historical data sequence within a preset time window (e.g., the past 5 minutes).

[0140] Calculate the robust mean μ_robust and robust standard deviation σ_robust of the historical data within the time window;

[0141] The robust mean (μ_robust) is the weighted average of the historical data sequence, and the weight of each data point is obtained by processing its historical abnormality score with a decay function; the lower the historical abnormality score of a data point, the smaller its weight;

[0142] The robust standard deviation (σ_robust) is calculated based on the weighted standard deviation formula with the same weight;

[0143] Calculate the standardized deviation M_t of the current reading (x_t) relative to the robust statistics;

[0144] M_t = |x_t - μ_robust| / σ_robust

[0145] Map the standardized deviation (M_t) to the instantaneous abnormality score S_instant between 0 and 1 through a preset monotonically decreasing function F: S_instant = F(M_t);

[0146] Wherein, the monotonically decreasing function F satisfies: when the standardized deviation does not exceed the preset threshold k based on the statistical confidence interval, the instantaneous abnormality score tends to 1; when the standardized deviation exceeds the preset threshold k, the instantaneous abnormality score tends to 0 as M_t increases; wherein, the value range of the preset threshold k is 2 to 3, and the specific value is adjusted according to the sensor type and historical measurement accuracy.

[0147] Temporal consistency score: to evaluate the smoothness and trend rationality of the sequence composed of the reading and the previous data; the acquisition method of the temporal consistency score includes: based on the predicted value of the sensor parameter sequence by recurrent neural network (RNN), calculate the deviation of the current reading and the predicted value; the smaller the deviation, the higher the temporal consistency score.

[0148] Spatial consistency score: to evaluate the difference between the reading and other sensor readings of the same parameter in spatial proximity; calculate the difference between the current sensor reading and the average value of the same parameter readings of the neighboring sensor cluster weighted by its spatial proximity and greenhouse airflow direction; the smaller the difference, the higher the spatial consistency score.

[0149] External fit score: assess the degree of fit of the reading to the expected value range derived from external meteorological data by physical model; input external meteorological data into a greenhouse physical model based on the law of conservation of energy to derive the expected value range of the environmental parameter in the greenhouse; calculate the degree of fit of the current sensor reading to the expected value range; the higher the degree of fit, the higher the external fit score.

[0150] The comprehensive trust score is obtained by weighted summation of the instantaneous abnormality score, the time consistency score, the spatial consistency score and the external fit score;

[0151] Wherein, the weight of each sub-score is dynamically configured according to the diagnostic accuracy rate of each scoring dimension to data anomaly in historical data.

[0152] According to the comprehensive trust scores of all sensors monitoring the same environmental parameter in the greenhouse, the second weight of each sensor of the environmental parameter is determined.

[0153] The comprehensive trust scores of all sensors monitoring the same environmental parameter are normalized, and the result is the second weight of each sensor reading.

[0154] The dynamic trust evaluation mechanism comprehensively judges the reliability of the sensor from the instantaneous, time, spatial and external four dimensions, avoids the limitations of traditional equal weight and single dimension evaluation, effectively filters abnormal data interference, and improves the single parameter estimation accuracy.

[0155] By calculating the standardized deviation by robust mean and robust standard deviation, the influence of abnormal values in historical data on the statistical benchmark can be effectively excluded (for example, a jump reading of a sensor occurs accidentally), and then the deviation is mapped to a score by a monotonically decreasing function, which can quickly identify short-term abnormalities of the current reading. For example, a humidity sensor in the greenhouse shows 95% humidity (far beyond the normal range) for a moment due to water droplets splashing, the algorithm calculates its standardized deviation M_t=4 (greater than the threshold k=3), the instantaneous abnormality score is reduced to 0.1, and the reading is marked as unreliable in time, avoiding its pulling up the humidity comprehensive estimation value, and ensuring the short-term accuracy of single parameter data.

[0156] Based on the RNN model to predict the trend of the previous data, by the deviation score of the current reading and the predicted value, data that is short-term normal but long-term trend inconsistent can be identified.

[0157] The average value of the adjacent sensor cluster is calculated by weighting according to spatial proximity and air flow direction, and by the difference score of the current reading and the average value, the case that only a single sensor is abnormal and the surrounding sensors are normal can be identified.

[0158] When scoring, the average of the neighboring cluster is calculated with a weight that considers the direction of the greenhouse air flow (for example, a sensor near a vent is more affected by the outdoor air flow). Instead of simply taking the arithmetic mean, the average of the neighboring cluster of a humidity sensor near the west vent of the greenhouse will be more referenced to the other sensors on the west side (affected by the same ventilation air flow) than to the east side sensors (less affected by the air flow), avoiding misjudgments of spatial correlation due to air flow differences, making the score more consistent with the actual greenhouse space environment and improving the accuracy of spatial consistency evaluation.

[0159] External meteorological data (such as outdoor temperature, wind speed) are converted into expected ranges of greenhouse internal parameters through energy conservation physical models. The degree of fit between the current readings and the ranges is scored to identify data that is logically inconsistent with the external environment. The physical model incorporates the characteristics of the greenhouse itself (such as the good heat retention of a glass greenhouse, so outdoor temperature changes have little effect on the indoor environment; the poor heat retention of a plastic greenhouse, so the effect is large) to derive the expected range.

[0160] In addition to evaluating the reliability of the data itself, the inherent physical laws of the greenhouse environment system can also be used as a verifier to filter out data combinations that are logically impossible. This physical constraint-based data fusion is key to ensuring the self-consistency of the system's cognition.

[0161] The environmental factors in the greenhouse do not exist in isolation, but are a complex system that interacts and influences each other. For example, an increase in air temperature is usually accompanied by a decrease in air humidity (especially when ventilation is used), while an increase in light intensity will accelerate the transpiration of crops, leading to a decrease in air humidity and an increase in temperature. Based on these principles, physical constraint verification is performed when fusing data.

[0162] By analyzing the interaction between temperature, humidity, and light intensity, it is checked whether there is a violation of the law of conservation of energy. For example, if multiple sensors simultaneously report a sharp rise in air temperature without external heat input, and humidity also abnormally rises, this may violate the law of conservation of energy, and such data combinations are considered abnormal.

[0163] When the light intensity reaches or approaches the light saturation point of tomatoes (about 70,000 lux), the transpiration of tomatoes will tend to saturate. At this time, if the system detects extremely high light intensity, but the air humidity is abnormally stable or even rising, it may indicate that there is a serious problem with the sensor data, because strong light should promote the evaporation of a large amount of water.

[0164] Data verification is combined with known crop physiological models. For example, for tomatoes, there is a complex nonlinear relationship between temperature and light intensity, which can be modeled by a quadratic function. If the light intensity in a certain area is extremely low, but the temperature is abnormally high, it may indicate a failure of the shading system, but it also violates the physiological logic that crops do not grow well in low light environments.

[0165] By introducing physical constraint check, those data combinations that seem normal but are logically contradictory are identified, so as to actively eliminate or correct these information, and ensure that the decision greenhouse environment data is real and in line with the physical law.

[0166] In a possible implementation, in the conflict resolution mechanism based on dynamic priority, the priority rules include: the priority of the instruction for ensuring the survival of crops is the highest, the priority of the instruction for adapting the growth of crops is second, and the priority of the instruction for optimizing resource consumption is the lowest.

[0167] In a possible implementation, the environmental parameters include temperature, humidity, light and CO2 concentration, and in the priority rules, the priority of temperature regulation is higher than that of humidity regulation, the priority of humidity regulation is higher than that of light and CO2 regulation; and the priority weight of each environmental parameter is dynamically adjusted according to the current crop growth stage and real-time environmental state.

[0168] Among the many environmental factors, temperature has the most direct and intense effect on the growth rate and physiological activity of crops, and therefore is established as the highest priority; the scientific basis for establishing temperature as the highest priority is that temperature is the dominant factor affecting physiological processes such as photosynthesis and respiration. For example, when the temperature is too high, even if the light and CO2 concentration are in the best state, the crop may stop growing due to heat stress. Conversely, when the temperature is too low, even if the humidity and light conditions are good, the metabolic activity of the crop will be greatly reduced. By establishing the asymmetric relationship between temperature and other environmental factors (such as humidity, CO2) in advance, the system can handle the strong coupling problem between them, for example, ventilation for cooling may cause the air humidity to decrease, at this time the system will prioritize to ensure that the temperature is within the safe range. After the temperature is effectively controlled, humidity, light intensity and carbon dioxide concentration will be processed in turn. This clear response sequence ensures that in the case of limited resources, the problem that has the greatest impact on the survival and growth of crops is solved first.

[0169] Although temperature has the highest priority, it is not always the same in all cases; the physiological needs of crops vary at different growth stages, and the importance of certain key parameters will change accordingly. Therefore, a dynamic priority weight adjustment mechanism is introduced, so that the control decision can be more flexible and accurate.

[0170] Dynamic adjustment based on crop growth stage: For example, during the growth of tomatoes, different stages have different sensitivities to the environment. During the seedling stage, a larger diurnal temperature difference (such as 8-10℃) is beneficial to the seedling; during the flowering stage, the stability of temperature is more important than intensity, because temperature exceeding 30℃ for 3 hours will cause pollen inactivation rate to exceed 60%, seriously affecting fruit setting rate. Therefore, when the crop is identified to enter the flowering stage, the intensity weight of temperature control will be automatically reduced, while the weight of light intensity and temperature stability will be increased to ensure the stability of pollination environment. During the fruit enlargement and color change stage, the control of diurnal temperature difference will be emphasized to promote sugar accumulation and lycopene synthesis, and the optimization weight of light intensity may be increased again.

[0171] Dynamic adjustment based on current environmental state: Real-time evaluation of environmental state, and adjustment of control weights of various parameters accordingly. For example, in the scenario of high temperature and high humidity combined stress, although the priority of temperature is higher, it is also realized that humidity is also a key factor leading to disease (such as 40% increase in tomato disease spot incidence) and growth restriction. At this time, the weight of humidity control is dynamically increased, and cooling and dehumidification are considered as equally important targets, and the micro-spraying system and ventilation equipment are coordinated to work together to achieve efficient cooling and dehumidification. Similarly, in the condition of low temperature and weak light, the light supplementing system is preferentially started, and then the heating equipment is coordinated to achieve the dual targets of temperature increase and light increase.

[0172] Embodiment 2 provides a greenhouse environment adaptive control system for implementing the greenhouse environment adaptive control method of the right embodiment 1, the system comprises:

[0173] A data acquisition module for acquiring multi-dimensional environmental parameters in the greenhouse through a distributed sensor network;

[0174] An instruction generation module for the central control unit to generate a set of device linkage instructions for a future preset time period based on the environmental parameters, a preset crop growth period database and weather prediction data, with the goal of minimizing a preset cost function, wherein the preset cost function integrates environmental control deviation and operating cost, and a model predictive control algorithm is used;

[0175] An execution module for issuing the set of device linkage instructions to each execution device for linkage control;

[0176] Wherein, when generating the set of device linkage instructions, a dynamic priority-based conflict resolution mechanism is used to determine the execution order of multiple device instructions.

[0177] The working principle and effect of the above technical solution are the same as those of the method embodiments of the present application, and will not be repeated here.

[0178] The embodiment of the present application further provides an electronic device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of the method according to any one of the embodiments of the present application when executing the computer program.

[0179] The embodiment of the present application further provides a computer readable storage medium for storing a computer program, the computer program is executed to implement the steps of the method according to any one of the embodiments of the present application, and the specific implementation manners and the achieved technical effects are the same as those of the method embodiments, and part of the contents will not be described herein.

[0180] In the present application, the readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, device or apparatus. The program product can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example but not limited to, an electrical, magnetic, optical, electromagnetic, infrared or semiconductor system, device or apparatus, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0181] Computer-readable storage media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The readable storage medium may also be any readable medium capable of sending, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, or any suitable combination thereof. Program code for performing operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar programming languages. The program code may be executed entirely on a user computing device, partially on an associated device, as a standalone software package, partially on a user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing devices can be connected to user computing devices via any type of network, including local area networks (LANs) or wide area networks (WANs), or they can be connected to external computing devices (e.g., via the Internet using an Internet service provider).

[0182] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the invention without departing from the principles and spirit of the invention, and all such changes should fall within the protection scope of the claims of the present invention.

Claims

1. A method for adaptive regulation of a greenhouse environment, characterized in that, The method comprises: Collecting multi-dimensional environmental parameters in the greenhouse; Based on the environmental parameters, a preset crop growth period database and weather prediction data, a model predictive control algorithm is used to generate a device linkage instruction set for a future preset time period, with the goal of minimizing a preset cost function that combines environmental regulation deviation and operating cost; The device linkage instruction set is issued to each execution device for linkage regulation and control; Wherein, in generating the device linkage instruction set, a dynamic priority-based conflict resolution mechanism is used to determine the execution order of multiple device instructions.

2. The method of adaptive regulation of the greenhouse environment according to claim 1, characterized in that, In the process of generating the device linkage instruction set, the model predictive control algorithm synchronously generates an environmental parameter reference trajectory for a future preset time period; this trajectory is the environmental target expected to be achieved by the device linkage instruction set, and the generation of the environmental parameter reference trajectory is based on the premise of meeting the physiological needs of crops, with the core goal of minimizing the deviation of the comprehensive physiological suitability index in the preset cost function.

3. The method of adaptive regulation of the greenhouse environment according to claim 2, characterized in that, The comprehensive physiological suitability index is obtained by the following method; including: Through the crop growth period database, the ideal range and stress threshold of each environmental parameter at the current crop growth stage are obtained; based on the ideal range and stress threshold, a single-factor suitability function is constructed for each environmental parameter, which maps the measured or predicted value of the environmental parameter to a suitability score between 0 and 1; According to the current crop growth stage, a dynamic first weight is assigned to the suitability score of each environmental parameter; The suitability scores of each environmental parameter are weighted and summed with their corresponding dynamic weights to obtain the comprehensive physiological suitability index.

4. The method of adaptive regulation of the greenhouse environment according to claim 2, characterized in that, The specific steps of the model predictive control algorithm include: Receive weather forecast data for a future preset time period as a feedforward input; Establish a dynamic prediction model of the greenhouse environment to simulate the changes in the environmental parameters inside the greenhouse after different combinations of execution device operations; Construct a multi-objective cost function, which includes a first term representing physiological benefit deviation and a second term representing economic cost; wherein the first term is the minimization of the deviation between the cumulative comprehensive physiological suitability index and the ideal value within the future preset time period, and the second term is the minimization of the total energy consumption of the execution device; Take the device operation sequence to be solved within the future preset time period as the decision variable, and based on the dynamic prediction model, solve the multi-objective cost function under the condition of meeting the physiological constraints of crops to obtain the optimal device operation sequence that minimizes the function value, which is the device linkage instruction set.

5. The method of adaptive regulation of the greenhouse environment according to claim 4, characterized in that, The dynamic prediction model of the greenhouse environment is constructed and trained by the following method, including: Using a multivariate time series composed of historical device operation sequences, historical environmental parameter sequences and historical external weather data as input, output includes predicted values of future environmental parameters and their confidence intervals, and the model is trained by minimizing the loss function; The dynamic prediction model uses a multi-scale time series fusion architecture to extract multi-scale features of environmental parameter instantaneous fluctuations, short-term trends and long-term regularities, and dynamically fuses the multi-scale features through an attention mechanism; The loss function comprises a first loss term and a second loss term, the first loss term is a mean square error of the predicted value and the measured value, and the second loss term is a coupling loss term; The second loss term is calculated by the following formula: where L coupling is a coupling loss term; is the model predicted change in environmental parameter matrix; C is a predefined coupling coefficient matrix; ||| | 2 is the L2 norm.

6. The method of adaptive regulation of the greenhouse environment according to claim 1, characterized by, After collecting the environmental parameters, the method further comprises a data fusion step, comprising: For multiple sensors monitoring the same environmental parameter, a dynamic trust evaluation mechanism is used to assign a second weight to each sensor reading; The second weight is used to perform weighted averaging on the readings of the multiple sensors to obtain a comprehensive estimated value of the environmental parameter; The comprehensive estimated values of all environmental parameters are combined into a comprehensive environmental state vector; Physical rules based on energy conservation and / or transpiration models are used to verify the comprehensive environmental state vector.

7. The method of adaptive regulation of the greenhouse environment according to claim 1, characterized in that, The second weight is obtained by the following method: Obtain the comprehensive trust score of the sensor; the comprehensive trust score is determined by the instantaneous abnormal score, the time sequence consistency score, the spatial consistency score, and the external fitting degree score; According to the comprehensive trust scores of all sensors monitoring the same environmental parameter in the greenhouse, the second weight of each sensor for the environmental parameter is determined.

8. The method of adaptive regulation of the greenhouse environment according to claim 1, characterized by, In the conflict resolution mechanism based on dynamic priority, the priority rules include: the priority of the instruction for ensuring the survival of crops is the highest, followed by the instruction for adapting to the growth of crops, and the priority of the instruction for optimizing resource consumption is the lowest.

9. The method of adaptive regulation of the greenhouse environment according to claim 7, characterized in that, The environmental parameters include temperature, humidity, light, and CO2 concentration, and in the priority rules, the priority of temperature regulation is higher than that of humidity regulation, the priority of humidity regulation is higher than that of light and CO2 regulation; and the priority weights of each environmental parameter are dynamically adjusted according to the current crop growth stage and the real-time environmental state.

10. A greenhouse environment self-adaptive regulation system for implementing the greenhouse environment self-adaptive regulation method according to any one of claims 1-9, characterized in that, The system comprises: A data acquisition module acquires multi-dimensional environmental parameters in the greenhouse; An instruction generation module generates a set of device linkage instructions for a future preset time period based on the environmental parameters, a preset crop growth period database, and weather prediction data, using a model predictive control algorithm with the objective of minimizing a preset cost function, wherein the preset cost function integrates environmental regulation deviation and operating cost; An execution module is used to issue the set of device linkage instructions to each execution device for linkage control; Wherein, when generating the set of device linkage instructions, a conflict resolution mechanism based on dynamic priority is used to determine the execution order of multiple device instructions.

Images