Tobacco growth environment multi-parameter intelligent regulation and control method driven by Internet of Things

Abstract

The invention discloses a tobacco growth environment multi-parameter intelligent regulation and control method driven by the Internet of Things, and relates to the technical field of agricultural Internet of Things and environment intelligent regulation and control, and the method comprises the steps: based on the environment parameters of a tobacco planting area collected in real time and crop stem flow data, combining the current growth stage and quality target of tobacco, and obtaining a tobacco growth environment multi-parameter intelligent regulation and control result; dynamically generating a target concentration change track of the key metabolite representing the internal physiological activity of the tobacco; taking the target concentration change track as an optimization target, carrying out reverse solution by using a crop physiological mechanism model for describing the relationship between the environment and the physiological process, and calculating a target track of environment parameter set values which are required for realizing the optimization target and change along with time; according to the target trajectory, combining with a pre-established equipment disturbance propagation model, and solving a field domain control inversion problem to generate a coordination action sequence of the environment regulation and control equipment; and the cooperative action sequence of the equipment is executed to control the corresponding environment regulation and control equipment to work.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of agricultural Internet of Things and environmental intelligent regulation, more particularly to the field of precise regulation of growth environment in tobacco agriculture, and specifically to an Internet of Things driven multi-parameter intelligent regulation method for tobacco growth environment. BACKGROUND

[0002] Tobacco is an important economic crop in China, and its quality and yield are directly related to the economic benefits and market competitiveness of the tobacco industry. The growth environment condition is a key factor affecting the growth and development of tobacco and the formation of intrinsic quality. Tobacco has significant differences in the demand for environmental parameters such as temperature, humidity, light, CO2 concentration, and soil moisture at different growth stages. If the environmental conditions deviate from the appropriate range, not only will the growth rate of tobacco be delayed and the risk of disease and insect pests increase, but also the imbalance of chemical components in tobacco leaves may occur, reducing the utilization rate of raw materials. With the advancement of modern agricultural intelligence, the use of Internet of Things technology to realize precise regulation of the growth environment of tobacco has become an important way to break through the limitations of traditional manual management, ensure the stability of tobacco quality, and improve production efficiency, which has important practical significance for promoting the development of tobacco agriculture towards high efficiency, high quality, and sustainability.

[0003] Although existing tobacco growth environment regulation technology has gradually integrated intelligent monitoring means, there are still obvious deficiencies. Most regulation schemes rely on preset environmental parameter thresholds or simple empirical models for control, and do not fully consider the dynamic changes in the growth state of tobacco, making it difficult to achieve accurate regulation.

[0004] In view of the above problems, no effective solutions have been proposed so far. SUMMARY

[0005] The embodiments of the present application provide an Internet of Things driven multi-parameter intelligent regulation method for tobacco growth environment to solve the above technical problems.

[0006] The present application provides an Internet of Things driven multi-parameter intelligent regulation method for tobacco growth environment, comprising: Based on the real-time collected environmental parameters and crop stem flow data of the tobacco planting area, combined with the current growth stage and quality target of tobacco, the target concentration change trajectory of the key metabolites representing the internal physiological activity of tobacco is dynamically generated; taking the target concentration change trajectory as the optimization target, the crop physiological mechanism model describing the relationship between the environment and the physiological process is used for reverse solving to calculate the target trajectory of the time-varying environmental parameter setting value required to achieve the optimization target; according to the target trajectory, combined with the pre-established device disturbance propagation model, the coordinated action sequence of the environmental regulation device is generated by solving a field control inversion problem; wherein the device disturbance propagation model defines the time and space disturbance characteristics of each device action on the environmental state at different positions in the planting space; the coordinated action sequence of the device is executed to control the corresponding environmental regulation device to work; the regulation effect is evaluated according to the time sequence change effect of the environmental parameters and the crop stem flow data, and the key physiological parameters in the crop physiological mechanism model and the spatial transmission characteristics in the device disturbance propagation model are dynamically calibrated based on the evaluation results.

[0007] Based on the embodiments provided in the present application, by dynamically generating the target concentration change trajectory of the key metabolites based on the real-time environmental parameters, crop stem flow data, and the current growth stage and quality target of tobacco, the regulation target can be directly anchored to the core demand of the internal physiological activity of tobacco, avoiding the disconnection between the regulation direction and the inherent growth mechanism of tobacco, and making the environmental regulation more suitable for the physiological nature of tobacco in forming the target quality at different growth stages. Taking the target concentration change trajectory as the optimization target, the crop physiological mechanism model is used for reverse solving to obtain the target trajectory, which can adaptively adjust the regulation target of the environmental parameters according to the dynamic changes of the tobacco metabolic activity, breaking through the limitations of relying on fixed thresholds or empirical models, and ensuring that the environmental parameters are always accurately matched with the real-time physiological demand of tobacco.

[0008] Combined with the device disturbance propagation model and by solving the field control inversion problem to generate the coordinated action sequence of the device, the spatial and temporal disturbance differences of each device action on the environmental state at different positions in the planting space are fully considered, which can effectively reduce the mutual interference of multiple devices working, improve the spatial uniformity of the environmental state of the planting area, and avoid the influence of local environmental fluctuations on the consistency of tobacco growth. By evaluating the regulation effect according to the time sequence change of the environmental parameters and the stem flow data, and dynamically calibrating the key parameters of the crop physiological mechanism model and the spatial transmission characteristics of the device disturbance propagation model, a closed-loop optimization mechanism is formed, which can adapt the regulation model to the changes of the tobacco growth state and the device working characteristics, ensuring the stability and accuracy of the long-term regulation effect, and helping to achieve the reliable achievement of the quality target of tobacco. BRIEF DESCRIPTION OF DRAWINGS

[0009] The accompanying drawings, which are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings: Figure 1 is a flow chart of an optional tobacco growth environment multi-parameter intelligent regulation and control method driven by an Internet of Things according to an embodiment of the application. Figure 2 is a flow chart of another optional tobacco growth environment multi-parameter intelligent regulation and control method driven by an Internet of Things according to an embodiment of the application. Figure 3 is a structural schematic diagram of an optional electronic device according to an embodiment of the application.

[0010] The implementation, functional features and advantages of the application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0011] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the application.

[0012] According to an aspect of an embodiment of the application, as shown in Figure 1 the application provides a tobacco growth environment multi-parameter intelligent regulation and control method driven by an Internet of Things, which comprises: S101, based on the real-time collected environment parameters of a tobacco planting area and crop stem flow data, and in combination with the current growth stage of tobacco and quality targets, a target concentration change trajectory of a key metabolite representing internal physiological activities of tobacco is dynamically generated; In S101, the control target is moved from the external environment parameters to the physiological activities of the crop itself. By fusing real-time environment data and the key physiological signal of stem flow, and in combination with the growth stage and quality targets, a target concentration change trajectory representing internal physiological activities is dynamically deduced. This is equivalent to establishing a physiological target for the whole regulation and control system, which is core to the internal state of the crop, quantitative and dynamically changing, rather than a fixed environment set point. The significance lies in changing the regulation and control behavior from “maintaining the environment” to “realizing the ideal physiological state of the crop”.

[0013] It needs to be explained that the "dynamic" of step S101 is reflected in two aspects, one is the real-time trigger of data-driven, and the generation process is driven by real-time collected environmental parameters and crop sap flow data. Data is obtained every 5-10 minutes (a monitoring period), when the data indicates that the physiological state of the crop changes or the external environment changes, the target concentration change trajectory will be triggered to recalculate and update. The second is the online refresh of the model parameters, the key parameters of the core model used to generate the trajectory are not fixed, but are dynamically called and fine-tuned in the preset parameter library according to the current growth stage of the crop and the historical environmental data, so that the generated trajectory can adapt to the dynamic changes of the physiological needs of the crop.

[0014] The core is to take the quality target as the final orientation and to take the real-time data as the calibration basis. First, a basic trajectory is generated based on the photosynthesis model and metabolic allocation strategy, and then the basic trajectory is corrected in real time through the stress detection and response mechanism, and finally the target concentration change trajectory that meets the real needs of the crop is output.

[0015] S102, taking the target concentration change trajectory as the optimization target, using the crop physiological mechanism model describing the relationship between the environment and the physiological process to perform reverse solving, calculating the target trajectory of the environmental parameter setting value required to achieve the optimization target changing with time; According to the real-time collected environmental parameters and sap flow data as the model input and feedback, the crop physiological mechanism model is used for reverse solving, and the target trajectory is continuously changed with time.

[0016] In the present application, the crop physiological mechanism model is an integrated concept, which does not refer to a single, fixed set of mathematical equations, but refers to a set of calculation framework for describing the key physiological processes of tobacco (such as photosynthesis, material transport, water transport, etc.) and their interaction with environmental factors. The specific implementation of the model can be embodied as the cooperative work of dynamic physiological causal diagram, simplified metabolic flow model for rapid calculation, and stem flow prediction model components.

[0017] In S102, a reverse thinking and mechanism model driven solving method is adopted. With the generated internal physiological target (target concentration change trajectory) as the terminal point, the crop physiological mechanism model is used for reverse deduction to calculate what kind of external environment (target trajectory) is needed to drive the crop to achieve the internal target. This is equivalent to deducing from "what the crop wants" to "what the environment should provide" in virtual space. The effect is that the generated target trajectory has a solid physiological basis, ensuring that any environmental intervention directly serves a specific physiological purpose, greatly improving the pertinence and scientificity of the control. S103, according to the target trajectory, combining the pre-established device disturbance propagation model, by solving a field control inversion problem to generate the coordinated action sequence of the environment regulation device; wherein the device disturbance propagation model defines the spatio-temporal disturbance characteristics of each device action on the environment state at different positions in the planting space; Wherein, the field control inversion problem of the present application is mathematically expressed as a constrained spatio-temporal optimization problem. The objective function is to minimize the root mean square error between the actual environment field and the expected environment field in the main area of the crop canopy; the constraint conditions include the physical constraints of the device itself, the energy consumption economy constraints and the logical mutual exclusion constraints between different device actions.

[0018] The influence of device action on the planting space is not uniform, but has spatio-temporal disturbance characteristics. In S103, by comparing the target trajectory with the current state, a "spatial expected field increment" is formed, and the optimal coordinated action sequence is solved in the pre-defined device action primitive library with this as the target. It realizes the mapping from one-dimensional environment target to three-dimensional space regulation, and avoids mutual interference between devices through the coordinated action sequence, ensuring that the environment in the whole space can accurately and efficiently approach the target trajectory.

[0019] S104, executing the coordinated action sequence of the device to control the corresponding environment regulation device to work; S105, evaluating the regulation effect according to the time sequence change effect of the environment parameters and the crop sap flow data, and dynamically calibrating the key physiological parameters in the crop physiological mechanism model and the spatial transmission characteristics in the device disturbance propagation model based on the evaluation results.

[0020] In S105, a self-learning and adaptive feedback optimization mechanism is constructed. It is not executed once, but continuously monitors the time sequence changes of environment parameters and crop sap flow, evaluates the regulation effect from two dimensions of "environment tracking accuracy" and "physiological response effect". Based on the evaluation results, the key parameters in the crop physiological mechanism model and the device disturbance propagation model are dynamically calibrated. Its core effect is to enable the system to overcome the uncertainty brought by the initial error of the model, the change of the device performance and the dynamic growth of the crop, so that the model becomes more and more close to the actual situation over time, thereby ensuring the stability and reliability of the long-term regulation effect.

[0021] Further, based on the real-time collected environment parameters and crop sap flow data of the tobacco planting area, combining the current growth stage and quality target of tobacco, the target concentration change trajectory of the key metabolites representing the internal physiological activity of tobacco is dynamically generated, including: The light intensity and temperature data in the real-time collected environment parameters are input into the preset photosynthesis primary productivity model to output the time-varying basic synthesis rate curve of carbohydrates; As a specific example, the present application can employ a nonrectangular hyperbolic light response model as the core of the photosynthesis primary productivity model. The model can better fit the light saturation phenomenon, and its mathematical expression contains photosynthetically active radiation flux density, maximum total photosynthetic rate, apparent quantum efficiency, convexity factor, and dark respiration rate, etc. parameters, all of which are known technologies, and this embodiment will not be described here.

[0022] By inputting the real-time collected light intensity and temperature into the model, the basic synthesis rate curve of carbohydrates can be output. The values of the model parameters are determined based on the published physiological research literature of different tobacco varieties.

[0023] The current growth stage of tobacco is taken as an index to query the preset metabolic allocation strategy database to obtain allocation parameters; and the allocation parameters are used to decompose the basic synthesis rate curve into a first synthesis rate sub-trajectory for growth and a second synthesis rate sub-trajectory for accumulation and transport; Among them, the metabolic allocation strategy database is realized in the form of a two-dimensional lookup table, the row index of which is the growth stage, and the column index is the allocation target. The data in the table is derived from long-term field experiments and literature integration.

[0024] In order to clearly define the allocation process, the present application introduces an allocation coefficient function with growth stage as a variable The specific formula for decomposing the basic synthesis rate curve with the allocation parameters is as follows:

[0025]

[0026] Among them, is the first synthesis rate sub-trajectory for growth, is the second synthesis rate sub-trajectory for accumulation and transport, is the basic synthesis rate curve that changes with time, is the growth allocation coefficient function related to the growth stage. Based on this formula, the abstract "metabolic allocation strategy" is converted into an accurate mathematical model, and the flow direction of photosynthetic products is guided through the dynamically changing coefficient, which is the core calculation step to realize "on-demand regulation".

[0027] The nicotine content set in the quality target is combined with the number of days from the current date to the planned harvest date to calculate the daily average demand rate of nicotine precursor substances, so as to generate a transport rate curve with a benchmark level; The purpose of this step is to decompose the final quality target (nicotine content) into the daily physiological tasks of the crop. The implementation process is as follows: Input: the target nicotine content per tobacco plant or per unit area (e.g., 2.5% of dry weight) set in the quality target, and the total number of days from the current date to the scheduled harvest date (e.g., 60 days).

[0028] Calculation: First, estimate the overall conversion efficiency from nicotine precursors (e.g., nicotine acid) to nicotine accumulation based on crop variety and historical data (e.g., estimated as 40%). Then, calculate the total amount of nicotine precursors needed to be synthesized during the remaining growth period. Conceptually, the formula is: total precursor demand = target nicotine content / overall conversion efficiency. Finally, distribute the total demand evenly to each day to obtain the daily demand rate. That is: daily demand rate = total precursor demand / remaining days.

[0029] Output: Based on the daily demand rate, the system generates a baseline translocation rate curve that is flat or has small fluctuations. This curve represents the ideal situation where the crop needs to steadily synthesize and translocate a certain amount of precursors every day to achieve the target quality at harvest.

[0030] Determine the initial metabolic flux trajectory set based on the baseline synthesis rate curve, the first synthesis rate sub-trajectory, the second synthesis rate sub-trajectory, and the translocation rate curve. Compare the real-time collected stem flow data with the pre-set normal range; when the data is abnormal, obtain the attenuation coefficient according to the pre-set stress response coefficient table to make global adjustment to all curves in the initial metabolic flux trajectory set, and output the target concentration change trajectory.

[0031] Wherein, the stress response coefficient table is a key empirical database. For example, when the system detects that the stem flow rate is continuously 10%-20% lower than the normal range, it is determined to be mild water stress, and the attenuation coefficient is taken as 0.9, the physiological basis of which is that partial stomatal closure leads to limited supply of photosynthetic substrates, requiring moderate down-regulation of synthesis targets; when the stem flow rate is continuously more than 20% lower than the normal range, it is determined to be severe water stress, and the attenuation coefficient is taken as 0.7, the basis of which is that significant stomatal closure leads to severe suppression of photosynthesis. When the system determines to enter a certain stress state, it reads the corresponding attenuation coefficient from the table and makes global adjustment to all curves in the initial metabolic flux trajectory set.

[0032] It should be noted that, taking the stem flow rate of mature tobacco as an example, its pre-set normal range can be pre-set as 800 grams to 1500 grams per hour. This range is a statistical range obtained by statistically analyzing the stem flow data of tobacco plants in a healthy state with sufficient water supply on a typical sunny day in many years of planting practice (e.g., taking the interval of 5% to 95% quantile).

[0033] The process of determining the initial set of metabolic flow trajectories is not to generate a single trajectory, but to construct a set of initial schemes with multiple focuses.

[0034] Base trajectory: directly use the base synthesis rate curve calculated by the photosynthesis model and allocation strategy.

[0035] Growth priority trajectory: on the basis of the base trajectory, the amplitude of the "first synthesis rate sub-trajectory" for growth is strengthened, forming a scheme that is biased towards promoting plant growth.

[0036] Quality priority trajectory: on the basis of the base trajectory, the amplitude of the "second synthesis rate sub-trajectory" for accumulation and transport and the "transport rate curve" is strengthened, forming a scheme that is biased towards secondary metabolite accumulation.

[0037] This set provides a variety of feasible initial choices for subsequent inverse solving.

[0038] Based on the embodiments provided in the present application, external environmental monitoring (such as light and temperature) and internal physiological activity (such as carbohydrate synthesis and nicotine precursor transport) are coupled at multiple levels, and preset models and databases are used to realize fine decomposition and adjustment of metabolic flow. First, the base synthesis rate is output by the photosynthesis model, then the metabolic pathway is dynamically allocated according to the growth stage, and finally the global calibration is performed by introducing the stress response coefficient through stem flow anomaly detection. The simulation of internal metabolic activity of tobacco is realized, which enables the control system to make decisions from the physiological level rather than only from the environmental level, improving the accuracy of control; through the comparison of real-time stem flow data with the normal range, the stress state can be quickly identified and the metabolic trajectory can be automatically adjusted, enhancing the self-adaptation ability in a variable environment; the nicotine content target is converted into the daily average demand rate, ensuring the continuity and traceability of quality control, and avoiding the control lag caused by ambiguous targets in traditional methods.

[0039] Further, the method further comprises: When the light intensity data are all lower than the light compensation point threshold value and the stem flow rate data all show a downward trend in the continuous three monitoring periods, it is determined to enter the weak light stress state; Under the weak light stress state, the peak value of the second synthesis rate sub-trajectory is reduced by the first preset proportion, and the duration of its synthesis rate greater than zero is extended by a fixed period; When the crown temperature data are all higher than the high temperature threshold value and the fluctuation amplitude of the stem flow rate data are all higher than the normal fluctuation range in the continuous two monitoring periods, it is determined to enter the high temperature stress state; Among them, for the flue-cured tobacco variety K326, the light compensation point threshold value is determined to be 35 μmol·m -2 ·s -1This value is obtained from the plant physiology handbook of this variety, and represents the balance between the amount of carbohydrate produced by photosynthesis and consumed by respiration in the leaves under this light intensity.

[0040] In this embodiment, the high temperature threshold is set to 32°C. This threshold is determined based on the consensus in tobacco cultivation that when the canopy temperature consistently exceeds this value, the photosystem II efficiency of the crop will be significantly reduced, the respiration consumption will be intensified, and the growth will be inhibited. For mature tobacco plants, the normal fluctuation range of stem flow rate is usually 800 grams to 1500 grams per hour during the day when the water supply is sufficient and the weather is sunny. This range is obtained by statistical division (such as taking the 5% to 95% quantile) of a large amount of stem flow data under historical normal growth conditions.

[0041] It needs to be explained that the weak light stress is determined by selecting three consecutive monitoring periods, because short-term light intensity fluctuations (such as a cloud passing by) will not immediately cause irreversible changes in the physiological state of the crop. The sustained low light intensity for two consecutive periods (about 15-30 minutes) and the decrease in stem flow can more reliably indicate that the crop has entered a sustained stress physiological response mode rather than a transient fluctuation.

[0042] The high temperature stress is determined by selecting two consecutive monitoring periods, because the damage of high temperature to the crop is fast and cumulative. When the canopy temperature exceeds the high temperature threshold for two periods (about 10-20 minutes), and the stem flow shows abnormally high fluctuations (indicating that the transpiration cooling system is in a stressed state), it indicates that the risk of heat stress is very urgent, and the system needs to respond immediately to avoid irreversible damage to the leaf tissue.

[0043] Under high temperature stress, the peak appearance period of the transport rate curve is adjusted from the daytime to the temperature minimum period in the early morning of the next day, and the peak amplitude is adjusted by a second preset proportion.

[0044] The first preset proportion, for example, is 20%. That is, under weak light stress, the system will reduce the peak of the "second synthesis rate sub-trajectory" for accumulation and transport by 20% to reflect the reality of reduced photosynthetic output. The fixed period, for example, is 2 hours. That is, the duration of the synthesis rate greater than zero in this sub-trajectory is extended by 2 hours in total, which simulates the physiological compensation mechanism of the crop to compensate for insufficient efficiency by extending the effective metabolic time under adversity. The second preset proportion under high temperature stress, for example, is 15%. Under high temperature stress, the peak amplitude of the "transport rate curve" is increased by 15%, and its appearance period is adjusted to the cool morning. This simulates the physiological adaptive behavior of the crop to avoid the high temperature stress period and accelerate the completion of the transport task at suitable temperatures.

[0045] Based on the embodiments provided in the present application, the stress state determination is combined with the metabolic allocation logic to optimize resource allocation by modifying the peak value and period of the synthesis rate sub-trajectory. The state determination is based on the environmental parameters and stem flow trends in the continuous monitoring period, and the peak value and duration of the second synthesis rate sub-trajectory (for accumulation and transport) are adjusted under stress, or the period of the transport rate curve is re-planned under high temperature. Under weak light stress, the energy saving mechanism of plants under low light conditions is simulated by reducing the peak value and prolonging the duration, reducing the invalid metabolic consumption, while maintaining the basic growth needs; under high temperature stress, the transport peak value is adjusted to the period with lower temperature, which takes advantage of the stability of night physiological activity, avoiding the inhibition of high temperature on the accumulation of nicotine precursors, thereby improving the reliability of quality regulation.

[0046] In some embodiments, the following correspondence can be established: the dynamic physiological causal graph is a qualitative or semi-quantitative topological structure of the crop physiological mechanism model, which describes the causal relationship network between various physiological and environmental variables, and is the core skeleton of the model. The simplified metabolic flow model is a metabolic process calculation core for rapid verification in the crop physiological mechanism model. It is an engineering simplification of the complete and complex mechanism model, focusing on the prediction of metabolic flow. The stem flow prediction model is a sub-module of the crop physiological mechanism model for simulating water transport and judging physiological stability.

[0047] Further, as shown in Figure 2 The target concentration change trajectory is used as the optimization target, and the crop physiological mechanism model describing the relationship between the environment and the physiological process is used for reverse solving to calculate the target trajectory of the environmental parameter setting value required to achieve the optimization target over time, including: S201, constructing a dynamic physiological causal graph, the nodes of which include environmental parameters and metabolic flow rates; In the dynamic physiological causal graph, the nodes: the key nodes include environmental parameter nodes (light intensity, carbon dioxide concentration, canopy temperature, air humidity) and metabolic material nodes (carbohydrate instantaneous synthesis rate, nicotine precursor transport rate). The edges: the edges in the graph represent the causal relationship between the nodes, and the direction is from cause to effect. For example, there is an edge from "light intensity" to "carbohydrate instantaneous synthesis rate", and an edge from "canopy temperature" to "nicotine precursor transport rate". Dynamic allocation of causal weights: the weight allocation algorithm considers the growth stage and stem flow stability. First, a basic weight matrix Wbase(Stage) is pre-stored in the system, where each element The base weight represents the causal influence of node i on node j at a specific growth stage Stage. Then, a stem flow stability adjustment factor γ (ApEn) is introduced, which is a function of the real-time stem flow approximate entropy ApEn. When ApEn is low (poor stability), the value of γ increases, thereby amplifying the weights of edges related to water and energy transport. The final dynamic weight Wdynamic = Wbase(Stage) multiplied by γ(ApEn).

[0048] S202, based on the current growth stage and the real-time stem flow stability, dynamically allocate the causal weights of the edges in the physiological causal graph; starting from the target node of the target concentration change trajectory, perform reverse graph search to identify the key environmental parameters with the highest influence weight and their sensitive periods, thereby focusing the full-parameter optimization problem on a key subset; In some embodiments, the system stores a base weight table that defines the baseline influence weight of each environmental parameter on the metabolic flow at different growth stages (for example, the weight of "light → carbohydrate synthesis" is the highest in the vigorous growth period). In runtime, the system introduces real-time stem flow stability as an adjustment factor. When the stem flow stability is high, the system trusts the base weight; when the stability decreases, the system automatically increases the causal weights of environmental parameters related to "canopy temperature", "air humidity" and other parameters that directly affect water transpiration and transport. This means that when the crop is "sick", the control system will focus on its core physiological stability.

[0049] Starting from the target node of the target concentration change trajectory to be achieved later (such as the "peak value of carbohydrate synthesis rate at 12 noon"), the system performs reverse tracing in the dynamic causal graph. Along the causal edges, the system finds the environmental parameter nodes (such as "light intensity at 11 am" and "CO2 concentration at 10 am") that point to the target node and have the highest current dynamic weight, and marks these parameters and the time periods they are in (i.e., "sensitive periods"). This process focuses the thousands of variables that need to be optimized to a few key environmental parameters and key time periods, greatly improving the subsequent solving efficiency.

[0050] S203, according to the real-time collected environmental parameters and the crop stem flow data to judge the physiological situation of the crop; according to different situations, call the corresponding optimization algorithm from the strategy library including global exploration strategy, conservative repair strategy and economic energy saving strategy, and generate multiple different candidate environmental parameter trajectories in parallel; Among them, the physiological situation includes: "stable growth situation": the canopy temperature is between 22-28℃, the stem flow approximate entropy is greater than 0.6, and the light is sufficient. "Water stress situation": the stem flow rate is continuously lower than 15% of the lower limit of the normal range, and the stem flow approximate entropy is lower than 0.5.

[0051] The strategy library includes: "Global Exploration Strategy," which corresponds to a genetic algorithm and is suitable for "stable growth conditions," aiming to broadly search for potentially better new environmental solutions. "Conservative Repair Strategy," corresponding to a strongly constrained sequential quadratic programming method, is suitable for stress conditions and aims to quickly and smoothly restore crops to a safe state during physiologically sensitive periods. "Economic and Energy-Saving Strategy," corresponding to a particle swarm optimization algorithm with energy consumption as a penalty, is suitable when environmental goals are easily achieved, prioritizing the reduction of operating costs.

[0052] S204 uses a simplified metabolic flow model to verify and screen candidate environmental parameter trajectories, and outputs an initial solution set. For each trajectory in the initial solution set, an adjoint method channel is activated for adjustment. The adjoint method channel uses an attention mechanism to prioritize gradient calculation and optimization during a defined sensitive period. At the same time, an imitation learning channel is activated to perform pattern matching and fragment fusion between the trajectory and historical successful trajectories. The Simplified Metabolic Flux Model is a real-time computational version of the crop physiological mechanism model. It linearizes complex differential equations around the operating point; focuses on core metabolic fluxes such as carbohydrate synthesis and transport, ignoring secondary pathways; and uses lookup tables for key physiological parameters instead of real-time calculation. It is a set of algebraic equations that take environmental parameters as input and predict metabolic flux values ​​as output, used to quickly verify the physiological plausibility of candidate environmental trajectories.

[0053] The adjoint method is a technique for efficiently calculating the gradient of the objective function by constructing an "adjoint system" corresponding to the original differential equation system. In this scenario, it is used to quickly determine how to adjust the environmental trajectory to minimize the error between the predicted metabolic flux and the target value.

[0054] The attention mechanism achieves focus on sensitive periods through a time-weighted function. Therefore, this invention constructs a loss function L that is solved inversely.

[0055]

[0056] in, This represents the loss function, used to quantify the overall bias. This represents the total number of time points within the optimization period. Indicates at time The attention weight is set to a higher value (e.g., 5.0) during sensitive periods and 1.0 during non-sensitive periods. Indicates at time Predicted metabolic flow rate (unit: μmol·m) -2 ·s - ¹). Indicates at time Target concentration change trajectory Target value (unit: μmol·m -2 ·s -¹). Reference values ​​for metabolic flux (unit: μmol·m) -2 ·s - ¹), for example, determined based on the maximum or average metabolic flow rate in historical data, for normalization. Indicates at time Environmental parameter values ​​(e.g., temperature in °C, light intensity in μmol·m⁻²) -2 ·s - ¹). Reference values ​​representing changes in environmental parameters (units relative to environmental parameters) Consistency), for example, is determined based on the maximum or typical variation of environmental parameters allowed in historical data, and is used for normalization. This represents the weighting coefficient, used to balance the importance of metabolic flux tracking error and the penalty term for changes in environmental parameters.

[0057] The selection criteria for historical successful trajectories are: environmental regulation records that ultimately resulted in a metabolic flux achievement rate higher than 0.9 and low energy consumption under similar past reproductive stages and external weather types. Pattern matching employs a dynamic time warping algorithm to align candidate trajectories with historical successful trajectories on the time axis. Fragment fusion involves replacing or weighting corresponding portions of candidate trajectories with fragments from historical successful trajectories during similar time periods in the matching process, in order to introduce proven effective regulation patterns.

[0058] S205, based on the real-time stem flow stability index, the solutions output by the adjoint method channel and the solutions output by the imitation learning channel are weighted and fused to generate a preliminary target trajectory; the preliminary target trajectory is input into the stem flow prediction model to predict the stem flow sequence generated after executing the trajectory; The stem flow prediction model can be implemented using a mechanism-based empirical regression model or a data-driven machine learning model. Such model construction methods are mature technologies in this field.

[0059] Specifically, one approach is to establish empirical formulas or lookup tables based on crop physiological mechanisms and field experiments, using environmental factors (such as light, temperature, and humidity) and crop growth stages as input variables, to achieve rapid estimation of stem flow rate. Another approach is to employ lightweight machine learning models, such as Support Vector Regression (SVR) or computationally efficient Gated Recurrent Unit (GRU) networks, trained using historical stem flow data and environmental data to obtain predictive models.

[0060] S206, calculate the approximate entropy of the predicted stem flow sequence. If the approximate entropy is lower than the safe entropy threshold, it is determined that the trajectory has the risk of causing physiological instability. When the risk of instability is detected, adjust the causal weights of the nodes related to water transport stability in the dynamic causal graph and trigger a new round of reverse solution process. S207, if the predicted approximate entropy is higher than or equal to the safety entropy threshold, output the current preliminary trajectory as the target trajectory.

[0061] wherein, through analysis of a large amount of stem flow data of healthy tobacco plants, it can be determined that the approximate entropy thereof is generally within a stable range. Therefore, 0.5 can be set as the safety entropy threshold. When the approximate entropy of the predicted stem flow sequence is lower than 0.5, it is determined that there is a high risk of instability.

[0062] When the instability risk is detected, the risky environmental trajectory is not directly output. Instead, a closed-loop correction process is started: Adjust the causal graph: significantly increase the causal weight of nodes such as "air humidity" and "canopy temperature" in the dynamic causal graph, which are directly related to the stability of water transport.

[0063] Trigger re-solution: use the adjusted causal graph (now paying more attention to stability) to re-perform reverse graph search and subsequent optimization solution.

[0064] Iteration: the newly generated environmental trajectory instinctively avoids parameter combinations that lead to instability of water transport (such as excessively high temperature and excessively low humidity), thereby generating a new trajectory that is safer and more robust at the source. This process can be repeated until the predicted stem flow stability meets the standard.

[0065] Based on the embodiments provided in the present application, the dynamic causal graph and weight distribution enable the system to focus on environmental parameters that have the greatest impact on metabolic flow, reducing computational complexity and improving solution efficiency; generating candidate trajectories in parallel and combining multiple optimization channels (such as companion methods and imitation learning) enhances the diversity and feasibility of trajectories and avoids local optimal solutions; through stem flow prediction and approximate entropy checking, physiological instability risks can be identified in advance and re-solved, ensuring the safety of the regulation process. Overall, this method combines reverse reasoning with machine learning to achieve high-precision mapping from physiological goals to environmental parameters.

[0066] Further, in the process of generating candidate environmental parameter trajectories and preliminary target trajectories, photoperiod light constraint conditions and carbon balance temperature constraint conditions are introduced; wherein, The photoperiod light constraint condition includes: when generating the light intensity parameter trajectory, the change curve thereof is constrained to simulate the daily variation pattern of natural light intensity, i.e., the light intensity nonlinearly rises from zero or a basic value to a daytime peak value within a preset first time period after sunrise, and nonlinearly falls from the daytime peak value to a nighttime basic value within a preset second time period before sunset. The non-linear rise / fall is simulated by a sigmoid growth curve (Logistic function). Because the daily variation of natural light intensity is not linear, but a process of slow-fast-slow, i.e. accelerated growth after sunrise, slowed growth around noon, and accelerated decay before sunset, the sigmoid curve can well fit this natural rhythm.

[0067] The carbon balance temperature constraint includes: when generating the night temperature parameter trajectory, the value is constrained to be calculated according to the accumulated virtual amount of daytime photosynthetic products, so that the predicted night respiration consumption of the crop at this temperature is maintained in a proportion range dynamically set according to the quality target.

[0068] The first period (morning light start): refers to the period from sunrise to noon, for example, from the local sunrise time to 11:00. The second period (evening light decay): refers to the period from noon to sunset, for example, from 15:00 to the local sunset time.

[0069] Daytime peak: refers to the highest value that the light intensity needs to reach between the first and second periods, for example, for light-loving crops such as tobacco, the peak value at noon can be set to 1200 μmol·m -2 ·s -1 at noon on a sunny day. Night base value: refers to the minimum light level provided during the night to maintain basic metabolism or certain special processes (such as dark fixation) of the crop, for example, 0 μmol·m -2 ·s - ¹ (complete darkness) or 50 μmol·m -2 ·s -1 when light is supplemented.

[0070] The dynamically set proportion range: refers to the proportion range of night respiration consumption to daytime photosynthetic products. This range is dynamically set according to the quality target. For example: when pursuing high carbon-nitrogen ratio and thick leaf quality, a lower proportion range (such as 10%-15%) is set to minimize night carbon consumption and promote material accumulation. When pursuing normal growth rate, a medium proportion range (such as 15%-25%) is set.

[0071] Based on the embodiments provided in the present application, the photosynthetic rhythm constraint requires that the light intensity curve simulate the nonlinear change of sunrise and sunset, and the carbon balance constraint dynamically sets the night temperature according to the daytime photosynthetic product virtual quantity to maintain the proportion of respiration consumption and photosynthetic product. The photosynthetic rhythm constraint makes the artificial light closer to the natural light environment, reduces the stress response of plants due to light mutation, and promotes the smooth progress of photosynthesis; the carbon balance temperature constraint ensures the matching of night respiration consumption and daytime energy accumulation through virtual quantity calculation, avoids energy waste or shortage, and thus supports the realization of the quality target. These constraint conditions embed the plant physiological rules into the parameter generation, and improve the biological rationality of the environment trajectory.

[0072] Further, according to the target trajectory, a field control inversion problem is solved to generate a coordinated action sequence of the environment regulation device, including: It should be noted that the pre-established device disturbance propagation model in the present application refers to a mathematical model system for quantifying the space-time influence of the action of the environment regulation device on the physical field (such as temperature field, humidity field, flow field) inside the greenhouse. The specific implementation form of the model system in the controller is a device coordinated action primitive library and a spatial effect atlas corresponding to each primitive in the library. The spatial effect atlas is established by a pre-system identification method and can be calibrated online during system operation. The process of generating the coordinated action sequence is essentially based on the device disturbance propagation model (specifically represented by the primitive library and the spatial effect atlas) to calculate the optimal device driving instructions required to achieve the expected environmental change through matching and optimization.

[0073] Specifically, the device coordinated action primitive is the "excitation source" or "input unit" of the device disturbance propagation model. Each primitive represents a certain, executable device action combination. The spatial effect atlas is the "response database" or "output prediction" of the device disturbance propagation model. Each atlas accurately describes the output results of the model when the corresponding primitive is used as input, that is, the environmental parameter changes caused in the entire space. The correspondence between the primitive library and the atlas set constitutes a complete, query-based "device disturbance propagation model". The controller does not need to directly solve complex physical partial differential equations, but queries this "primitive-atlas" database to predict the effect of device action and combines the solution.

[0074] In the present application, solving a field control inversion problem has the core mathematical connotation that: given a desired environmental field change (i.e. spatial expected field increment) defined in the spatial and temporal domains, from a complex solution space composed of device physical actions, an optimal device control sequence is found to minimize the error between the synthesized environmental field generated by the sequence and the desired field.

[0075] The traditional inversion method can directly solve the inverse problem of the partial differential equation system, which is complex to calculate and difficult to ensure real-time. The present application creatively converts this problem into a two-level optimization problem to solve it efficiently: Discrete inversion in the element space (corresponding to the screening step): The infinite-dimensional device control sequence solution space is contracted to a solution space subset composed of a limited number of functionally explicit device cooperative action elements. By matching the spatial effect atlas with the spatial expected field increment, a high-quality and feasible initial solution range is quickly locked. This step is equivalent to solving the existence and general form of the inverse problem.

[0076] Continuous inversion in the parameter space (corresponding to the optimization adjustment step): Based on the selected high-quality element initial solution, fine-tuning is performed in its continuous parameter space (opening, power, and duration) to achieve the transition from coarse matching to fine matching, and finally accurately solve the optimal control instructions.

[0077] Therefore, the screening and optimization adjustment steps together constitute the unique and efficient solving path designed by the present application to solve the field control inversion problem.

[0078] A set of device cooperative action elements is predefined, each element representing a cooperative action mode of a group of environmental regulation devices at a specific time sequence; a spatial effect atlas is established for each element through system identification method, which is used to describe the change in the distribution of environmental parameters in the future period of time on the three-dimensional spatial grid of the tobacco planting area after executing this element; Among them, the device cooperative action element is a predefined device cooperative action combination. For example, the top cooling element: the action sequence is "open the skylight to 45% opening, delay 90 seconds, and then start the top two circular flow fans to 70% power, and run for 8 minutes". The horizontal uniform field element: the action sequence is "start the horizontal circulating fans on the east and west sides at the same time, the east fan power is set to 60%, and the west fan power is set to 40%, and run for 10 minutes.

[0079] For the system identification method, a subspace identification method can be used to establish the model. Experimental process: in the system initialization stage, each device cooperative action element is automatically executed in turn, and the data of all environmental sensors on the three-dimensional grid points of the greenhouse are collected at a frequency of 1 second. Based on these input-output data, a multi-input multi-output state space model is fitted for each element using a subspace identification algorithm (such as N4SID), which is the mathematical core of the spatial effect atlas of the element.

[0080] In each control cycle, the target trajectory's change in a short future time domain is compared with the predicted environment parameter values of each spatial grid point based on current sensor data, and the expected environment parameter change amount needed to be achieved on the main spatial grid points of the crop canopy in the short time domain is calculated, denoted as spatial expected field increment; In each control cycle (e.g., every 5 minutes), the system looks ahead to a short future time domain (e.g., the next 30 minutes). It compares the target trajectory's change in this period with the predicted environment parameter values of each grid point based on the current situation, and obtains how much the environment needs to be increased or decreased (e.g., 2°C warming) in the crop canopy area in the future. The spatial distribution of this required change is the spatial expected field increment.

[0081] From each device cooperative action primitive, at least one candidate cooperative action primitive whose spatial effect pattern best matches the spatial distribution pattern of the spatial expected field increment is selected. The spatial distribution pattern of the spatial expected field increment (e.g., a large amount of warming in the middle canopy and a small amount of warming in the edge) is compared with the spatial effect pattern of each primitive in the primitive library for similarity matching. It selects several primitives whose effect patterns best match the expected change pattern as candidates by calculating the spatial distribution correlation coefficient.

[0082] The action parameters of the selected candidate cooperative action primitives are optimized and adjusted, including device opening degree, power, and duration. The goal of the adjustment is to minimize the overall error between the adjusted combined spatial effect of the primitives and the spatial expected field increment, while considering the total energy consumption of the devices and the frequency of action as optimization penalty terms. The action parameters of the selected candidate primitives are fine-tuned (e.g., adjusting "70% power" of the fan to "65% power" or adjusting "8 minutes on" to "7.5 minutes"). The goal of the adjustment is to minimize the overall error between the adjusted combined effect of these primitives and the "spatial expected field increment". At the same time, in the optimization objective, the total energy consumption of the devices and the frequency of device start-stop in a short time are considered as penalty terms to ensure that the solution is accurate, economical, and friendly to device life.

[0083] The optimal cooperative action primitives and their parameters obtained by optimization and adjustment are converted into environmental control device control instructions and executed by the corresponding devices.

[0084] Mathematically, the spatial effect pattern can be expressed as a linear time-invariant system. For a given device cooperative action primitive, the environment parameter change amount ΔE(x, y, z, τ) caused by it at a specific spatial point (x, y, z) can be modeled as: ΔE(x, y, z, τ) = H(x, y, z, τ) * U(τ); ΔE(x,y,z,τ) represents the environmental parameter variation (e.g. temperature variation) at spatial point (x,y,z) and future time τ. H(x,y,z,τ) represents the spatial impulse response function of the primitive at point (x,y,z). It encapsulates the spatio-temporal characteristics of the disturbance propagation. Its physical meaning is: the environmental parameter variation at spatial point (x,y,z) and time τ caused by a unit impulse of the device input signal U(τ). Thus, the dimension of H is environmental parameter variation dimension / time (e.g. °C / s for temperature regulation). The input signal U(τ) is the normalized device action intensity, ranging from 0 to 1. U(τ) represents the input signal of the device collaborative action primitive (e.g. power percentage change over time). * denotes the convolution operation.

[0085] This model quantitatively describes the dynamic relationship between device action and spatial environmental field variation. It is the basis for solving the field control inversion problem, enabling the system to predict the spatio-temporal consequences of actions.

[0086] Regarding the best match on the spatial distribution pattern, the quantitative standard for matching is to calculate the spatial cross-correlation coefficient. The system calculates the Pearson correlation coefficient of the spatial effect pattern of each primitive and the spatial expected field increment in two dimensions: one is the spatial distribution pattern, and the other is the trend over time. The primitive with the highest comprehensive correlation coefficient is selected as the candidate.

[0087] Based on the embodiments provided in the present application, a spatial effect pattern is established for each primitive to describe the disturbance characteristics of device action on the three-dimensional environmental parameter in space, and the primitive parameters are screened and optimized by calculating the spatial expected field increment. The device collaborative action primitive abstracts complex device operations into standardized patterns, simplifying the logic of multi-device collaborative control and improving the generation efficiency of action sequences; the spatial effect pattern enables the system to predict the spatio-temporal distribution of device action in the planting area, achieving precise spatial regulation of environmental parameters and avoiding local over-regulation or under-regulation; the optimization process considers device energy consumption and action frequency, reducing operating costs, and through the inversion problem solution, the matching degree of the action sequence and the environmental target is ensured.

[0088] Further, the method further comprises a feedforward interference cancellation control based on weather forecasts, comprising: receiving short-term weather forecast data, converting the external environmental disturbance predicted to occur in the future period into a predicted disturbance field with a specific incoming direction, spatial distribution pattern and intensity change waveform; wherein, constructing the predicted disturbance field comprises: the process of converting the short-term weather forecast data into a predicted disturbance field with spatio-temporal characteristics is a quantization process from point to surface and from scalar to field. Specifically: Incoming direction: Determine the main incoming direction of the disturbance based on the wind direction forecast. For example, if the forecast is southeast wind, then the disturbance field is set to propagate from the southeast side of the greenhouse to the northwest side.

[0089] Spatial distribution pattern: Different spatial distribution patterns are preset based on the structure of the greenhouse and the type of disturbance. For example, for instantaneous strong light, it is modeled as a two-dimensional plane field that propagates uniformly downward from the top of the greenhouse; for gust, it is modeled as a gradient field that penetrates from the windward side window to the inside and decays with distance.

[0090] Intensity variation waveform: Based on the data of the predicted intensity variation over time, the "waveform" of the disturbance is generated. For example, the forecast shows that a thick cloud will cover the sun in 10 minutes and last for 20 minutes, then the light disturbance is quantified as a trapezoidal wave whose intensity linearly decreases from the current value to a low value after 10 minutes, and then linearly recovers after maintaining for 20 minutes.

[0091] Through the above implementation method, an abstract weather forecast data is concretized into a physical field model with a clear propagation path, distribution range and intensity variation in the physical space of the greenhouse.

[0092] In order to counteract the effect of the predicted disturbance field in the crop canopy area, one or more device cooperative action primitives are searched from the device cooperative action primitives to form a pre-action sequence; the design goal of the pre-action sequence is to make the net environmental change minimum after the generated regulation field and the predicted disturbance field are superimposed in the main spatial area of the crop canopy; It should be noted that the process of reverse search is constructed as an optimization problem with the goal of counteracting effect. The goal of this optimization problem is to find one or more primitives from the primitive library to make the superimposition effect of the generated regulation field and the predicted disturbance field in the crop canopy area minimum. Since the primitive library is discrete, the calculation amount of directly solving the combinatorial optimization problem is very large.

[0093] Therefore, a two-stage greedy search algorithm is adopted in this embodiment. The first stage performs fast screening to calculate the maximum counteracting potential that each primitive can produce when acting alone, and selects the top N primitives with the highest potential. The second stage performs combination evaluation to arrange and combine the N primitives, evaluate the joint counteracting effect of different combinations, and finally select the primitive combination with the best effect and the least total number of device actions to form the "pre-action sequence". This method significantly improves the search efficiency while ensuring the counteracting effect.

[0094] The pre-action sequence and the cooperative action sequence are weighted and fused; wherein the weight coefficient of the feedforward control is dynamically calculated based on the average deviation of the meteorological forecast data and the actual monitoring data in the past preset period, and the smaller the deviation, the higher the weight.

[0095] Based on the embodiments provided in the present application, meteorological forecast data is received, a predicted interference field is generated, and then a device cooperative action primitive is searched for the purpose of offsetting the interference, and a pre-action sequence and a feedback control sequence are weighted and fused. The feedforward control can actively respond to the external environmental changes (such as temperature fluctuations or wind speed changes) that are about to occur, reduces the hysteresis of the feedback control, and improves the stability of the environmental parameters; the weighting fusion mechanism dynamically adjusts the weight of the feedforward according to the prediction accuracy, avoiding the control misalignment caused by prediction errors; through the superposition optimization of the interference field and the regulation and control field, more uniform environmental control is achieved in the crop canopy area, and the overall regulation and control quality is improved.

[0096] Further, the feedforward interference offset control further includes the following safety strategies: A safety boundary based on historical operation data is set for the single action amplitude of each environmental regulation device to ensure that the action does not cause a dramatic shock to the environmental parameters; Among them, the safety boundary of the single action amplitude of the device is set based on two layers of information. The first layer is the physical limit of the device, which comes from the device manual, such as the maximum power of the heater and the maximum opening of the skylight. The second layer is the physiological safety range of the crop, which is obtained by analyzing the historical operation data, that is, the upper and lower limits of the operation amplitude that does not cause a dramatic fluctuation in the stem flow rate of the crop or an overshoot of the environmental parameters are taken as the soft constraints of the safety boundary. The final safety boundary is the more stringent one of the physical limit and the physiological safety range.

[0097] A device action logic conflict detection mechanism is established, when the calculated action sequence includes mutually exclusive actions, according to the priority of environmental parameter regulation, the low-priority action is automatically shielded or replaced; Among them, the device action logic conflict mainly includes functional exclusion and resource exclusion. Functional exclusion refers to the physical effect of the action being opposite, for example, starting the heater and opening the wet curtain to cool down, releasing carbon dioxide and opening the skylight to ventilate. Resource exclusion refers to the limitation of power capacity or water system, which does not allow some high-power or high-water consumption devices to run simultaneously.

[0098] The priority rules of environmental parameter regulation follow the following principles: first, to ensure the survival of the crop, second, to promote the growth of the crop, and finally, to optimize energy consumption. Specifically, the priority of relieving stress (such as high temperature, low temperature) is the highest; second, the parameters that affect the core metabolic process (such as photosynthesis), mainly light and carbon dioxide; then humidity and air flow and other parameters that affect the comfort of the microenvironment. When a conflict occurs, the system automatically shields or delays the action of low priority to ensure the realization of high priority targets.

[0099] A confidence condition for the effectiveness of feedforward control is set, when the predicted interference intensity is lower than the sensor noise level or the key meteorological data is missing, the feedforward control is automatically suspended, and only the feedback control is used to maintain the system operation.

[0100] Based on the embodiments provided in the present application, the safety boundary limits the single action amplitude of the device, the conflict detection mechanism identifies the mutually exclusive actions and automatically processes, and the confidence condition suspends the feedforward when the prediction is unreliable. This design creates the following effects in technical implementation: the safety boundary prevents the device damage or environmental mutation caused by the over-action of the device; the conflict detection mechanism eliminates the contradiction between the device actions through priority logic, avoiding system oscillation caused by control instruction conflict; and the confidence condition dynamically manages the feedforward control based on the deviation between the sensor data and the prediction, automatically degrades to feedback control when the data is unreliable, and improves the reliability in uncertain environment.

[0101] Further, the regulation effect is evaluated according to the time sequence change effect of both the environmental parameters and the crop stem flow data, and the key physiological parameters in the crop physiological mechanism model and the spatial transmission characteristics in the device disturbance propagation model are dynamically calibrated based on the evaluation results, including: In each evaluation period, three evaluation indexes of environmental parameter tracking error, metabolic flow realization degree and stem flow stability maintenance degree are calculated; Among them, the environmental parameter tracking error is obtained by calculating the root mean square error between the actual monitored environmental parameter values at each spatial position in the sliding time window and the corresponding target trajectory values; the metabolic flow realization degree is obtained by inputting the actual environmental parameter data into the simplified metabolic flow model to simulate the metabolic flow prediction value, and then calculating the correlation coefficient of the prediction value and the corresponding target concentration change trajectory value; and the stem flow stability maintenance degree is obtained by calculating the approximate entropy of the actual collected stem flow rate data in the time window, and comparing it with the preset expected stability range.

[0102] Among them, the environmental parameter tracking error: this index calculates the root mean square value of the difference between the actual monitored environmental parameter values (such as temperature, humidity) at each spatial position in the sliding time window (such as the past 1 hour) and the corresponding values of the target trajectory. It directly reflects the accuracy of environmental control.

[0103] Metabolic flow realization degree: this index uses correlation coefficient instead of tracking error, which is based on the core idea of the present application. Metabolic flow is the ultimate internal goal pursued by the present application, and environmental parameters are means. Correlation coefficient can measure the consistency of the actual environment driven metabolic flow change trend and the target concentration change trajectory. A high positive correlation coefficient means that even if the absolute value is biased due to uncontrollable factors, the system correctly guides the metabolic flow to be high when it needs to be high and low when it needs to be low, which is more in line with the nature of physiological regulation. Its calculation method is to input the actual environmental data into the simplified metabolic flow model to obtain the predicted metabolic flow sequence, and then calculate the Pearson correlation coefficient between the sequence and the target concentration change trajectory sequence.

[0104] Stem flow stability maintenance degree: This index evaluates by calculating the approximate entropy of the actual collected stem flow rate data in the same time window, and comparing it with a preset expected stability range (such as 0.5 to 1.0). The purpose is to ensure that the regulation behavior maintains the stability of the crop water transport system, avoiding physiological instability.

[0105] Based on the embodiments provided in the present application, the root mean square error, correlation coefficient and approximate entropy in the sliding time window are calculated, which respectively reflect the environmental control accuracy, physiological target achievement degree and plant physiological state stability. Multi-index evaluation covers the whole chain regulation effect from environment to physiology, helping the system to identify weak links; environmental parameter tracking error directly measures the accuracy of environmental control, providing a basis for equipment adjustment; metabolic flow realization degree verifies the achievement of physiological target through model simulation, ensuring the effectiveness of the regulation strategy; stem flow stability maintenance degree quantifies the health of plant water transport using approximate entropy, preventing physiological stress risk.

[0106] Further, the method further comprises: Based on the evaluation index, parameter sensitivity analysis is performed on the crop physiological mechanism model and the equipment disturbance propagation model to identify the top K physiological parameters that have the greatest impact on the metabolic flow realization degree, which constitute a set of to-be-calibrated physiological parameters, and the top M spatial transfer parameters that have the greatest impact on the environmental parameter tracking error, which constitute a set of to-be-calibrated equipment parameters. In some embodiments, the Morris screening method is used for parameter sensitivity analysis. This method is a highly efficient global sensitivity analysis method that quickly screens out key parameters that have the greatest impact on the output by perturbing multiple parameters in a reasonable space in a regular manner and observing the changes in the model output (such as metabolic flow realization degree, environmental parameter tracking error). Compared with the more accurate but computationally expensive Sobol method, the Morris method is more suitable for the computational efficiency requirements of the online calibration scenario while ensuring the identification of major sensitive parameters.

[0107] The recursive least squares method with a forgetting factor is used to take the environmental parameter tracking error and the metabolic flow realization degree as the joint optimization target to perform online estimation and update of the parameters in the set of to-be-calibrated physiological parameters and the set of to-be-calibrated equipment parameters, and write the updated parameter values into the crop physiological mechanism model and the equipment disturbance propagation model respectively, completing the model calibration.

[0108] Among them, recursive least squares with a forgetting factor was used for online estimation of key model parameters selected from the pool. The application process is as follows: First, the model to be calibrated is linearized around the operating point, transforming its parameter estimation problem into a linear regression problem. Then, the recursive least squares method iteratively updates the estimated values ​​of the model parameters using new monitoring data (such as environmental parameters and stem flow data). The "forgetting factor" assigns higher weight to new data, enabling the algorithm to track the dynamic characteristics of the system caused by slow changes in crop growth or the environment. The initial value of the forgetting factor is set to 0.99, and fine-tuned at different growth stages: during rapid crop changes, such as the transplanting and seedling establishment period, a smaller forgetting factor (e.g., 0.95) is used to quickly track changes; during stable stages, such as the maturity and harvest period, a larger forgetting factor (e.g., 0.995) is used to maintain the stability of parameter estimation and prevent excessive fluctuations.

[0109] The dynamic calibration process also includes the following adaptive mechanisms: Set the start conditions for model calibration. When the moving average of the tracking error of environmental parameters exceeds the first threshold for three consecutive evaluation periods, or the metabolic flow realization is lower than the second threshold for two consecutive evaluation periods, or the stem flow stability maintenance exceeds the expected stability range for two consecutive evaluation periods, the model calibration process is triggered. Different calibration intensity parameters are set according to different growth stages. During the transplanting and seedling establishment period, the forgetting factor of the recursive least squares method is increased to enable the model parameters to adapt to changes in crop status more quickly. During the maturity and harvest period, the forgetting factor is reduced to maintain the stability of the model parameters. Set reasonable physical range constraints for each parameter to be calibrated to ensure that the calibrated parameter values ​​conform to the physiological characteristics of crops and the physical characteristics of equipment. When the calibration results exceed the reasonable range, use boundary values ​​to replace them and record the abnormal state.

[0110] Based on the embodiments provided in this application, a set of key parameters that have the greatest impact on evaluation indicators is identified, and then recursive estimation is performed with tracking error and achievement rate as the objectives. This design achieves the following technical benefits: dynamic calibration enables the model to adapt to physiological changes and environmental drift during tobacco growth, maintaining the model's predictive accuracy; parameter sensitivity analysis focuses on key parameters, reducing the computational burden of calibration and improving efficiency; the recursive least squares method with a forgetting factor ensures the real-time nature of parameter estimation and a reasonable balance with historical data, allowing the model to be continuously optimized to adapt to actual regulatory needs.

[0111] According to another aspect of the embodiments of this application, an electronic device for implementing the above-described IoT-driven multi-parameter intelligent control method for tobacco growth environment is also provided. This electronic device may be... Figure 3 The terminal device or server shown. This embodiment uses this electronic device as an example of a server.Figure 3 As shown in the figure, the electronic device includes a memory 402, a processor 404 and a transmission device 406. The memory 402 stores a computer program. The processor 404 is configured to execute the steps of any of the method embodiments described above by using the computer program.

[0112] Optionally, in the embodiment, the electronic device can be located in at least one of the network devices in the computer network.

[0113] Optionally, the transmission device 406 is configured to receive or send data via a network. The network can include a wired network and a wireless network. In an example, the transmission device 406 includes a network interface controller (NIC) which can be connected to other network devices and routers through a network cable so as to communicate with the Internet or a local area network. In an example, the transmission device 406 is a radio frequency (RF) module which is configured to communicate with the Internet through a wireless manner. In addition, the electronic device further includes a display 408 and a connection bus 410 which is configured to connect the modules in the electronic device.

[0114] The above merely describes the preferred embodiments of the present application, and is not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation based on the content of the present application specification and drawings, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for intelligent multi-parameter control of tobacco growth environment driven by the Internet of Things, characterized in that, include: Based on real-time collected environmental parameters and crop stem flow data of tobacco planting areas, combined with the current growth stage and quality targets of tobacco, the target concentration change trajectory of key metabolites characterizing the internal physiological activities of tobacco is dynamically generated. Using the target concentration change trajectory as the optimization objective, the crop physiological mechanism model describing the relationship between the environment and physiological processes is used for inverse solution to calculate the target trajectory of the environmental parameter setpoints required to achieve the optimization objective over time. Based on the target trajectory and combined with a pre-established equipment disturbance propagation model, a coordinated action sequence of environmental control equipment is generated by solving a field control inversion problem; wherein, the equipment disturbance propagation model defines the spatiotemporal disturbance characteristics of each equipment action on the environmental state at different locations within the planting space; Execute the coordinated action sequence of the device to control the corresponding environmental control equipment to operate; The regulatory effect is evaluated based on the temporal changes in environmental parameters and crop stem flow data, and the key physiological parameters in the crop physiological mechanism model and the spatial transmission characteristics in the equipment disturbance propagation model are dynamically calibrated based on the evaluation results.

2. The IoT-driven intelligent multi-parameter control method for tobacco growth environment according to claim 1, characterized in that, The method, based on real-time collected environmental parameters and crop stem flow data of the tobacco planting area, combined with the current growth stage and quality targets of the tobacco, dynamically generates target concentration change trajectories of key metabolites characterizing the internal physiological activities of tobacco, including: The light intensity and temperature data from the real-time collected environmental parameters are input into the preset primary productivity model of photosynthesis to output the basic synthesis rate curve of carbohydrates over time. Using the current growth stage of tobacco as an index, a preset metabolic allocation strategy database is queried to obtain allocation parameters; the basic synthesis rate curve is decomposed into a first synthesis rate sub-trajectory for growth and a second synthesis rate sub-trajectory for accumulation and transport using the allocation parameters. The average daily demand rate of nicotine precursors is calculated by combining the nicotine content set in the quality target with the number of days from the current date to the planned harvest date, in order to generate a translocation rate curve. Based on the basic synthesis rate curve, the first synthesis rate sub-trajectory, the second synthesis rate sub-trajectory, and the transport rate curve, an initial set of metabolic flux trajectories is determined. The real-time collected crop stem flow data is compared with a preset normal range; when the data is abnormal, the attenuation coefficient is obtained according to the preset stress response coefficient table to globally adjust all curves in the initial metabolic flow trajectory set and output the target concentration change trajectory.

3. The IoT-driven intelligent multi-parameter control method for tobacco growth environment according to claim 2, characterized in that, The method further includes: When the light intensity data is below the light compensation point threshold and the stem flow rate data shows a downward trend for three consecutive monitoring cycles, it is determined that the state of weak light stress has been entered. Under low light stress, the peak value of the second synthesis rate sub-trajectory is reduced by a first preset ratio, while the duration of its synthesis rate being greater than zero is extended by a fixed period of time. When the canopy temperature data exceeds the high temperature threshold and the fluctuation range of the stem flow rate data exceeds the normal fluctuation range in two consecutive monitoring cycles, it is determined that the state of high temperature stress has been entered. Under high-temperature stress, the peak occurrence period of the transport rate curve is adjusted from daytime to the lowest temperature period in the early morning of the next day, and its peak amplitude is adjusted according to a second preset ratio.

4. The IoT-driven intelligent multi-parameter control method for tobacco growth environment according to claim 2, characterized in that, The optimization objective is to use the target concentration change trajectory as the optimization target, and to perform inverse solving using a crop physiological mechanism model describing the relationship between the environment and physiological processes, to calculate the target trajectory of the environmental parameter setpoints required to achieve the optimization objective over time, including: Construct a dynamic physiological cause-and-effect graph, whose nodes include environmental parameters and metabolic flow rates; Based on the current reproductive stage and real-time stalk flow stability, the causal weights of the edges in the physiological causal graph are dynamically allocated; a reverse graph search is performed starting from the target node of the target concentration change trajectory to identify the key environmental parameters and their sensitive periods that have the highest impact on achieving the target, thereby focusing the full parameter optimization problem on the key subset; The physiological state of the crop is determined based on real-time collected environmental parameters and crop stem flow data; based on different states, corresponding optimization algorithms are called from the strategy library, including global exploration strategy, conservative repair strategy and economic energy-saving strategy, to generate multiple different candidate environmental parameter trajectories in parallel. The candidate environmental parameter trajectories are validated and screened using a simplified metabolic flow model, and an initial solution set is output. For each trajectory in the initial solution set, an adjoint method channel is activated for adjustment. The adjoint method channel prioritizes gradient calculation and optimization during a defined sensitive period through an attention mechanism. At the same time, an imitation learning channel is activated to perform pattern matching and segment fusion between the trajectory and historical successful trajectories. Based on the real-time stem flow stability index, the solution output by the accompanying method channel and the solution output by the imitation learning channel are weighted and fused to generate a preliminary target trajectory. The preliminary target trajectory is input into the stem flow prediction model to predict the stem flow sequence generated after executing the trajectory; The approximate entropy of the predicted stem flow sequence is calculated. If the approximate entropy is lower than the safe entropy threshold, it is determined that the trajectory poses a risk of physiological instability. When the risk of instability is detected, the causal weights of the nodes related to water transport stability in the dynamic causal graph are adjusted, and a new round of reverse solution process is triggered. If the predicted approximate entropy is higher than or equal to the safety entropy threshold, then the current preliminary target trajectory is output as the target trajectory.

5. The IoT-driven intelligent multi-parameter control method for tobacco growth environment according to claim 4, characterized in that, In the process of generating the candidate environmental parameter trajectories and the preliminary target trajectories, photosynthetic rhythm illumination constraints and carbon balance temperature constraints are introduced; wherein... The photosynthetic rhythm light constraint conditions include: when generating the light intensity parameter trajectory, the constraint curve of its change needs to simulate the diurnal variation pattern of natural light intensity, that is, the light intensity rises non-linearly from zero or the base value to the daytime peak value in a preset first period after sunrise, and falls non-linearly from the daytime peak value to the nighttime base value in a preset second period before sunset. The carbon balance temperature constraint conditions include: when generating the nighttime temperature parameter trajectory, constraining its value to be calculated based on the virtual amount of cumulative photosynthetic products during the day, so that the predicted nighttime respiration consumption of crops at this temperature and the virtual amount of cumulative photosynthetic products during the day are maintained within a ratio range dynamically set according to the quality target.

6. The IoT-driven intelligent multi-parameter control method for tobacco growth environment according to claim 4, characterized in that, The step involves generating a coordinated action sequence of environmental control equipment based on the target trajectory and a pre-established equipment disturbance propagation model by solving a field control inversion problem, including: A set of equipment collaborative action primitives are predefined, each primitive representing a collaborative action mode of a set of environmental control equipment under a specific time sequence; a spatial effect map is established for each primitive through a system identification method, the spatial effect map being used to describe the distribution of changes in environmental parameters over a future period of time on a three-dimensional spatial grid of the tobacco planting area after the execution of this primitive; In each control cycle, the change of the target trajectory in a short time domain in the future is compared with the environmental parameter values ​​of each spatial grid point predicted based on the current sensor data. The expected change of environmental parameters that needs to be achieved at the main spatial grid points of the crop canopy in the short time domain is calculated and denoted as the spatial expected field increment. From the cooperative action primitives of various devices, at least one candidate cooperative action primitive whose spatial effect map best matches the spatial distribution pattern of the spatial expected field increment is selected. The motion parameters of the selected candidate cooperative action primitives are optimized and adjusted; wherein, the motion parameters include equipment opening degree, power and duration; the goal of the adjustment is to minimize the overall error between the synthesized spatial effect of the primitives after adjustment and the expected spatial field increment, while taking the total energy consumption of the equipment and the frequency of action as optimization penalty terms; The optimal collaborative action primitives and their parameters obtained through optimization and adjustment are converted into control commands for environmental control equipment and sent to the corresponding equipment for execution.

7. The IoT-driven intelligent multi-parameter control method for tobacco growth environment according to claim 6, characterized in that, The method also includes feedforward disturbance cancellation control based on weather forecasts, including: It receives short-term weather forecast data and converts the predicted external environmental disturbances that will occur in the future into a predicted disturbance field with a specific incoming direction, spatial distribution pattern and intensity variation waveform; With the goal of counteracting the effect of the predicted interference field in the crop canopy area, a pre-action sequence is formed by backward searching one or more device cooperative action primitives from each device cooperative action primitive; the design goal of the pre-action sequence is to minimize the net environmental change after the regulation field it generates is superimposed on the predicted interference field in the main spatial area of ​​the crop canopy. The pre-action sequence and the coordinated action sequence are weighted and fused; wherein, the weight coefficient of the feedforward control is dynamically calculated based on the average deviation between the meteorological forecast data and the actual monitoring data in the past preset period. The smaller the deviation, the higher the weight.

8. The IoT-driven intelligent multi-parameter control method for tobacco growth environment according to claim 7, characterized in that, The feedforward interference cancellation control also includes the following security strategies: Set a safety boundary based on historical operating data for the single action range of each environmental control device. Establish a device action logic conflict detection mechanism. When the calculated action sequence includes mutually exclusive actions, automatically block or replace low-priority actions based on the priority adjusted by environmental parameters. Set confidence conditions for feedforward control to take effect. When the predicted interference intensity is lower than the sensor noise level or key meteorological data is missing, feedforward control will be automatically suspended, and the system will be maintained by feedback control alone.

9. The IoT-driven intelligent multi-parameter control method for tobacco growth environment according to claim 4, characterized in that, The evaluation of the regulatory effect based on the temporal changes of both environmental parameters and crop stem flow data, and the dynamic calibration of key physiological parameters in the crop physiological mechanism model and spatial transfer characteristics in the equipment disturbance propagation model based on the evaluation results, includes: Within each evaluation period, three evaluation indicators are calculated: environmental parameter tracking error, metabolic flow realization degree, and stem flow stability maintenance degree. The environmental parameter tracking error is obtained by calculating the root mean square error between the actual monitored environmental parameter values ​​at each spatial location within the sliding time window and the corresponding target trajectory values; the metabolic flow realization degree is obtained by inputting the actual environmental parameter data into the simplified metabolic flow model for forward simulation to obtain the metabolic flow prediction value, and then calculating the correlation coefficient between the prediction value and the corresponding target concentration change trajectory value; the stem flow stability maintenance degree is obtained by calculating the approximate entropy of the actual collected stem flow rate data within the time window and comparing it with the preset expected stability range.

10. The IoT-driven intelligent multi-parameter control method for tobacco growth environment according to claim 9, characterized in that, The method further includes: Based on the evaluation indicators, parameter sensitivity analysis was performed on the crop physiological mechanism model and the equipment disturbance propagation model to identify the top K physiological parameters that have the greatest impact on the realization of metabolic flux as a set of physiological parameters to be calibrated, and the top M spatial transfer parameters that have the greatest impact on the tracking error of environmental parameters as a set of equipment parameters to be calibrated. Using a recursive least squares method with a forgetting factor, and taking environmental parameter tracking error and metabolic flux realization as joint optimization objectives, the parameters in the set of physiological parameters to be calibrated and the set of equipment parameters to be calibrated are estimated and updated online. The updated parameter values ​​are then written into the crop physiological mechanism model and the equipment perturbation propagation model, respectively, to complete the model calibration.

Images