Greenhouse environment control method and system

Abstract

The invention relates to the technical field of greenhouse environment control, in particular to a greenhouse environment control method and system. The method comprises the following steps: obtaining data of a crop cultivation stage and data of a soil monitoring sensor to obtain soil state monitoring filling data; then, identifying the abnormal fluctuation state of the soil acidity and alkalinity, and estimating the root growth inhibition degree according to the state to obtain root growth inhibition gradient data; secondly, predicting the soil looseness requirement according to the root inhibition data, and adjusting the soil acidity and alkalinity to realize balance adaptation; and then, based on the soil state data, carrying out adaptive control training, constructing a soil environment regulation and control model, and sending the soil environment regulation and control model to a cloud platform to execute greenhouse environment control. According to the invention, the greenhouse environment control technology is optimized, so that the greenhouse environment control technology is more perfect.

Description

Technical Field

[0001] The present invention relates to the technical field of greenhouse environment control, and particularly to a greenhouse environment control method and system. Background Art

[0002] Greenhouse environment control technology realizes the all-round regulation of crop growth by precisely adjusting factors such as temperature, humidity, light, and carbon dioxide concentration, thereby increasing the yield and quality of agricultural crops and reducing the impact of the external environment on agricultural production. The environment control method based on intelligent technology has gradually become the focus of research. By combining advanced technologies such as Internet of Things technology, sensors, data analysis, and artificial intelligence, environmental factors such as soil and air can be monitored in real time, and the micro-environment conditions in the greenhouse can be dynamically adjusted, so as to provide the best growth environment for crops. This can not only optimize resource utilization and reduce energy consumption, but also avoid the inhibition or damage of crop roots by precisely regulating key factors such as soil pH, humidity, and temperature, and improve the growth potential and stress resistance of crops. However, there is a problem in a traditional greenhouse environment control method that the recognition accuracy of the degree of root growth inhibition is low, resulting in a large error in the transformation of the soil in the greenhouse. Summary of the Invention

[0003] Based on this, it is necessary to provide a greenhouse environment control method and system to solve at least one of the above technical problems.

[0004] To achieve the above object, a greenhouse environment control method, the method includes the following steps: Step S1: Obtain the data of the crop cultivation stage in the greenhouse; based on the data of the crop cultivation stage, and collect the soil state data through the soil monitoring sensor to obtain the soil state monitoring filling data; Step S2: Identify the abnormal fluctuation state of the soil pH value for the soil state monitoring filling data to obtain the abnormal fluctuation state of the soil pH value; estimate the degree of root growth inhibition based on the abnormal fluctuation state of the soil pH value to obtain the root growth inhibition gradient data; Step S3: Predict the demand for root soil looseness based on the root growth inhibition gradient data to obtain the predicted data of the demand for root soil looseness; adapt the abnormal fluctuation state of the soil pH value according to the predicted data of the demand for root soil looseness to obtain the soil pH value balance adaptation data; Step S4: Perform adaptive control training on the soil state variables according to the soil pH value balance adaptation data and the predicted data of the demand for root soil looseness to obtain the soil state variable control training data; construct a soil environment regulation model based on the soil state variable control training data to obtain the soil environment regulation model; send the soil environment regulation model to the cloud platform to execute the greenhouse environment control method.

[0005] Preferably, step S1 includes the following steps: Step S11: Obtain the data of the crop cultivation stage in the greenhouse; Step S12: Evaluate the root system status of the crops between different cultivation stages for the crop cultivation stage data to obtain the root system status of the crops between different cultivation stages; Step S13: Based on the root system status of the crops and through the soil monitoring sensor, collect the soil status data to obtain the soil status monitoring data; Step S14: Fill in the missing values for the soil status monitoring data to obtain the soil status monitoring filled data.

[0006] Preferably, step S2 includes the following steps: Step S21: Analyze the soil penetration structure for the soil status monitoring filled data to obtain the soil penetration structure status data; Step S22: Use the preset soil pH anomaly recognition model to identify the abnormal fluctuation status of the soil pH for the soil status monitoring filled data to obtain the abnormal fluctuation status of the soil pH; Step S23: Quantify the soil hardening degree for the soil penetration structure status data based on the abnormal fluctuation status of the soil pH to obtain the soil hardening degree quantification data; Step S24: Measure the loss of soil microbial activity according to the abnormal fluctuation status of the soil pH and the soil hardening degree quantification data to obtain the microbial activity loss measurement data; Step S25: Estimate the degree of root growth inhibition for the root system status of the crops based on the abnormal fluctuation status of the soil pH, the soil hardening degree quantification data, and the microbial activity loss measurement data to obtain the root growth inhibition gradient data.

[0007] Preferably, step S23 includes the following steps: Step S231: Perform multi-period clustering processing on the abnormal fluctuation status of the soil pH to obtain the pH fluctuation distribution data; Step S232: Fit the trend of the pH double-peak fluctuation intensity for the pH fluctuation distribution data to obtain the pH double-peak fluctuation intensity trend; Step S233: Quantify the loss of soil colloid cohesion for the soil penetration structure status data according to the pH double-peak fluctuation intensity trend to obtain the soil colloid cohesion loss data; Step S234: Reconstruct the soil penetration resistance equivalent based on the soil colloid cohesion loss data to obtain the soil penetration resistance equivalent reconstruction data; Step S235: Perform integral processing on the soil structure density for the soil penetration resistance equivalent reconstruction data to obtain the soil structure density integral data. Step S236: Quantify the degree of soil hardening based on the reconstructed data of soil infiltration resistance equivalent values and the integral data of soil structure compactness to obtain the quantified data of soil hardening degree.

[0008] Preferably, step S24 includes the following steps: Step S241: Analyze the reduction of air / water circulation space in soil porosity for the quantified data of soil hardening degree to obtain the reduction data of air / water circulation space; Step S242: Conduct a regression analysis on the increase in soil bulk density for the quantified data of soil hardening degree to obtain the regression data of soil bulk density increase; Step S243: Simulate and estimate the inhibition of enzyme activity based on the abnormal fluctuation state of soil pH to obtain the estimated data of enzyme activity inhibition; Step S244: Estimate the equivalent loss of soil microbial metabolism based on the estimated data of enzyme activity inhibition, the regression data of soil bulk density increase, and the reduction data of air / water circulation space to obtain the estimated data of equivalent metabolic loss; Step S245: Measure the loss of soil microbial activity based on the estimated data of equivalent metabolic loss and the estimated data of enzyme activity inhibition to obtain the measured data of microbial activity loss.

[0009] Preferably, step S25 includes the following steps: Step S251: Conduct a time-series analysis of the mutation intensity of the abnormal fluctuation state of soil pH to obtain the time-series intensity data of pH mutation; Step S252: Estimate the incremental effect of ion toxicity based on the time-series intensity data of pH mutation to obtain the estimated data of incremental ion toxicity effect; Step S253: Evaluate the attenuation of soil oxygen exchange based on the quantified data of soil hardening degree to obtain the soil oxygen exchange attenuation data; Step S254: Analyze the reduction of nutrient dissolution effectiveness based on the estimated data of incremental ion toxicity effect, the soil oxygen exchange attenuation data, and the measured data of microbial activity loss to obtain the reduction data of nutrient dissolution effectiveness; Step S255: Estimate the degree of root growth inhibition of crop roots based on the reduction data of nutrient dissolution effectiveness to obtain the root growth inhibition gradient data.

[0010] Preferably, step S254 includes the following steps: Conduct an analysis of the probability of heavy metal valence migration for the estimated data of incremental ion toxicity effect to obtain the probability data of heavy metal valence migration; Conduct a redox potential coupling analysis on the soil oxygen exchange attenuation data and the measured data of microbial activity loss according to the probability data of heavy metal valence migration to obtain the potential coupling loss coefficient; Perform pore structure degradation gradient analysis based on soil oxygen exchange attenuation data to obtain pore structure degradation gradient data; Perform nutrient adsorption site attenuation modeling based on the potential coupling loss coefficient and pore structure degradation gradient data to obtain adsorption site attenuation data; Perform nutrient dissolution effectiveness reduction analysis based on the adsorption site attenuation data to obtain nutrient dissolution effectiveness reduction data.

[0011] Preferably, step S3 includes the following steps: Step S31: Normalize the root growth inhibition gradient data to obtain normalized root growth inhibition gradient data; Step S32: Predict the root soil looseness requirement based on the normalized root growth inhibition gradient data to obtain predicted root soil looseness requirement data; Step S33: Control the stage-by-stage increment of root soil microorganisms based on the normalized root growth inhibition gradient data and the predicted root soil looseness requirement data to generate stage-by-stage increment control data for microorganisms; Step S34: Adapt the abnormal fluctuation state of soil pH according to the predicted root soil looseness requirement data and the stage-by-stage increment control data for microorganisms to obtain soil pH balance adaptation data.

[0012] Preferably, step S4 includes the following steps: Step S41: Perform logical learning on the predicted root soil looseness requirement data, the stage-by-stage increment control data for microorganisms, and the soil pH balance adaptation data respectively to obtain root soil looseness logical learning data, stage-by-stage increment control learning data for microorganisms, and soil pH balance adaptation learning data respectively; Step S42: Perform adaptive control training on soil state variables based on the root soil looseness logical learning data, the stage-by-stage increment control learning data for microorganisms, and the soil pH balance adaptation learning data to obtain soil state variable control training data; Step S43: Construct a soil environment regulation model based on the soil state variable control training data to obtain a soil environment regulation model; Step S44: Send the soil environment regulation model to the cloud platform to execute the greenhouse environment control method.

[0013] Preferably, the present invention also provides a greenhouse environment control system for executing the greenhouse environment control method as described above. The greenhouse environment control system includes: A soil state acquisition module for obtaining data on the crop cultivation stage in the greenhouse; based on the data on the crop cultivation stage, and collecting soil state data through soil monitoring sensors to obtain soil state monitoring filling data; The root growth inhibition degree estimation module is used to identify the abnormal fluctuation state of soil pH value for the soil state monitoring filling data, and obtain the abnormal fluctuation state of soil pH value; estimate the root growth inhibition degree based on the abnormal fluctuation state of soil pH value, and obtain the root growth inhibition gradient data; The pH balance adaptation module is used to predict the root soil looseness demand based on the root growth inhibition gradient data, and obtain the root soil looseness demand prediction data; adapt the abnormal fluctuation state of soil pH value according to the root soil looseness demand prediction data to obtain the soil pH balance adaptation data; The soil environment regulation model construction module is used to perform adaptive control training of soil state variables based on the soil pH balance adaptation data and the root soil looseness demand prediction data, and obtain the soil state variable control training data; construct a soil environment regulation model based on the soil state variable control training data, and obtain the soil environment regulation model; send the soil environment regulation model to the cloud platform to execute the greenhouse environment control method.

[0014] The beneficial effects of the present invention are as follows. During the process of cultivating crops in a greenhouse, the first step is to collect data on the growth stage of the crops through a sensor system. By recording multiple factors such as the growth cycle of the crops, climatic conditions, light, humidity, etc., the growth state of the crops can be comprehensively understood. This data provides a basis for subsequent soil state monitoring and optimization. Based on these data, data such as soil temperature and humidity, pH value, and oxygen content collected by soil monitoring sensors can be further combined to comprehensively understand the state of the soil and fill in various parameters of the soil. These data are particularly important for precise farmland management and can help agricultural managers monitor and adjust the environment in real time, improving agricultural production efficiency and crop yields. By analyzing the collected soil state data, the acidity and alkalinity of the soil can be identified and monitored. The acidity and alkalinity of the soil are one of the key factors affecting crop growth, and too high or too low acidity and alkalinity will have an adverse impact on the development of crop roots and nutrient absorption. By monitoring the fluctuations in soil acidity and alkalinity through an algorithm, abnormal fluctuations can be detected in a timely manner, and then the abnormal state of soil acidity and alkalinity can be identified. Next, based on the identified abnormal fluctuation state, the degree of root growth inhibition can be estimated to obtain root growth inhibition gradient data. This step can provide precise soil management suggestions for agricultural managers to help adjust the acidity and alkalinity of the soil to improve the root growth environment. After obtaining the predicted data on the looseness requirements of the root soil and the adaptation data on the balance of soil acidity and alkalinity, the next step is to perform adaptive control training on soil state variables. This training process will automatically adjust the control parameters according to different soil states and crop growth requirements, thereby achieving precise control of the soil environment. Through adaptive control training, the system can quickly respond under different environmental conditions and adjust parameters such as soil humidity, acidity and alkalinity, and looseness to optimize the crop growth conditions to the greatest extent. After training, the obtained soil environment regulation model can effectively predict and regulate various variables in the soil environment, and finally form a stable and intelligent soil management system to ensure that the crops in the greenhouse grow in the optimal environment. Therefore, the present invention makes an optimized treatment of a traditional greenhouse environment control method, solves the problem that the traditional greenhouse environment control method has a low recognition accuracy of the degree of root growth inhibition, resulting in a large error in the transformation of the soil in the greenhouse, improves the recognition accuracy of the degree of root growth inhibition, and reduces the error in the transformation of the soil in the greenhouse. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic flow chart of the steps of a greenhouse environment control method; Figure 2 is Figure 1 a detailed implementation step flow chart of step S2 in Figure 3 is Figure 1Schematic diagram of the detailed implementation steps of step S3 in the method. Detailed implementation manners

[0016] Please refer to Figures 1 to 3 , a greenhouse environment control method, the method comprising the following steps: Step S1: Obtain the crop cultivation stage data in the greenhouse; based on the crop cultivation stage data, and collect soil state data through a soil monitoring sensor to obtain soil state monitoring filling data; Step S2: Identify the abnormal fluctuation state of soil pH value for the soil state monitoring filling data to obtain the abnormal fluctuation state of soil pH value; estimate the root growth inhibition degree based on the abnormal fluctuation state of soil pH value to obtain root growth inhibition gradient data; Step S3: Predict the root soil looseness requirement based on the root growth inhibition gradient data to obtain root soil looseness requirement prediction data; perform soil pH balance adaptation on the abnormal fluctuation state of soil pH value according to the root soil looseness requirement prediction data to obtain soil pH balance adaptation data; Step S4: Perform soil state variable adaptive control training according to the soil pH balance adaptation data and the root soil looseness requirement prediction data to obtain soil state variable control training data; construct a soil environment regulation model based on the soil state variable control training data to obtain a soil environment regulation model; send the soil environment regulation model to the cloud platform to execute the greenhouse environment control method.

[0017] In an embodiment of the present invention, referring to Figure 1 as described, it is a schematic diagram of the step flow of a greenhouse environment control method of the present invention. In this example, the greenhouse environment control method comprises the following steps: Step S1: Obtain the crop cultivation stage data in the greenhouse; based on the crop cultivation stage data, and collect soil state data through a soil monitoring sensor to obtain soil state monitoring filling data; In the embodiments of the present invention, in the process of obtaining the data of the crop cultivation stage in the greenhouse, first, the image recognition and acquisition module installed in the greenhouse operation management system is used to collect the growth images of the crops at different time periods. The image acquisition device uses an industrial-grade CMOS image sensor with more than 5 million pixels, and takes 24 hours as a sampling period. The image data of the crop roots and surface parts is collected every 2 hours. After the image acquisition is completed, the built-in YOLOv5 (You Only Look Once version 5) model is used to identify and classify the crop growth characteristics in the image. The identified characteristics include but are not limited to the number of leaves, chlorophyll color value, stem height, and root color difference. After the identification is completed, the current cultivation stage of the crop is determined according to the preset growth period division standard. This standard is based on the growth cycle division standards of different crops published by the National Agricultural Science Data Center, and time stamp synchronization processing is carried out in combination with the current environmental control data in the greenhouse to complete the time series data archiving of the crop cultivation stage. Next, the multi-parameter soil monitoring sensor buried 10 cm deep in the crop root zone is used to collect the soil state data at the current time point. The soil state data includes 7 basic variables such as temperature, humidity, pH value, electrical conductivity (EC), ammonia nitrogen concentration, nitrate nitrogen concentration, and phosphorus and potassium ion concentration. The collection cycle is synchronized with the image collection. After the data is transmitted to the edge computing node through the RS-485 industrial communication interface, data missing detection is carried out. The detection rule uses the Z-score method to identify abnormal missing points, and the missing point filling method uses the K-nearest mean interpolation algorithm to perform 5th-order distance weighted averaging on the same parameter data points in adjacent time periods to obtain complete and non-missing soil state monitoring filling data.

[0018] In another embodiment, in the process of greenhouse environment control, first, it is necessary to obtain the data of the crop cultivation stage in the greenhouse. This data is obtained through the growth cycle archive table corresponding to the crop variety set manually. The table has been stage-labeled according to the crop variety number, sowing time, and preset growth stage division standard. Each stage includes the seedling stage, growth stage, flowering stage, fruiting stage, etc., and corresponds to the range values of the soil physical and chemical parameters required for the growth stage. After obtaining the current cultivation stage of the crop, the real-time soil state parameters are collected in combination with the soil monitoring sensors arranged in the greenhouse. The sensors are deployed at three depths of 10 cm, 20 cm, and 30 cm from the surface layer, and data on five dimensions of soil temperature, moisture, pH, electrical conductivity, and oxygen concentration are continuously collected and recorded every five minutes. The time series interpolation algorithm is used to linearly interpolate and fill in the missing data during the sensing interval to construct the soil state monitoring filling data on a unified time axis. The data structure is a complete set of state parameters corresponding to each monitoring depth at each moment.

[0019] Step S2: Identify the abnormal fluctuation state of soil pH value for the soil state monitoring filling data to obtain the abnormal fluctuation state of soil pH value; estimate the degree of root growth inhibition based on the abnormal fluctuation state of soil pH value to obtain the root growth inhibition gradient data; In the embodiment of the present invention, during the process of identifying the abnormal fluctuation state of soil pH value for the soil state monitoring filling data, discriminant analysis is performed using a Random Forest model. The input of the model is the pH value sequence data for 48 consecutive hours, and the output is the abnormal fluctuation state level label. The labels are divided into five levels, representing "extremely low fluctuation", "low fluctuation", "moderate fluctuation", "high fluctuation", and "severe fluctuation" respectively. After the identification is completed, the results are cross-mapped and analyzed with the root growth sensitivity matrix corresponding to the current crop growth period. A root sensitive response regression model (implemented by support vector regression SVR) is used to quantitatively estimate the degree of root growth inhibition. The model takes the abnormal fluctuation level of soil pH value, the historical root nutrient absorption rate, and the soil temperature and humidity linkage value as input variables, and the output result is the root growth inhibition gradient data between 0 and 1. The higher this value, the more severe the root inhibition. This regression model has been cross-validated five times at 1000 different measured points of crop roots, and it is determined that its estimation error is controlled within ±5%.

[0020] In another embodiment, based on the above soil state monitoring filling data, the differential sliding window algorithm is used to detect abnormal fluctuations in the pH value sequence. Specifically, with a two-hour window as the unit, the difference between the maximum and minimum values, the average change amplitude, the ratio of range to standard deviation of the pH sequence within the window are extracted, and a benchmark threshold is set to determine whether there is an abnormal mutation fluctuation. If the range suddenly rises beyond the threshold in two consecutive windows, it is marked as the abnormal fluctuation state of soil pH value, and the abnormal marked time period and the corresponding depth data are output. Subsequently, based on comprehensive parameters such as the oxygen concentration change gradient, conductivity change trend, and water evaporation rate within the abnormal fluctuation interval, the degree of root growth inhibition is estimated. The multi-factor weighted cumulative addition method is used to perform weighted product superposition on the duration of oxygen concentration below the average value, the frequency of water loss rate higher than the benchmark ratio, and the number of times of conductivity surge, and the quantified root growth inhibition gradient data is output. The corresponding gradient value is obtained for each sampling depth, indicating its inhibition intensity level.

[0021] Step S3: Predict the root soil looseness requirement based on the root growth inhibition gradient data to obtain the root soil looseness requirement prediction data; perform soil pH balance adaptation on the abnormal fluctuation state of soil pH value according to the root soil looseness requirement prediction data to obtain the soil pH balance adaptation data; In the embodiments of the present invention, during the process of predicting the soil looseness requirement based on the root growth inhibition gradient data, the BP neural network (Back Propagation Neural Network) is used as the core prediction algorithm. The input layer is set with 4 neurons, and the corresponding input variables are the root inhibition gradient value, the crop growth stage code, the current soil bulk density value, and the change value of the soil infiltration rate in the previous 48 hours. The hidden layer is set with two layers, and the number of neurons in each layer is 8 and 4 respectively. The output layer is a single neuron representing the looseness prediction value, and the value ranges from 0 to 10, indicating the demand level of the current root system for looseness. The training samples of the network model are sourced from the measured data of the root zone compaction changes in the greenhouse within a three-year cycle. After the output is completed, the value is passed into the multi-factor soil pH regulation adaptation module. This module conducts a cross-comparison based on the current abnormal pH fluctuation level and the soil buffer capacity curve, and uses the ion balance adjustment model to determine the type of regulator to be added and the application dose per unit volume. The additive types are limited to three types: calcium and magnesium carbonates, humic acid salts, and organic acids. The adjustment direction and amplitude are determined through the intersection point of the buffer curve. Finally, the output is the specific soil pH balance adaptation data, which is packaged in the standard JSON structure format for subsequent model construction phase calls.

[0022] In another embodiment, the root growth inhibition gradient data is used as the input, and the demand prediction is carried out through the looseness requirement mapping table. The table records the crop type, the current cultivation stage, and the empirical values of soil physical structure adjustment corresponding to the historical root inhibition data. The linear interpolation method is used to match the interval where the current gradient value is located, and the predicted data of the root soil looseness requirement is output, including the required soil porosity, the adjustment range of compaction, and the recommended ventilation frequency. Subsequently, based on the linkage relationship between the demand data and the aforementioned soil pH abnormal fluctuation state data, the soil pH balance adaptation analysis is carried out. The analysis content includes the soil buffer capacity index (judged by the fluctuation of bicarbonate ion concentration) in the abnormal pH interval, the change law of the redox state, and the rhizosphere microbial distribution structure. A logical combination analysis is performed on them, and whether there are physical and chemical conditions for effectively adjusting the soil acid-base state is judged through the cross-judgment of the logical matrix. If it is judged to be in an adjustable state, the required amount of acid-base regulator to be put in and the adjustment duration are calculated by combining the predicted pH adjustment range and the buffer capacity, and the soil pH balance adaptation data is generated.

[0023] Step S4: Perform adaptive control training on the soil state variables according to the soil pH balance adaptation data and the predicted data of the root soil looseness requirement to obtain the soil state variable control training data; construct a soil environment regulation model based on the soil state variable control training data to obtain the soil environment regulation model; send the soil environment regulation model to the cloud platform to execute the greenhouse environment control method.

[0024] In the embodiments of the present invention, during the process of adaptively controlling the training of soil state variables according to the soil pH balance adaptation data and the root soil looseness requirement prediction data, a reinforcement learning framework is adopted to construct the training process, and Proximal Policy Optimization (PPO) is used as the main training algorithm. The input features include: the pH balance adjustment factor value, the looseness prediction level, the historical regulation feedback error, and the change in the current crop root absorption rate. The environmental state is the real-time values of 7 basic soil variables collected currently. The behavior strategies include 3 combinations of regulation schemes: the microbial application scheme, the buffer agent adjustment scheme, and the soil deep plowing cycle adjustment scheme. Each strategy combination corresponds to different weight allocation rules. The reward function takes the reduction amplitude of the root inhibition gradient as the main index. The training data is asynchronously and parallelly trained on the GPU cluster for 48 hours. After the training is completed, the policy model parameter file is exported, and this parameter file is integrated into the overall soil environment regulation model through the model packaging module. After the soil environment regulation model is constructed, a TensorRT optimization structure is constructed through the model integrator. This structure integrates the reinforcement learning policy model and the historical feedback prediction sub-model. The model interface call format is configured as the RESTfulAPI protocol at the model deployment layer. The model is called by the edge device at regular intervals to execute the model inference process, and the inference cycle is set to once every 6 hours. The inference results returned include the regulation intensity value, the regulation target variable value, and the expected root state change amount. After the inference is completed, the model deployment component uploads the optimized model parameter file to the cloud platform server. The server receiving end is built based on the Alibaba Cloud IoT suite, and the model file upload channel is implemented using the MQTT protocol. After receiving the model, the cloud platform will automatically distribute the parameters to the greenhouse environment control execution modules of each shed, and trigger relevant automatic operation modules such as fertilization, pH regulators, and soil tillage devices through the PLC control port, so as to realize the implementation operation of the model output and complete a round of complete greenhouse environment regulation task logic loop.

[0025] In another embodiment, the obtained soil pH balance adaptation data and the predicted data of the root soil looseness requirement are respectively subjected to feature discretization processing to construct an input vector group. The pH balance data includes the pH adjustment amount, the regulator type code, and the application period; the looseness requirement data includes the compaction correction amount, the target value of the change in porosity, and the ventilation control parameter. All input vectors are uniformly converted into interval-type discrete data. After being processed by the maximum entropy normalization method, conditional rule training is carried out using state classification labels (such as "strong acid + high compaction", "slightly alkaline + insufficient ventilation"). The state change influence path under the feature combination is extracted by the information gain ratio principle to complete the adaptive control training of the soil state variables. The output result is the control strategy weight distribution table and threshold structure corresponding to each state combination, generating the soil state variable control training data. Based on the adaptive control training data, a soil environment regulation model is constructed. The mapping relationship between each group of input feature combinations and the control output in the control strategy weight table is represented by a tree structure. In the regulation model, the root node is set as the soil state input item, the first layer is the acid-base state control logic, and the second layer is the physical structure regulation logic. Each leaf node is connected to specific control execution instructions, such as the pH regulator dosing channel, the ventilation device opening and closing frequency setting channel, the water regulating valve control instruction, etc. After the regulation model is constructed, structure verification and boundary testing are carried out to verify whether the output instructions are reasonable when abnormal input states occur. Finally, the soil environment regulation model is encapsulated and uploaded to the control center of the agricultural cloud platform according to the preset data interaction interface standard through a unified transmission protocol. After receiving the model, the platform automatically deploys it to the corresponding greenhouse control module to complete the model access and put it into operation, realizing the full-process automatic execution of the entire greenhouse environment control method.

[0026] Step S1 includes the following steps: Step S11: Obtain the crop cultivation stage data in the greenhouse; Step S12: Evaluate the crop root states between different cultivation stages for the crop cultivation stage data to obtain the crop root states between different cultivation stages; Step S13: Based on the crop root states, collect soil state data through soil monitoring sensors to obtain soil state monitoring data; Step S14: Fill in the missing values in the soil state monitoring data to obtain the filled soil state monitoring data.

[0027] In the embodiments of the present invention, in the process of obtaining the data of the crop cultivation stage in the greenhouse, first, multi-angle high-definition camera devices are installed on both sides of the main channel in the greenhouse. The camera devices adopt the Sony IMX477 sensor module with 12 million pixels, integrated with automatic exposure and automatic white balance adjustment algorithms. The shooting frequency of each device is set to once every 3 hours. The camera is installed at a height of 1.2 meters from the ground and tilted at an angle of 35 degrees to cover the whole plant image of the crop. The length of the shooting area covers 12-meter crop rows. The images are transmitted to the image processing module in real time through the local edge gateway. In the image processing module, the ResNet50 deep neural network is loaded to extract the stem and leaf structure information and color distribution characteristics of the crops in the image. At the same time, the three-dimensional spatial structure data obtained by the LiDAR (Light Detection and Ranging) device is combined to calibrate the plant height and stem thickness. All the extracted features are input into the crop growth stage recognition model according to different crop types. The model adopts a multi-class logistic regression structure and is trained based on the standardized crop staging data released by the Chinese Academy of Agricultural Sciences, and outputs the growth stage label of the crop corresponding to the current image. The label information is divided into six stages: "seedling stage", "tillering stage", "jointing stage", "heading stage", "filling stage", and "maturity stage". Each label value is automatically associated with the system timestamp and then stored in the time series database InfluxDB for data synchronization calibration for subsequent root system evaluation and soil monitoring. In the process of evaluating the root system status of crops between different cultivation stages for the data of the crop cultivation stage, root system image acquisition units are arranged equidistantly in the crop root zone. This unit consists of a 2MP micro camera module with infrared fill light function and a micro industrial endoscope. The depth of the probe inserted into the soil is set to 15 cm, and the probe is inserted into a position 3 cm away from the plant center at a 45-degree angle through a flexible catheter. The acquisition frequency is set to once a day. After the image is acquired, it is transmitted to the root system image processing server. The server runs a root system segmentation model based on the U-Net structure. Through this model, the root system contour in the image can be effectively separated from the soil background. Subsequently, an image analysis algorithm based on OpenCV is used to calculate the root length density, average diameter, number of branches, root hair density, and color distribution histogram. The extracted features are combined with the growth stage information obtained previously and input into the root system status evaluation model. The model is a decision tree ensemble model with multiple input layers, and outputs the root system activity level and the change gradient of the root system growth trend at the current stage. The status level is divided into levels 0 to 5, and the gradient value indicates whether there are obvious characteristics of growth stagnation, decline, or rapid growth of the root system at the current stage. All evaluation results are stored in the PostgreSQL database after unified numbering for the next-stage soil status acquisition scheduling module to call.During the process of collecting soil status data based on the root system status of crops and through soil monitoring sensors, the SoilSens multi-parameter sensing unit provided by SIEMENS is adopted. Each sensor unit integrates a temperature sensor (NTC thermistor), a soil moisture sensor (frequency domain reflectance type FDR), a pH electrode, an electrical conductivity EC probe, and a nitrate nitrogen ion-selective electrode. The sensors are installed 10 cm deep from the ground surface in the root zone of the plants. The sensor spacing is one group every 2 meters, and each group of three sensors is located in the main root, lateral root, and non-planted area of the crop respectively. Each sensor is connected to the edge gateway through the RS-485 interface. The data collection period is once per hour, and the collected parameters are uploaded to the greenhouse central control system by the gateway using the Modbus RTU protocol. After the data is uploaded, the acquisition scheduling module discriminates the priority according to the root system status level. If the root system status is evaluated as level 4 or 5, that is, when the root system is abnormal, the collection period will be shortened to once every 30 minutes. All the original data formats are uniformly converted into the standard CSV format and timestamped, and then stored in the local time series database. During the process of filling in the missing values of the soil status monitoring data, first, the data integrity detection module analyzes the continuity and interval validity of the collected data. The detection module uses the Sliding Window algorithm to set 24 hours as the window interval, and judges whether the continuous collection times of each sensor parameter within the window meet 96 times. If the continuous times are insufficient or the missing measurement exceeds 5%, it is marked as missing. Subsequently, for all the data segments marked as missing, a method combining time series interpolation and spatial interpolation is used for filling. For the time series interpolation part, a fifth-order Lagrange polynomial is used to fit and calculate the filling value for the context values. For the spatial interpolation part, the Kriging spatial interpolation method is used, and the data at the corresponding time points of the sensors within a radius of 5 meters around are weighted and calculated according to the variogram. The average value after combining the two interpolation methods is used as the final filling value. All the filling results are relabeled and attached with the interpolation source type. The soil status monitoring data after filling is packaged into a structured data table, and after integrity verification through the CRC checksum, it is used as the input data source for the next module.

[0028] Step S2 includes the following steps: Step S21: Analyze the soil penetration structure of the filled soil status monitoring data to obtain soil penetration structure status data; Step S22: Use the preset soil pH anomaly recognition model to identify the abnormal fluctuation state of the soil pH of the filled soil status monitoring data to obtain the abnormal fluctuation state of the soil pH; Step S23: Quantify the soil hardening degree based on the abnormal fluctuation state of the soil pH for the soil penetration structure status data to obtain the soil hardening degree quantification data; Step S24: measuring soil microbial activity loss according to the abnormal fluctuation state of soil pH and the quantitative data of soil hardening degree, and obtaining microbial activity loss measurement data; Step S25: estimating the degree of root growth inhibition of crop roots based on the abnormal fluctuation state of soil pH, the quantitative data of soil hardening degree and the measurement data of microbial activity loss, and obtaining root growth inhibition gradient data.

[0029] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes: Step S21: performing soil permeability structure analysis on soil state monitoring filling data to obtain soil permeability structure state data; In the embodiment of the present invention, in the process of performing soil permeability structure analysis on the soil state monitoring filled data, firstly, the filled soil state data including multi-dimensional parameter data such as moisture content, conductivity, resistivity, density and temperature are synchronized and aligned according to the timestamp, and the aligned data are input into the three-dimensional soil permeability analysis module, which is based on the linkage operation of the improved Green-Ampt permeability model and the dynamic physical statistics model. The Green-Ampt model part uses the moisture content difference and the pressure head estimation parameter to calculate the permeability rate per unit time, and the dynamic physical statistics module generates a spatial thermal map of the hydraulic conductivity of the soil per square meter through the historical measurement data of the sensor, and uses a convolutional neural network (CNN) to extract the soil stratification characteristics therefrom. The CNN network contains 4 convolutional layers and 2 pooling layers, and integrates Batch Normalization batch normalization and ReLU activation function are used to enhance the permeability recognition ability of boundary areas. Finally, the permeability distribution curve is obtained according to different depth levels. The analysis results are recorded in JSON format, including the hydraulic conductivity (in cm / day), capillary suction distribution (in kPa) and effective porosity (in %) of each depth interval, and the permeability structure is marked as high permeability, medium permeability and low permeability for subsequent quantitative call of hardening degree.

[0030] Step S22: using a preset soil pH abnormality recognition model to identify the abnormal fluctuation state of soil pH on the soil state monitoring filling data to obtain the abnormal fluctuation state of soil pH; In the embodiments of the present invention, during the process of using a preset soil pH anomaly recognition model to identify the abnormal fluctuation state of soil pH for the soil state monitoring filling data, a pH fluctuation anomaly recognition model with a fully connected feedforward neural network (FeedforwardNeural Network) structure is used for the recognition operation. The model structure includes an input layer, two hidden layers, and an output layer. The input layer receives 432 groups of pH value data at 10-minute intervals within the past 72 hours as a feature sequence. The first hidden layer contains 128 neuron nodes and uses the tanh activation function. The second hidden layer contains 64 neuron nodes and uses the sigmoid function for non-linear mapping. When training the model, the manually labeled abnormal fluctuation data of the ground sampling pH monitoring points is used as the label set. The recognition rule is defined as follows: if the change amplitude of the pH value exceeds 1.2 units within 24 hours and the duration exceeds 2 hours, it is determined as an acid-base abnormal fluctuation. The output is a boolean value array and an abnormal duration matrix. After the abnormal recognition, further classification is performed according to the pH value fluctuation direction (acidic or alkaline) and frequency. Finally, the recognition result is marked with a timestamp and written into the database to provide a basis for subsequent soil hardening analysis module.

[0031] Among them, the preset soil pH anomaly recognition model is a classification model constructed based on supervised learning. In the construction stage, historical data of the greenhouse environment is first collected and cleaned. The data sources include soil pH sensor data collected in multiple cycles, irrigation and fertilization records in the corresponding time periods, crop types and cultivation stage records, temperature and humidity change records, and manually labeled pH anomaly events. The sampling frequency of the used pH data is once every 30 minutes, and the collection cycle is not less than 180 days. The data cleaning steps include outlier removal, missing value interpolation filling, and unit normalization processing. After that, the data is grouped according to crop types and seasons, and feature extraction operations are performed separately. The extracted features include instantaneous pH value, first derivative, three-hour moving average, six-hour standard deviation, maximum change rate, extreme value difference within the day-night cycle, etc., a total of 12-dimensional features. In the model construction stage, Gradient Boosting Decision Tree (GBDT) is used as the basic classifier. The manually labeled abnormal fluctuation samples are used as positive examples, and the samples with a fluctuation range less than ±0.3 and stable for more than 12 hours are used as negative examples to construct the training set. During the model training process, cross-validation is used to tune the parameters, controlling the tree depth to be 6, the learning rate to be 0.1, the subsampling rate to be 0.8, the training rounds to be set to 150 rounds, and the loss function to be the logarithmic loss function. After training, the accuracy is evaluated on an independent validation set. The evaluation indicators include accuracy, recall rate, and F1-score. Finally, the model accuracy reaches 92.3%, which is solidified as the core algorithm of the greenhouse soil pH anomaly recognition module. In the online recognition stage, the real-time collected soil status monitoring filled data is input into the model for reasoning and judgment, and the binary classification recognition result and the abnormal type label are output. The abnormal type labels include three situations: "rapid acid enhancement", "continuous alkaline fluctuation", and "short-term high-frequency fluctuation". At the same time, the fluctuation score value is output as the subsequent model reference input.

[0032] Step S23: Quantify the soil hardening degree of the soil penetration structure state data based on the abnormal fluctuation state of the soil pH, and obtain the quantified soil hardening degree data; In the embodiments of the present invention, in the process of quantifying the soil hardening degree based on the abnormal fluctuation state of soil pH value for the data of soil infiltration structure state, first, the effective porosity value and the capillary suction value in the soil infiltration structure state are used as the main input variables, and the abnormal direction and the fluctuation intensity in the pH value fluctuation state are input into the hardening degree quantification model as adjustment factors. This model is a non-linear regression model constructed based on Support Vector Regression (SVR). The model parameters are obtained through training after determining the hardness grade of the actually collected soil samples. The sample hardness value is measured by the Proctor compaction test method, calibrated by the standard compaction energy and the percentage of soil sample compactness. The quantified output value is a continuous value between 0 and 1. The closer the value is to 1, the harder the soil. After combining the numerical results with the spatial position data, a two-dimensional soil hardening isogram is formed and output as a GeoTIFF format layer file, which is used as the reference value input for the geographical factor compensation module in the measurement of microbial activity.

[0033] Step S24: Measure the loss of soil microbial activity according to the abnormal fluctuation state of soil pH value and the quantified data of soil hardening degree, and obtain the measured data of the loss of microbial activity; In the embodiments of the present invention, in the process of measuring the loss of soil microbial activity according to the quantification data of the abnormal fluctuation state of soil pH and soil hardening degree, first, three features, namely, the pH value fluctuation amplitude, the fluctuation direction (acidic enhancement or alkaline enhancement), and the fluctuation duration per hour in the abnormal fluctuation state data of pH, are extracted to construct a feature matrix. The quantification value of soil hardening degree within the corresponding time period, the effective porosity, and the capillary suction in the soil infiltration structure are combined as multi-dimensional input variables and fed into the microbial activity loss prediction model. The prediction model adopted is a time series prediction model based on the structure of Long Short-Term Memory (LSTM). This model includes 1 input layer, 2 layers of LSTM memory units, and 1 fully connected regression output layer. The input data is organized in the form of a sliding window in chronological order. Each time window is 24 hours, with each hour as a time step, a total of 24 steps. The data within the window includes pH value fluctuation characteristics, hardening quantification data, and porosity values. The model is trained using the artificially measured values of historical microbial activity as the supervision signal. The Adam optimizer is used in the training process, and the learning rate is set to 0.0005. The loss function is the mean square error (MSE). The model prediction output is the percentage of microbial activity loss per unit volume of soil within the next 6 hours. The prediction result is output as tabular data in CSV format, where each row represents the predicted activity loss value at a sensor sampling point position and the corresponding timestamp. At the same time, an abnormal determination label field is output. The determination rule is that when the activity loss rate is greater than 30%, it is marked as "moderate activity decay", and when the activity loss rate exceeds 50%, it is marked as "severe activity decay". Finally, this prediction data is used as the input for the root growth inhibition degree estimation module for invocation.

[0034] Step S25: Estimate the root growth inhibition degree of the crop root system based on the abnormal fluctuation state of soil pH, the quantification data of soil hardening degree, and the measurement data of microbial activity loss to obtain root growth inhibition gradient data.

[0035] In an embodiment of the present invention, in the process of estimating the degree of root growth inhibition of crop roots based on the abnormal fluctuation state of soil pH, the quantified data of soil hardening degree, and the measured data of microbial activity loss, a fusion model is constructed to uniformly standardize the three types of data sources and input them into the root inhibition evaluation neural network model. The neural network adopts a three-input parallel channel structure, and inputs the acid-base fluctuation feature vector (including the maximum fluctuation amplitude, duration, and fluctuation frequency), the hardening quantification feature vector (including the average hardening coefficient and the maximum hardening area), and the microbial loss vector (including the average activity loss rate and the coordinates of the maximum activity loss point) respectively. Each channel contains two fully connected layers with 128 dimensions and 64 dimensions, and feature splicing is performed at the end. After splicing, it is sent to two hidden layers and an output layer, and the output is the root growth inhibition gradient value. The gradient value ranges from 0 to 1 and is divided into five grades, from no inhibition to extremely severe inhibition. After adding geographical coordinates and time identifiers to the gradient data, it is output as a time series heat map file for subsequent calling by the root soil looseness demand prediction module.

[0036] Step S23 includes the following steps: Step S231: Perform multi-period clustering processing on the abnormal fluctuation state of soil pH to obtain the pH fluctuation distribution data; Step S232: Fit the pH double-peak fluctuation intensity trend of the pH fluctuation distribution data to obtain the pH double-peak fluctuation intensity trend; Step S233: Quantify the loss of soil colloid cohesion of the soil infiltration structure state data according to the pH double-peak fluctuation intensity trend to obtain the soil colloid cohesion loss data; Step S234: Reconstruct the soil infiltration resistance equivalent based on the soil colloid cohesion loss data to obtain the soil infiltration resistance equivalent reconstruction data; Step S235: Perform soil structure compactness integration processing on the soil infiltration resistance equivalent reconstruction data to obtain the soil structure compactness integration data; Step S236: Quantify the soil hardening degree according to the soil infiltration resistance equivalent reconstruction data and the soil structure compactness integration data to obtain the soil hardening degree quantification data.

[0037] In the embodiments of the present invention, the operation of performing multi-period clustering processing on the abnormal fluctuation state of soil pH is carried out by constructing a clustering framework based on time series features. First, the abnormal fluctuation state of soil pH is sliced with a 24-hour time window to ensure that each sample contains complete daily change characteristics. Subsequently, statistical features are extracted from the pH change sequence within each time window, including 10 feature vectors such as mean, variance, range, skewness, kurtosis, maximum rising speed, maximum falling speed, rising duration, falling duration, and oscillation frequency. Then, the K-means Clustering algorithm is used to perform unsupervised clustering processing on the samples. The number of clusters is set to 4, and the initial centroid is initialized in the K-means++ manner. The clustering convergence criterion is that the change in the sum of squared errors is less than 0.001, obtaining the clustered pH fluctuation distribution data. Each category corresponds to a different type of pH fluctuation pattern, including high-frequency small-amplitude fluctuation type, sudden increase and sudden decrease type, stable high-value type, and stable low-value type. Each clustering category is accompanied by a fluctuation period diagram and a peak marker sequence as output. The operation of fitting the pH double-peak fluctuation intensity trend for the pH fluctuation distribution data is realized by introducing a nonlinear regression function. First, peak recognition processing is performed on the pH change sequence with an obvious double-peak morphology in the clustering results. The extreme value detection method of a sliding window with a window width of 2 hours is used to identify the two main pH peak points within a day. Subsequently, a fitting model based on the double Gaussian Kernel function is constructed to fit the central position, fluctuation amplitude, and diffusion width of the two peaks for each pH sequence, and a double-peak fluctuation intensity trend parameter vector is output. The fitting result of each sequence includes the main peak center time, the secondary peak center time, the main and secondary peak values, the time difference between peaks, the main and secondary peak diffusion factors, and the fitting residual. Those with a fitting residual greater than the set threshold (0.2) are determined as non-typical double-peak patterns and do not participate in subsequent processing.

[0038] In the operation of quantifying the loss of soil colloid cohesion for soil infiltration structure state data according to the trend of the double-peak fluctuation intensity of acidity and alkalinity, first, select the soil permeability data and soil dissolved organic carbon (DOC) concentration data collected synchronously with the pH fluctuation period, perform a first-order difference process on the permeability change, statistically calculate the degree of permeability reduction within the double-peak period and perform normalization processing. At the same time, calculate the standard deviation of the DOC concentration fluctuation within the corresponding period, establish the colloid cohesion loss coefficient, and calculate the soil colloid cohesion loss data for each observation point by weighted combination of the degree of permeability decline and the standard deviation of DOC fluctuation. This data represents the degree of disintegration of the soil microstructure caused by charge neutralization imbalance and organic matter destruction in the colloid structure under the action of the acid-base double peak fluctuation. The output result is quantified on a percentage scale, ranging from 0 to 100. The operation of reconstructing the soil infiltration resistance equivalent value based on the soil colloid cohesion loss data is carried out by simulating the resistance effect suffered by water conduction in the multi-layer soil structure. Using the multi-layer one-dimensional water infiltration model, the input parameters include the initial soil moisture content, saturated hydraulic conductivity, porosity, and colloid loss factor. By numerically iterating to simulate the water infiltration time delay under different colloid loss conditions, the equivalent infiltration resistance time is obtained, and the equivalent resistance time under different colloid loss levels is interpolated to construct the resistance equivalent curve. The output is the soil infiltration resistance equivalent reconstruction data, which is a two-dimensional matrix with the horizontal axis being the depth and the vertical axis being the resistance level, and the unit is minutes.

[0039] The operation of performing soil structure compactness integration processing on the soil infiltration resistance equivalent reconstruction data uses numerical integration methods to perform weighted averaging on the resistance data in different depth intervals. The integration window is set as a 5-cm depth sliding window, and the data weighting method within each window is a gradient decay weight, with the weight coefficient linearly decreasing from the center to the boundary. After processing, a continuous depth compactness curve is generated, and quadratic interpolation is performed on the curve to obtain the soil structure compactness integration data, which is used to describe the tightness of the soil interlayer structure. The numerical unit is the integration density value (dimensionless). The operation of quantifying the soil hardening degree based on the soil infiltration resistance equivalent reconstruction data and the soil structure compactness integration data is achieved by establishing a linear regression quantification model. Align the resistance equivalent matrix and the compactness integration data according to the depth coordinate, extract the maximum resistance value and the corresponding integration density value of the key depth segments, and input them into the trained multi-variable linear regression model. The regression weights are obtained by training with the manually labeled data of the known hardening degree in the historical samples. The model output is the soil hardening degree quantification data, and the quantification value range is from 0 to 1, divided into three categories: mild (<0.3), moderate (0.3 - 0.7), and severe (>0.7). The hardening degree of each measurement point is output in this numerical form, accompanied by time tags and geographical positioning tags for use in subsequent calculation of the root growth inhibition gradient.

[0040] Step S24 includes the following steps: Step S241: Conduct an analysis on the reduction of air / water circulation space of soil porosity for the quantified data of soil hardening degree to obtain the reduction data of air / water circulation space; Step S242: Conduct a regression analysis on the increase of soil bulk density for the quantified data of soil hardening degree to obtain the regression data of soil bulk density increase; Step S243: Conduct a simulation estimation on the inhibition of enzyme activity based on the abnormal fluctuation state of soil pH value to obtain the estimated data of enzyme activity inhibition; Step S244: Conduct an estimation on the equivalent loss of soil microbial metabolism based on the estimated data of enzyme activity inhibition, the regression data of soil bulk density increase, and the reduction data of air / water circulation space to obtain the estimated data of equivalent loss of metabolism; Step S245: Conduct a measurement on the loss of soil microbial activity based on the estimated data of equivalent loss of metabolism and the estimated data of enzyme activity inhibition to obtain the measurement data of microbial activity loss.

[0041] In the embodiments of the present invention, during the process of analyzing the reduction of the air and water circulation space of soil porosity based on the quantified data of soil hardening degree, first, a multi-layer time series data structure is constructed based on the quantified data of soil hardening degree. Each node of this structure represents the corresponding relationship between the hardening degree and soil depth at a specific moment, with the unit being the average hardening value within a millimeter thickness range. After constructing this time series, a pore structure distribution inference function based on the principal component decomposition algorithm is called. Using the principal component mapping relationship between the hardening degree and porosity distribution in 196 historical samples, interpolation fitting is performed on the soil layers with depths between 0 and 30 cm respectively. The estimated porosity is calculated every 2 cm. After estimating the porosity, the air and water channel connectivity classification processing is carried out using the pore connectivity discrimination rule. This discrimination rule sets the channel threshold based on the limit value range of the Poisson distribution. If the non-capillary porosity at a certain depth point is lower than 12%, it is determined as a fractured channel; if it is higher than 22%, it is determined as a connected channel. Finally, normalization weighting is performed in the vertical structure according to whether the channels are connected or not to obtain the reduction rate of the air and water circulation space of each layer. The reduction rate is expressed as a percentage value, with the unit being the proportion of the loss of connection in the pore structure per cubic decimeter. The processing method for the regression analysis of soil bulk density increment based on the quantified data of soil hardening degree is carried out based on the first-order difference processing and the multi-variable linear regression method. First, the input quantified data of soil hardening degree is two-dimensionally tensorized and expanded according to the soil depth. The expansion format is a matrix of depth dimension multiplied by time dimension, with the unit being the hardening value corresponding to a millimeter thickness. After completing the tensorization, the first-order central difference method is used to process the change gradient between each time segment to extract the growth rate of the hardening degree at different times. Subsequently, a fixed-form linear regression equation is constructed to express the bulk density increment value as a linear combination of the hardening value and its first derivative. The parameters are fitted by the least squares regression method. The training samples come from the measured data of bulk density and hardening degree of 640 groups from 23 different plots in history. In each group of data, the bulk density is in grams per cubic centimeter, and the hardening value is expressed as the compaction rate per unit thickness. The finally output regression data of bulk density increment is presented in tabular form, where each row represents the predicted value of bulk density increment at a certain depth.

[0042] The process of simulating and estimating enzyme activity inhibition based on the abnormal fluctuation state of soil pH is completed through a combination of logical condition judgment and time window filtering. At the beginning of the process, three basic dimensional data indicators, namely the average pH value within 7 consecutive days, the maximum daily fluctuation amplitude, and the number of consecutive days exceeding the standard, are first extracted from the pH fluctuation state. Based on the experimental data of soil microenvironment biochemical reactions, a set of judgment conditions is constructed. It is defined that if the average pH value is less than 5.2 or greater than 8.5, the daily fluctuation is greater than 0.7, and the number of consecutive days exceeding the standard is more than 3 days, it is defined as a high inhibition level state. The judgment of the medium inhibition level state is that the average pH value is between 5.2 and 5.8 or between 8.0 and 8.5, the fluctuation value is above 0.5, and the number of consecutive days exceeding the standard is more than 2 days. The low inhibition state is other data points that do not meet the high or medium conditions. After the classification judgment is completed, a corresponding inhibition coefficient is assigned to each state, which are set to 0.8 for high inhibition, 0.5 for medium, and 0.2 for low inhibition. The output result is a sequence of inhibition coefficients corresponding to each data window, with the unit of dimensionless ratio, ranging from 0 to 1. The operation method for estimating the equivalent loss of soil microbial metabolism based on the enzyme activity inhibition estimation data, soil bulk density increment regression data, and air moisture circulation space loss data is processed by a weighted average method based on weight factor allocation. First, the three types of input data are aligned in chronological order and subjected to standard normalization processing. The normalization method is range normalization, and the normalized data falls within the range of 0 to 1. After normalization, weights are assigned to each index. The enzyme inhibition weight is set to 0.5, the bulk density increment weight is set to 0.3, and the pore structure loss weight is set to 0.2. In the weighted processing stage, the method of multiplying each item and summing is used. The equivalent loss value of metabolism for each record is the sum of the products of the three normalized values and their corresponding weights, with the unit of dimensionless value, ranging from 0 to 1, where 1 represents complete loss of metabolic function and 0 represents no impact on metabolic function. The output result is a two-dimensional array, where the first dimension represents time and the second dimension represents the depth interval. The processing method for measuring the loss of soil microbial activity based on the equivalent loss of metabolism estimation data and enzyme activity inhibition estimation data is completed based on exponentially weighted moving average calculation. First, the equivalent loss value of metabolism and the corresponding enzyme activity inhibition coefficient at the same time point are extracted. Before the combined processing, amplitude coordination processing is performed on the two input data. The method is to multiply the metabolic loss data by a coefficient of 1.2 to match the intensity of enzyme inhibition. After the coordination processing, an exponentially weighted function structure dominated by enzyme activity is constructed. The enzyme inhibition coefficient at the most recent time point is set as the initial weight value of 1. Pushing forward five days, the daily decay factor is set to 0.85. The weighted average of the metabolic loss value at the corresponding time point multiplied by its enzyme weight is the measured value of microbial activity loss, with the unit of dimensionless ratio value. The final output value falls between 0 and 1. The higher the value, the more severe the weakening degree of microbial activity function. The output format is a two-dimensional table sorted by date and depth, and each record contains the activity loss value and the corresponding time label.

[0043] Step S25 includes the following steps: Step S251: Conduct a mutation intensity time series analysis on the abnormal fluctuation state of soil pH to obtain pH mutation time series intensity data; Step S252: Estimate the increment of ion toxicity effect based on the pH mutation time series intensity data to obtain ion toxicity effect increment estimation data; Step S253: Evaluate the attenuation of soil oxygen exchange based on the soil hardening degree quantification data to obtain soil oxygen exchange attenuation data; Step S254: Conduct an analysis of the reduction in nutrient dissolution effectiveness based on the ion toxicity effect increment estimation data, soil oxygen exchange attenuation data, and microbial activity loss measurement data to obtain nutrient dissolution effectiveness reduction data; Step S255: Estimate the degree of root growth inhibition of the crop root system based on the nutrient dissolution effectiveness reduction data to obtain root growth inhibition gradient data.

[0044] In the embodiments of the present invention, in the process of performing mutation intensity time series analysis on the abnormal fluctuation state of soil pH, first, the obtained data of the abnormal fluctuation state of pH is reconstructed into a time series with a daily granularity to construct a one-dimensional time series array structure. The data length is not less than 90 days, and each data point is the average pH value of the corresponding day. Then, a first-order difference operation is performed on this array to calculate the pH difference between each day and the previous day, with the unit being the pH change amplitude. Next, a time period mutation index is constructed using a sliding window method, and the window size is set to 7 days. The difference between the maximum pH difference and the minimum pH difference within each sliding window is calculated and recorded as the mutation amplitude value of this window. Subsequently, the abnormal mutation sections are screened out through an extreme value detection algorithm (judgment based on the mean and standard deviation). The fluctuation rate of each day is recalculated in the mutation sections. Finally, the maximum fluctuation rate in each abnormal section is extracted as the mutation intensity, and after summarizing all the mutation intensities, a pH mutation time series intensity data table is constructed. This data table is indexed by date, with the mutation intensity value as the main value, and the unit is the daily pH change rate value. The data structure is a time series data frame structure.In the process of estimating the increment of ion toxicity effect based on the temporal intensity data of pH mutation, first, each record item of the mutation intensity temporal data is called and standardized using the z-score method, that is, each data value is subtracted by the full sample mean and then divided by the standard deviation. Then, the weighted processing is applied to each standardized mutation intensity value using the ion sensitivity factor, and the value of this sensitivity factor is obtained from the correspondence table of soil ion types and pH mutation sensitivity established through experiments in the early stage. For example, the sensitivity factor under the condition of chloride ion as the main ion is set to 1.3, under the condition of sodium ion is 1.6, and under the condition of ammonium ion is 1.1. After mixing the sensitivity factors proportionally according to the current ion structure ratio of the soil solution, the weighted processing is carried out on the standardized mutation intensity value to obtain the estimated value of the increment of ion toxicity effect for each time period. The unit of this estimated value is the relative toxicity change coefficient, and the output form is in the time series format. Each record item includes the date, mutation intensity value, mixed sensitivity factor value, and the final value of the increment of ion toxicity effect. In the process of evaluating the attenuation of soil oxygen exchange based on the quantified data of soil hardening degree, first, the hardening value matrix of the corresponding depth interval is extracted from the quantified data of hardening degree. The structure of this matrix is a two-dimensional data grid composed of the time axis and the depth axis. Subsequently, the hardening values are extracted in each depth layer according to the time series, and the discrete hardening values are converted into the continuous oxygen flux reduction coefficient using the difference approximation function, which is calibrated according to the compaction degree and oxygen flux reduction function established by soil mechanics experiments. In each depth layer, the decline amplitude of the oxygen exchange capacity is calculated according to the change rate of compaction degree. After outputting the oxygen flux reduction ratio, the reduction ratios of all depth layers are aggregated and weighted average processing is carried out. The weights are set according to the distribution ratio of the active layer of crop roots. For example, the weight from 0 to 15 cm is set to 0.6, from 15 to 30 cm is set to 0.3, and below 30 cm is set to 0.1. The finally calculated soil oxygen exchange attenuation value is expressed as the reduction ratio of the oxygen flux compared with the normal state, with the unit of percentage, and the structure is a time series array with one record item per day.

[0045] In the process of analyzing the reduction of nutrient dissolution effectiveness based on the estimated data of ion toxicity effect increment, the data of soil oxygen exchange attenuation, and the measured data of microbial activity loss, first, time alignment is performed on the three items of data to ensure that there is a corresponding data point for the three at the same time scale. Then, a weighted aggregation method based on conditional weight factor mapping is used for comprehensive processing. Specifically, first, range normalization is performed on each input data item to unify the dimensions, and then the influence weights of each factor are set according to the experimental data of soil nutrient dissolution influence sensitivity. The influence weight of ion toxicity is set to 0.4, the influence weight of oxygen exchange is set to 0.35, and the influence weight of microbial activity is set to 0.25. At each time point, the product of the three items of data and their weights is calculated and summed respectively to obtain the reduction value of nutrient dissolution effectiveness at this time point. This value is expressed as a ratio number between 0 and 1. The closer it is to 1, the more serious the weakening of the effective dissolution function. The output result is a continuous sequence structure indexed by time. Each record item includes the reduction ratio, the corresponding time label, and the records of the three original input values. In the process of estimating the degree of root growth inhibition of crop roots based on the data of nutrient dissolution effectiveness reduction, first, a time series of 15 consecutive days in the data of nutrient dissolution effectiveness reduction is extracted as the analysis window. In each analysis window, the maximum value, the minimum value, and the average value are extracted, and the standard deviation is calculated as the fluctuation index. Then, a growth inhibition threshold discrimination condition is set for the average reduction value within the window. When the average value is between 0.3 and 0.5, it is defined as mild inhibition; when it is between 0.5 and 0.7, it is defined as moderate inhibition; when it is greater than 0.7, it is defined as severe inhibition. Then, the window smoothing algorithm is used to slide the entire time series and record the inhibition level corresponding to each time period. After determining the level, each level is mapped to a quantitative gradient value through a level mapping function. Mild is set to 0.3, moderate is 0.6, and severe is 0.9. This gradient value is the estimated value of the degree of root growth inhibition. The output result is a two-dimensional time series array. The first dimension is the date, and the second dimension is the root growth inhibition gradient value, with the unit of dimensionless ratio value.

[0046] Step S254 includes the following steps: Perform an analysis on the probability of heavy metal valence migration for the estimated data of ion toxicity effect increment to obtain the data of heavy metal valence migration probability; Based on the data of heavy metal valence migration probability, perform an analysis of redox potential coupling on the data of soil oxygen exchange attenuation and the measured data of microbial activity loss to obtain the potential coupling loss coefficient; Perform an analysis on the degradation gradient of pore structure based on the data of soil oxygen exchange attenuation to obtain the data of pore structure degradation gradient; Based on the potential coupling loss coefficient and the data of pore structure degradation gradient, perform modeling on the attenuation of nutrient adsorption sites to obtain the data of adsorption site attenuation; Based on the adsorption site decay data, the nutrient dissolution effectiveness reduction analysis is carried out to obtain the nutrient dissolution effectiveness reduction data.

[0047] In the embodiments of the present invention, in the process of analyzing the heavy metal valence migration probability of the ion toxicity effect increment estimation data, first, the heavy metal ion concentration increment data with time tags is extracted from the ion toxicity effect increment estimation data, including the daily increment concentration values of common elements such as copper ions, lead ions, cadmium ions, zinc ions, etc., with the unit of milligrams per kilogram of soil. Then, based on the redox reaction paths of the common valence states of each heavy metal ion in the soil, the corresponding valence conversion path diagram is constructed. Each valence node in the path diagram is connected to its reachable conversion states. Then, based on the pre-set critical intervals of soil pH value and redox potential change, the environmental conditions at each time point are mapped and judged. The current pH value and redox potential value are substituted into the valence change interval judgment logic to judge the conversion possibility of each valence state in the current state. The migration probability coefficient is scored on a five-level gradient and normalized, and then the migration probability coefficient is multiplied by the current ion concentration increment value to reflect the current valence migration activity. Finally, the migration probability data of each heavy metal element in each valence state at each time point is output as a three-dimensional structure. The first dimension is time, the second dimension is the type of heavy metal element, and the third dimension is the valence state and its corresponding migration probability value. In the process of performing redox potential coupling analysis on the soil oxygen exchange decay data and the microbial activity loss measurement data according to the heavy metal valence migration probability data, first, the three types of data are aligned according to the time series, and a ternary coupling data structure is constructed. Each time point corresponds to a data item, which includes the mean value of the migration probabilities of various heavy metal valence states, the corresponding soil oxygen exchange decay value, and the microbial activity loss value at that moment. Then, at each time point, a weighted coupling function is applied to process the three items. The function is constructed using a weighted product structure. The coupling calculation process is to multiply the weighted mean value of the heavy metal migration probability by the oxygen exchange decay value and the microbial activity loss value respectively, and multiply by a preset coupling factor. The coupling factor is set according to the experimental data of the reaction sensitivity between heavy metals and microorganisms. Different metals correspond to different coefficients. Copper is set to 1.2, cadmium is set to 1.5, and lead is set to 1.4. Then, the coupling terms of multiple heavy metals are summed to obtain the potential coupling loss coefficient at this time point, which is output as a one-dimensional array, with the index being the time tag and the value being the coupling loss coefficient.

[0048] In the process of analyzing the pore structure degradation gradient based on the soil oxygen exchange attenuation data, first, extract the daily attenuation rate values from the oxygen exchange attenuation data, and establish a data mapping according to the functional relationship between porosity and aeration rate in soil physical experiments. This mapping table maps different degrees of aeration rate reduction to the corresponding degrees of pore structure change. After establishing the mapping relationship, perform mapping conversion on the daily aeration rate reduction values to obtain the pore structure degradation coefficient. Then, calculate the change trend of the degradation coefficient for every consecutive 10 days through the moving window algorithm, and take the difference between the maximum and minimum degradation coefficients of each window as the gradient amplitude to obtain the corresponding daily degradation gradient value. This gradient value is used to measure the rate and intensity of pore structure deterioration. The final result is output as an array in time series format, containing daily gradient values and corresponding time tags. The unit of the gradient value is a dimensionless ratio, and the structure is a one-dimensional array. In the process of modeling the attenuation of nutrient adsorption sites based on the potential coupling loss coefficient and pore structure degradation gradient data, first, perform normalization processing on the potential coupling loss coefficient and pore structure degradation gradient. The normalization method is range normalization, subtracting the minimum value from each value and then dividing by the range. Then, perform weighted average processing on the two normalized values at each time point. The weights are set as 0.6 for the potential coupling coefficient ratio and 0.4 for the pore structure gradient ratio to obtain the potential attenuation factor of the adsorption site at each time point. Subsequently, map this factor to the adsorption capacity loss table determined through experiments in the early stage. The table records the adsorption site loss rates corresponding to different attenuation factors. For example, when the factor is 0.7, the adsorption capacity drops to 60% of the original value, and when the factor is 0.9, the adsorption capacity drops to 40% of the original value. Take the mapped adsorption site loss rate as the output result. The structure is a one-dimensional time series, with the index being the time tag and the value being the adsorption site attenuation ratio. In the process of analyzing the reduction of nutrient dissolution effectiveness based on the adsorption site attenuation data, first, couple the adsorption site attenuation ratio with the nutrient concentration change curve, and perform a multiplication operation on the daily nutrient element release rate and adsorption site attenuation rate in the concentration change curve to obtain the daily actual soluble nutrient release value. Then, perform a ratio operation on this release value and the theoretical release value to obtain the daily nutrient dissolution effectiveness value. Then, perform a ratio conversion on this effectiveness value and the maximum effectiveness value within the entire cycle to obtain the current reduction degree, and finally form the nutrient dissolution effectiveness reduction data. This data is indexed by date, with the reduction ratio as the value, the unit being a dimensionless ratio, and the output structure is a time series array for subsequent evaluation operations of root absorption effectiveness.

[0049] Step S3 includes the following steps: Step S31: Normalize the root growth inhibition gradient data to obtain the normalized root growth inhibition gradient data; Step S32: Predict the root soil looseness requirement based on the normalized root growth inhibition gradient data to obtain the predicted root soil looseness requirement data; Step S33: Perform phased increment control of root system soil microorganisms based on the normalized data of the root system growth inhibition gradient and the predicted data of the root system soil looseness requirement, and generate phased increment control data of microorganisms; Step S34: Perform soil pH balance adaptation on the abnormal fluctuation state of soil pH according to the predicted data of the root system soil looseness requirement and the phased increment control data of microorganisms to obtain soil pH balance adaptation data.

[0050] As an example of the present invention, refer to Figure 3 As shown, in this example, the said step S3 includes: Step S31: Normalize the root system growth inhibition gradient data to obtain normalized root system growth inhibition gradient data; In the embodiment of the present invention, during the process of normalizing the root system growth inhibition gradient data, first extract the daily inhibition gradient values in the continuous time series from the root system growth inhibition gradient data. The inhibition gradient value is a dimensionless percentage, indicating the degree of root system growth inhibition. The original distribution of the numerical range is different, and there is a large gap between the maximum value and the minimum value, which affects subsequent processing. To unify the dimension and distribution range, the range normalization method is used to process all data points. Subtract the minimum value in the time series from each original inhibition gradient value, and then divide the obtained difference by the difference between the maximum value and the minimum value to obtain the normalized value. The distribution range of the normalized value is between zero and one. The data structure after normalization is the same one-dimensional array as the original data, only the value changes. The index is the time label, and the value is the normalized inhibition gradient data. This data serves as the input basis for subsequent prediction of the root system soil looseness requirement.

[0051] Step S32: Predict the root system soil looseness requirement based on the normalized root system growth inhibition gradient data to obtain predicted data of the root system soil looseness requirement; In the embodiment of the present invention, during the process of predicting the root system soil looseness requirement based on the normalized root system growth inhibition gradient data, first compare and fit the normalized daily root system inhibition gradient value with the actual soil looseness change value in the historical experimental data. The historical data records the soil looseness change law corresponding to the root system under different gradient values, and the unit is the millimeter value of the average pore spacing between soil particles. According to this law, a discrete mapping table is constructed to map the normalized gradient value to the soil looseness requirement interval. Subsequently, the sliding window prediction method is introduced. Each 7 days is a cycle window. Take the average value of the normalized gradient values within each cycle, and generate the looseness prediction value corresponding to the cycle according to the looseness mapping value corresponding to the average value. This value is the target loose demand threshold corresponding to the root system within the current cycle. The output structure is periodic time series data, and each cycle is a prediction data point, with the unit of millimeter.

[0052] Step S33: Perform phased incremental control of rhizosphere soil microorganisms based on the normalized data of the root growth inhibition gradient and the predicted data of the rhizosphere soil looseness requirement, and generate phased incremental control data for microorganisms. In the embodiment of the present invention, during the process of performing phased incremental control of rhizosphere soil microorganisms based on the normalized data of the root growth inhibition gradient and the predicted data of the rhizosphere soil looseness requirement, first, time alignment processing is performed on the two input data sources. The daily normalized inhibition gradient data is merged with the looseness prediction value within its corresponding prediction period to form a two-dimensional array. Then, based on the regression relationship in the experimental data of the microbial loosening effect, the quantitative relationship between the microbial quantity and the increase in soil looseness is extracted. At each time point, the required increase value of looseness is obtained by looking up the table, and then the incremental amplitude of the required microbial quantity is obtained by reverse lookup according to this value. The microbial species are limited to two types: nitrogen-fixing bacteria and phosphorus-solubilizing bacteria. An incremental conversion table of the microbial quantity and the increase ratio of looseness is established respectively. Among them, the nitrogen-fixing bacteria are set to increase 0.01 mm of looseness per unit quantity, and the phosphorus-solubilizing bacteria are set to increase 0.015 mm of looseness per unit quantity. Then, the target incremental value is divided by the conversion ratio of the two types of bacteria to obtain the unit quantity of the microbial population that needs to be controlled and input daily. Finally, the phased incremental control data for microorganisms is output, and the structure is a two-dimensional table. The first dimension is the time index, and the second dimension is the microbial species and the corresponding unit quantity value.

[0053] Step S34: Perform soil pH balance adaptation on the abnormal fluctuation state of soil pH according to the predicted data of the rhizosphere soil looseness requirement and the phased incremental control data for microorganisms, and obtain soil pH balance adaptation data.

[0054] In the embodiments of the present invention, in the process of adapting the abnormal fluctuation state of soil pH by using the predicted data of the looseness demand of the root system soil and the microbial phased increment control data, first, the daily predicted looseness value and the corresponding microbial control quantity value are extracted, and the synergistic influence relationship between the two on soil pH in each stage is analyzed. Based on the pH fluctuation historical database, the data of the change in the pH buffering interval caused by the microbial increment in the same type of greenhouse soil is extracted. Based on the measured regression relationship between the change in the unit quantity of the bacterial species and the pH change value, the current microbial input quantity is substituted at each time point, and the theoretical pH influence amplitude at this time point is calculated according to the corresponding regression coefficient. Then, according to the fluctuation amplitude value of the abnormal fluctuation of soil pH in the current time period, the adaptation correction value is calculated, and the correction value is the difference between the two. A positive value indicates an acidic tendency, and a negative value indicates an alkaline tendency. Subsequently, the correction value is coupled with the daily predicted looseness value for analysis, and the daily pH change trend is filtered for trend. The daily pH fluctuation balance adjustment trend value is output as a time series data structure, with the value being the pH correction value for daily balance adaptation, the unit being the pH numerical change amount, the index being the time tag, and the structure being a one-dimensional array, providing basic data support for the subsequent environmental control module to adjust the pH compensation strategy.

[0055] Step S4 includes the following steps: Step S41: Respectively perform logical learning on the predicted data of the looseness demand of the root system soil, the microbial phased increment control data, and the soil pH balance adaptation data, and respectively obtain the root system soil looseness logical learning data, the microbial phased increment control learning data, and the soil pH balance adaptation learning data; Step S42: Perform adaptive control training on the soil state variables according to the root system soil looseness logical learning data, the microbial phased increment control learning data, and the soil pH balance adaptation learning data, and obtain the soil state variable control training data; Step S43: Construct a soil environment regulation model based on the soil state variable control training data, and obtain the soil environment regulation model; Step S44: Send the soil environment regulation model to the cloud platform to execute the greenhouse environment control method.

[0056] In the embodiments of the present invention, during the process of logical learning of the predicted data of the looseness requirement of the root system soil, the data of the phased increment control of microorganisms, and the data of the adaptation of the soil pH balance, the independent logical relationships of the three data sets are extracted respectively. The predicted data of the looseness requirement of the root system soil contains the periodic looseness change values, which are input as time series. Five statistical features, namely the maximum, minimum, average, change range, and change trend of each period, are extracted in units of cycles to form a set of feature vectors; in the data of the phased increment control of microorganisms, the daily input amount of the bacterial community is used as the input, and statistical features such as the species density, the fluctuation range of the daily increment, and the gradient change ratio with the time series are extracted to construct the feature space of the species and quantity structure distribution of the bacterial community; in the data of the adaptation of the soil pH balance, the pH adjustment value is used as the input, and four types of indicators, namely the daily adjustment range, the difference between adjacent days, the cumulative correction amount of the cycle, and the amplitude offset, are extracted as the feature dimensions. The feature aggregation analysis algorithm is applied to the three types of data respectively. Through the neighborhood feature merging and the classification rule extraction, the logical rules of the feature sequences in the adjacent time periods are induced, and the output form is a feature relationship matrix, where each row represents a feature dimension and each column represents the result of the feature interaction rule. Finally, the logical learning data of the root system soil looseness, the learning data of the phased increment control of microorganisms, and the learning data of the adaptation of the soil pH balance are obtained. During the process of the adaptive control training of the soil state variables according to the logical learning data, the above three logical learning data are input into the control variable training structure in the form of a feature matrix. First, a state label column is introduced into the feature matrix, which corresponds to the classification annotation of the real state of the soil in each time period. The classification labels include seven typical soil states such as "too acidic", "insufficient looseness", "excessive bacterial community", "suitable looseness", etc. The discriminant analysis method is used to establish a logical mapping function. Based on the Bayesian discriminant logic, a conditional probability distribution model is constructed. The maximum likelihood method is relied on to train the logical correspondence relationship between the current input feature combination and the target state. A mapping result is generated in each period, and the error is minimized by comparing with the original state label. The training is repeated until the error converges. The result of the training is output as the control training data of the soil state variables. The data structure includes the mapping weight vector, the state switching threshold sequence, and the cycle rule index table, which provides a stable data basis for the subsequent construction of the control strategy.

[0057] In the process of constructing a soil environment regulation model based on soil state variable control training data, the state mapping weights and thresholds in the aforementioned training data are used as the core parameters of the model. The correspondence between the feature items in the feature matrix and the control variable output is fixed as a functional expression structure. In the model structure, the input layer is defined as a seven-dimensional input vector, corresponding to the main influencing indicators in three types of feature data respectively. The output layer is defined as six types of environmental control response types, including the regulation intensity of looseness, the regulation amount of the proportion of microbial population release, the release amount of pH neutralizer, the regulation coefficient of ventilation frequency, the regulation amount of water penetration factor, and the control time of buffer reaction delay. Three layers of control weight combination layers are set in the middle of the model to perform feature weighted fusion on each type of output response respectively. After the final model is constructed, it is stored in the form of a function structure and named the soil environment regulation model. The model file includes four types of structural contents: input structure, output control response function, weight set, and state mapping table, and can be directly called. When sending the soil environment regulation model to the cloud platform to execute the greenhouse environment control method, the model file is encapsulated in a standard format and converted into a general structured model file package. Necessary structure descriptions, input requirements, and output interface standards are embedded in the file. The file is connected to the cloud control system deployment platform through a remote data communication protocol. The platform automatically identifies the model structure and deploys it to the soil parameter adjustment module, sets the daily operation cycle and data input interface of the model, docks with the data acquisition channels such as soil temperature, humidity, pH, looseness, etc. in the real-time monitoring system in the greenhouse, imports the latest environmental monitoring data regularly every day, calls the regulation model to perform state identification and output adjustment suggestions, and finally the execution system of the greenhouse is controlled by the cloud platform to perform adjustment actions to achieve the whole process closed-loop control.

[0058] The present invention also provides a greenhouse environment control system for executing the greenhouse environment control method as described above. The greenhouse environment control system includes: A soil state acquisition module, configured to obtain the crop cultivation stage data in the greenhouse; based on the crop cultivation stage data, and through soil monitoring sensors, collect soil state data to obtain soil state monitoring filling data; A root growth inhibition degree estimation module, configured to identify the abnormal fluctuation state of soil pH in the soil state monitoring filling data to obtain the abnormal fluctuation state of soil pH; based on the abnormal fluctuation state of soil pH, estimate the root growth inhibition degree to obtain root growth inhibition gradient data; A pH balance adaptation module, configured to predict the root soil looseness requirement based on the root growth inhibition gradient data to obtain root soil looseness requirement prediction data; according to the root soil looseness requirement prediction data, perform soil pH balance adaptation on the abnormal fluctuation state of soil pH to obtain soil pH balance adaptation data; The soil environment regulation model construction module is used to perform adaptive control training on soil state variables according to the data adapted to soil pH balance and the predicted data of the root soil looseness requirement, so as to obtain the soil state variable control training data; construct a soil environment regulation model based on the soil state variable control training data to obtain the soil environment regulation model; and send the soil environment regulation model to the cloud platform to execute the greenhouse environment control method.

[0059] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A greenhouse environment control method, characterized in that, It includes the following steps: Step S1: Obtain the crop cultivation stage data in the greenhouse; based on the crop cultivation stage data, and through the soil monitoring sensor, collect the soil state data to obtain the soil state monitoring filling data; Step S2: Identify the abnormal fluctuation state of soil pH value for the soil state monitoring filling data to obtain the abnormal fluctuation state of soil pH value; Estimate the degree of root growth inhibition based on the abnormal fluctuation state of soil pH value to obtain the root growth inhibition gradient data; Step S3: Predict the demand for soil looseness of roots based on the root growth inhibition gradient data to obtain the predicted data of the demand for soil looseness of roots; adapt the abnormal fluctuation state of soil pH value according to the predicted data of the demand for soil looseness of roots to obtain the soil pH value balance adaptation data; Step S4: Conduct soil state variable adaptive control training according to the soil pH value balance adaptation data and the predicted data of the demand for soil looseness of roots to obtain the soil state variable control training data; construct a soil environment regulation model based on the soil state variable control training data to obtain the soil environment regulation model; send the soil environment regulation model to the cloud platform to execute the greenhouse environment control method.

2. The greenhouse environment control method according to claim 1, wherein, Step S1 includes the following steps: Step S11: Obtain the crop cultivation stage data in the greenhouse; Step S12: Evaluate the root state of crops between different cultivation stages for the crop cultivation stage data to obtain the root state of crops between different cultivation stages; Step S13: Based on the root state of crops, and through the soil monitoring sensor, collect the soil state data to obtain the soil state monitoring data; Step S14: Fill in the missing values for the soil state monitoring data to obtain the soil state monitoring filling data.

3. The greenhouse environment control method according to claim 2, wherein Step S2 includes the following steps: Step S21: Analyze the soil penetration structure for the soil state monitoring filling data to obtain the soil penetration structure state data; Step S22: Use the preset soil pH value abnormal identification model to identify the abnormal fluctuation state of soil pH value for the soil state monitoring filling data to obtain the abnormal fluctuation state of soil pH value; Step S23: Quantify the soil hardening degree for the soil penetration structure state data based on the abnormal fluctuation state of soil pH value to obtain the soil hardening degree quantification data; Step S24: Measure the loss of soil microbial activity according to the abnormal fluctuation state of soil pH value and the soil hardening degree quantification data to obtain the microbial activity loss measurement data; Step S25: Estimate the degree of root growth inhibition for the root state of crops based on the abnormal fluctuation state of soil pH value, the soil hardening degree quantification data, and the microbial activity loss measurement data to obtain the root growth inhibition gradient data.

4. The greenhouse environment control method according to claim 3, characterized in that, Step S23 includes the following steps: Step S231: Conduct multi-period clustering processing on the abnormal fluctuation state of soil pH value to obtain the pH value fluctuation distribution data; Step S232: Fit the pH value double-peak fluctuation intensity trend for the pH value fluctuation distribution data to obtain the pH value double-peak fluctuation intensity trend; Step S233: Quantify the loss of soil colloid cohesion for the soil infiltration structure state data according to the trend of the double-peak fluctuation intensity of soil pH, and obtain the soil colloid cohesion loss data; Step S234: Reconstruct the soil infiltration resistance equivalent based on the soil colloid cohesion loss data, and obtain the soil infiltration resistance equivalent reconstruction data; Step S235: Perform integral processing on the soil structure compactness of the soil infiltration resistance equivalent reconstruction data to obtain the soil structure compactness integral data; Step S236: Quantify the degree of soil hardening according to the soil infiltration resistance equivalent reconstruction data and the soil structure compactness integral data, and obtain the soil hardening degree quantification data.

5. The greenhouse environment control method according to claim 3, characterized in that, Step S24 includes the following steps: Step S241: Analyze the reduction of the air / water circulation space of soil porosity for the soil hardening degree quantification data to obtain the air / water circulation space reduction data; Step S242: Perform regression analysis on the increase in soil bulk density for the soil hardening degree quantification data to obtain the soil bulk density increase regression data; Step S243: Estimate the inhibition of enzyme activity based on the abnormal fluctuation state of soil pH to obtain the enzyme activity inhibition estimation data; Step S244: Estimate the equivalent loss of soil microbial metabolism based on the enzyme activity inhibition estimation data, the soil bulk density increase regression data, and the air / water circulation space reduction data to obtain the equivalent loss of metabolism estimation data; Step S245: Measure the loss of soil microbial activity according to the equivalent loss of metabolism estimation data and the enzyme activity inhibition estimation data to obtain the microbial activity loss measurement data.

6. The greenhouse environment control method according to claim 3, wherein, Step S25 includes the following steps: Step S251: Analyze the temporal sequence of the mutation intensity of the abnormal fluctuation state of soil pH to obtain the pH mutation temporal sequence intensity data; Step S252: Estimate the increase in ion toxicity effect based on the pH mutation temporal sequence intensity data to obtain the ion toxicity effect increase estimation data; Step S253: Evaluate the attenuation of soil oxygen exchange based on the soil hardening degree quantification data to obtain the soil oxygen exchange attenuation data; Step S254: Analyze the reduction of nutrient dissolution effectiveness according to the ion toxicity effect increase estimation data, the soil oxygen exchange attenuation data, and the microbial activity loss measurement data to obtain the nutrient dissolution effectiveness reduction data; Step S255: Estimate the degree of root growth inhibition of crop roots according to the nutrient dissolution effectiveness reduction data to obtain the root growth inhibition gradient data.

7. The greenhouse environment control method according to claim 6, characterized in that, Step S254 includes the following steps: Analyze the probability of heavy metal valence migration for the ion toxicity effect increase estimation data to obtain the heavy metal valence migration probability data; Perform redox potential coupling analysis on the soil oxygen exchange attenuation data and the microbial activity loss measurement data according to the heavy metal valence migration probability data to obtain the potential coupling loss coefficient; Analyze the degradation gradient of pore structure based on the soil oxygen exchange attenuation data to obtain the pore structure degradation gradient data; Model the attenuation of nutrient adsorption sites according to the potential coupling loss coefficient and the pore structure degradation gradient data to obtain the adsorption site attenuation data; Based on the adsorption site decay data, nutrient dissolution effectiveness reduction analysis is carried out to obtain nutrient dissolution effectiveness reduction data.

8. The greenhouse environment control method according to claim 1, wherein Step S3 includes the following steps: Step S31: Normalize the root growth inhibition gradient data to obtain the normalized root growth inhibition gradient data; Step S32: Predict the root soil looseness demand based on the normalized root growth inhibition gradient data to obtain the predicted root soil looseness demand data; Step S33: Control the phased increment of root soil microorganisms according to the normalized root growth inhibition gradient data and the predicted root soil looseness demand data to generate phased increment control data for microorganisms; Step S34: Adapt the abnormal fluctuation state of soil pH according to the predicted root soil looseness demand data and the phased increment control data for microorganisms to obtain the soil pH balance adaptation data.

9. The greenhouse environment control method according to claim 8, characterized in that Step S4 includes the following steps: Step S41: Conduct logical learning on the predicted root soil looseness demand data, the phased increment control data for microorganisms, and the soil pH balance adaptation data respectively to obtain the root soil looseness logical learning data, the phased increment control learning data for microorganisms, and the soil pH balance adaptation learning data; Step S42: Perform adaptive control training on soil state variables according to the root soil looseness logical learning data, the phased increment control learning data for microorganisms, and the soil pH balance adaptation learning data to obtain the soil state variable control training data; Step S43: Construct a soil environment regulation model based on the soil state variable control training data to obtain the soil environment regulation model; Step S44: Send the soil environment regulation model to the cloud platform to execute the greenhouse environment control method.

10. A greenhouse environment control system, characterized in that, For executing the greenhouse environment control method as described in claim 1, the greenhouse environment control system includes: A soil state acquisition module, configured to obtain the crop cultivation stage data in the greenhouse; based on the crop cultivation stage data, and collect soil state data through soil monitoring sensors to obtain the soil state monitoring filling data; A root growth inhibition degree estimation module, configured to identify the abnormal fluctuation state of soil pH in the soil state monitoring filling data to obtain the abnormal fluctuation state of soil pH; estimate the root growth inhibition degree based on the abnormal fluctuation state of soil pH to obtain the root growth inhibition gradient data; A pH balance adaptation module, configured to predict the root soil looseness demand based on the root growth inhibition gradient data to obtain the predicted root soil looseness demand data; adapt the abnormal fluctuation state of soil pH according to the predicted root soil looseness demand data to obtain the soil pH balance adaptation data; A soil environment regulation model construction module, configured to perform adaptive control training on soil state variables according to the soil pH balance adaptation data and the predicted root soil looseness demand data to obtain the soil state variable control training data; construct a soil environment regulation model based on the soil state variable control training data to obtain the soil environment regulation model; send the soil environment regulation model to the cloud platform to execute the greenhouse environment control method.

Images