A method for intelligent monitoring of environmental big data

Abstract

This invention relates to the field of environmental big data detection and intelligent control technology, and discloses an intelligent monitoring method for environmental big data. The method involves acquiring temperature and humidity measurements over a period of time for prediction, obtaining predicted temperature and humidity values; based on the given temperature and humidity values, obtaining temperature prediction error values ​​and humidity prediction error values ​​respectively; and using a dynamic decoupling strategy to obtain temperature decoupling error values ​​and humidity decoupling error values ​​respectively. Through tracking decoupling processing, a first environmental big data controller and a second environmental big data controller obtain actual temperature and humidity control values. These actual temperature and humidity control values ​​serve as control inputs for temperature and humidity control devices, respectively, to intelligently regulate the temperature and humidity of the environmental big data. Compared with existing technologies, this invention monitors environmental big data from three aspects: detection, decoupling, and control, improving the accuracy and robustness of environmental big data parameter monitoring.

Description

Technical Field

[0001] This invention relates to the field of environmental big data detection and intelligent control technology, specifically to an intelligent monitoring method for environmental big data. Background Technology

[0002] Precise monitoring of temperature and humidity using environmental big data is a key factor in ensuring high-quality production, maintaining the freshness of food products, and extending shelf life. This improves production efficiency and the freshness of food during transportation and storage, ensures consumers have access to fresh and safe food, increases the market value of products, and prevents food from being damaged by dryness or excessive moisture.

[0003] Precise temperature and humidity monitoring technology is crucial for maintaining product quality throughout the entire supply chain. However, most environmental big data control relies on traditional PID control methods, resulting in low accuracy in temperature and humidity control. These methods lack a dynamic coupling mechanism based on changes in temperature and humidity parameters, leading to low economic efficiency. Specifically:

[0004] 1) Traditional PID control: insufficient accuracy and difficulty in handling temperature and humidity coupling.

[0005] In production scenarios such as tobacco curing, where temperature and humidity control requirements are extremely high, most existing control systems are based on traditional PID control algorithms. However, the curing process exhibits significant time-varying characteristics and strong coupling between temperature and humidity—temperature changes cause humidity fluctuations, and humidity adjustments, in turn, affect temperature. Traditional PID systems struggle to achieve coordinated optimization of both. This limitation leads to unsatisfactory control accuracy, directly impacting product qualification and yield rates. Similar issues are prevalent in plant factory environmental control, where the strong coupling between temperature and humidity variables makes precise regulation difficult using traditional control methods.

[0006] 2) Sensor monitoring: limited accuracy and insufficient robustness

[0007] In the field of greenhouse environmental monitoring, traditional systems generally suffer from low monitoring accuracy and poor robustness due to limitations in sensor precision and insufficient data fusion capabilities. Existing research indicates that the application of IoT technology in environmental monitoring still faces challenges such as insufficient sensor precision and difficulty in ensuring data consistency. When data from a single sensor becomes distorted, the lack of effective cross-validation and anomaly removal mechanisms leads to a decline in overall monitoring reliability and makes it difficult to cope with environmental fluctuations under complex operating conditions. Summary of the Invention

[0008] Purpose of the invention: This invention provides an intelligent monitoring method for environmental big data, which enables the detection and management of environmental big data and precise control of temperature and humidity to meet the different environmental big data requirements of products at different stages, thereby improving the accuracy and robustness of environmental big data parameter monitoring.

[0009] Technical solution: This invention discloses an intelligent monitoring method for environmental big data, comprising:

[0010] Step 1: Obtain temperature and humidity measurements over a period of time, perform preprocessing, and obtain a temperature and humidity dataset;

[0011] Step 2: Based on the temperature and humidity dataset, predict the temperature and humidity using the corresponding PSO-optimized DRNN neural network models to obtain the predicted values ​​of temperature and humidity;

[0012] Step 3: Based on the given temperature and humidity values, obtain the temperature prediction error and humidity prediction error values ​​respectively. Then, based on the temperature prediction error and humidity prediction error values, obtain the temperature decoupling error and humidity decoupling error values ​​respectively through a dynamic temperature and humidity decoupling strategy.

[0013] Step 4: Input the temperature prediction error value and error change rate, and the temperature decoupling error value and error change rate to the first environmental big data controller, and obtain the actual temperature control quantity after tracking and decoupling processing; input the humidity prediction error value and error change rate, and the humidity decoupling error value and error change rate to the second environmental big data controller, and obtain the actual humidity control quantity after tracking and decoupling processing;

[0014] Step 5: The actual temperature control value and the actual humidity control value are used as the control inputs of the temperature regulation device and the humidity regulation device, respectively, to intelligently regulate the temperature and humidity based on big data of environmental data.

[0015] Furthermore, the first environmental big data controller and the second environmental big data controller have the same structure.

[0016] Furthermore, both the first environmental big data controller and the second environmental big data controller structure include a PSO-optimized DRNN neural network model-PID controller, a PSO-optimized TS fuzzy neural network model, and a pair of PSO-optimized DRNN neural network models.

[0017] Temperature prediction error and its rate of change, or humidity prediction error and its rate of change, serve as inputs to the PSO-optimized DRNN neural network model-PID controller. The PSO-optimized DRNN neural network model calculates the error gradient of each neuron based on error backpropagation, adjusts the network weights and thresholds according to the error gradient, and finally outputs the PID incremental parameters Kp, Ki, and Kd. The PID controller controls the temperature or humidity based on the PID incremental parameters Kp, Ki, and Kd, and outputs the temperature or humidity tracking control quantity.

[0018] Temperature decoupling error value and its rate of change, or humidity decoupling error value and its rate of change, are used as inputs to the PSO-optimized TS fuzzy neural network model. The output value of the PSO-optimized TS fuzzy neural network model is used as input to the PSO-optimized DRNN neural network model. The temperature and humidity errors caused by coupling are adjusted through recursive dynamic control of the combination of the PSO-optimized TS fuzzy neural network model and the PSO-optimized DRNN neural network model. The output values ​​of the PSO-optimized TS fuzzy neural network model and the PSO-optimized DRNN neural network model are dynamically weighted and used as the temperature or humidity decoupling control quantity.

[0019] The output value of the superimposed tracking control quantity and decoupling control quantity, after being recursively dynamically controlled by a PSO-optimized DRNN neural network model, is weighted and summed with the accumulated tracking control quantity and decoupling control quantity to obtain the final actual control quantity.

[0020] Furthermore, in step 3, the dynamic decoupling strategy for temperature and humidity is implemented using a PSO-optimized TS fuzzy neural network model. The PSO-optimized TS fuzzy neural network model optimizes the parameters of the TS fuzzy neural network model through a particle swarm optimization algorithm to obtain the center and width values ​​of the weights and membership functions of the TS fuzzy neural network model. The optimized TS fuzzy neural network model is then used to achieve dynamic decoupling of temperature and humidity prediction error values ​​through fuzzy inference and neural networks.

[0021] Furthermore, the error calculation function is defined as follows during the training process of the TS fuzzy neural network model:

[0022] ;

[0023] in, This is the expected output. It is the actual output, the weights of the neural network during the training process. The center of the membership function and width The parameters are corrected as follows:

[0024] ;

[0025] The corresponding change in parameter correction is:

[0026] ;

[0027] in, , , For the change in the corresponding parameter, , These represent the (k-1)th and (k-2)th times, respectively. , For learning rate, and and weight, and The value is:

[0028] ;

[0029] Where l is the disturbance factor, with a value of (0,1], e is the error, i.e. the temperature prediction error value and the humidity prediction error value, ec is the error change rate, i.e. the temperature prediction error change rate and the humidity prediction error change rate, E is the absolute value of the maximum error value, and EC is the absolute value of the maximum error change rate.

[0030] Furthermore, the PSO-optimized DRNN neural network model is improved by setting two delay unit groups on the basis of the PSO-optimized DRNN neural network model. Each delay unit group consists of 1 to n delay units connected in pairs. The parameters of the PSO-optimized DRNN neural network model are used as the input of delay unit group 1, and the output of each delay unit in delay unit group 1 is used as the input of the PSO-optimized DRNN neural network model. The output of the PSO-optimized DRNN neural network model is used as the input of delay unit group 2, and the output of each delay unit in delay unit group 2 is used as the input of the PSO-optimized DRNN neural network model. The output value of the PSO-optimized DRNN neural network model is used as the predicted value of the final input parameters.

[0031] Furthermore, the PSO-optimized DRNN neural network model is to optimize the parameters of the DRNN neural network model using the PSO algorithm. That is, the particle position with the best fitness output by the PSO algorithm is used as the optimal solution, and then it is mapped to the weights of the DRNN neural network model and trained. When the prediction error of the DRNN neural network model meets the condition or reaches the maximum number of iterations, the predicted value is output.

[0032] Furthermore, when the output values ​​of the PSO-optimized TS fuzzy neural network model and the PSO-optimized DRNN neural network model are dynamically weighted, the sum of their weight coefficients is 1.

[0033] Furthermore, when the output value of the superimposed tracking control quantity and decoupling control quantity is subjected to recursive dynamic control by a PSO-optimized DRNN neural network model, and then weighted and summed with the accumulated tracking control quantity and decoupling control quantity, the sum of the weight coefficients is 1.

[0034] Furthermore, temperature and humidity measurements over a period of time are obtained by using multiple temperature and humidity sensors placed in the environment.

[0035] Compared with the prior art, the present invention has the following obvious advantages:

[0036] This invention first uses a PSO-optimized DRNN neural network model to predict temperature and humidity in environmental big data. The predicted values ​​are compared with given values ​​to generate temperature and humidity errors and their rate of change. Based on these errors, decoupling predictions are performed to obtain temperature decoupling errors and humidity decoupling errors. For temperature and humidity, the actual control output of the environmental big data controller is achieved based on the temperature and humidity prediction errors and their rate of change, as well as the temperature and humidity decoupling errors and their rate of change. This invention addresses environmental big data detection, decoupling, and control from three aspects: dynamic, accurate, and robust perspectives. Based on the time-delay, time-lag, and coupling characteristics of environmental big data changes, it performs prediction, decoupling, and control in a dynamic, accurate, and robust manner. All three stages are dynamic recursive processes, consistent with the time-delay, nonlinear, and dynamic change mechanisms of environmental big data parameters, thus improving the scientific rigor, accuracy, dynamism, and robustness of environmental big data parameter detection, prediction, decoupling, and control.

[0037] II. This invention uses a PSO-optimized TS fuzzy neural network model to dynamically decouple temperature and humidity data from environmental big data. It leverages the robustness of fuzzy inference and the accuracy of the neural network within the PSO-optimized TS fuzzy neural network model to decouple temperature and humidity errors, improving the robustness, accuracy, and dynamism of temperature and humidity error decoupling. Furthermore, it enhances the dynamic performance, stability, and speed of the TS fuzzy neural network model for the current moment. , and When correcting parameter changes, different weights are applied to the changes at different time points to synthesize the data, reflecting the dynamics and stability of the neural network model's output. Simultaneously, the learning factor is dynamically adjusted based on the error and its changes. The sine function component of the dynamic adjustment reflects the stability of the nonlinear changes in the learning factor, while the exponential component reflects the rapid adjustment of the learning factor based on error changes and their rate of change. The TS fuzzy neural network model exhibits good learning speed, capable of tracking given values ​​in real time. The network parameters are synthesized with different weights based on changes at different time points, reflecting the dynamics and stability of the network output.

[0038] Third, this invention achieves intelligent adjustment of temperature and humidity errors caused by coupling by combining the PSO-optimized TS fuzzy neural network model and the PSO-optimized DRNN neural network model in the environmental big data controller through recursive dynamic control. The PSO-optimized DRNN neural network model of the environmental big data controller performs recursive dynamic control on the sum of parameter tracking control quantity and parameter decoupling control quantity. One controls from the dynamic changes of temperature and humidity parameters, and the other performs decoupling control based on coupling characteristics, thereby improving the accuracy and robustness of environmental temperature and humidity parameter control.

[0039] IV. The improvement of the PSO-optimized DRNN neural network model in the environmental big data controller designed in this invention includes a dynamic recursive layer and delay unit groups 1-2, which constitute the input and output delays. These respectively realize the dynamic delay prediction of the parameter decoupling control quantity and the dynamic delay prediction of the cumulative sum of the parameter tracking control quantity and the parameter decoupling control quantity. These, together with the real-time values ​​of the parameter decoupling control quantity and the cumulative sum of the parameter tracking control quantity and the parameter decoupling control quantity, constitute a two-level recursive control. That is, it considers the real-time value of the current control value and the delay influence of the past value of the control quantity on the current environmental big data. This control method is consistent with the dynamic delay characteristics and nonlinear changes of environmental temperature and humidity parameters, and improves the dynamism, accuracy and robustness of environmental big data control.

[0040] V. The present invention provides an improved PSO-optimized DRNN neural network model for predicting temperature and humidity. This model includes a dynamic recursive layer and two delay unit groups 1-2. The two delays are connected in series to form the input and output delays, which respectively realize the dynamic delay prediction of temperature and humidity of environmental big data. This prediction mechanism is consistent with the nonlinearity, time delay and dynamic change characteristics of environmental temperature and humidity parameters, thereby improving the dynamism, accuracy and robustness of temperature and humidity parameter prediction. Attached Figure Description

[0041] Figure 1 This invention relates to an environmental big data intelligent monitoring system;

[0042] Figure 2 The improved PSO optimized DRNN neural network model of this invention. Detailed Implementation

[0043] Combined with appendix Figure 1-2 The technical solution of this application will be further described below:

[0044] This invention discloses an intelligent monitoring method for environmental big data, see [link to relevant documentation]. Figure 1 Specifically, it includes the following steps:

[0045] Step 1: Preprocess the temperature and humidity measurements obtained over a period of time by using multiple temperature and humidity sensors set in the environment to obtain a temperature and humidity dataset.

[0046] Step 2: Based on the temperature and humidity dataset, predict the temperature and humidity using the corresponding PSO-optimized DRNN neural network model to obtain the predicted values ​​of temperature and humidity.

[0047] In this embodiment, for ease of description, PSO-optimized DRNN neural network model 1 is used for temperature prediction, and PSO-optimized DRNN neural network model 2 is used for humidity prediction. Furthermore, PSO-optimized DRNN neural network model 1 and PSO-optimized DRNN neural network model 2 have the same structure. Improvements are made to PSO-optimized DRNN neural network models 1 and 2. For specific improvements, please refer to [link to details]. Figure 2 :

[0048] An improved PSO-optimized DRNN neural network model is proposed by setting up two delay unit groups based on the PSO-optimized DRNN neural network model. Each delay unit group consists of 1 to n delay units connected in pairs. The parameters of the PSO-optimized DRNN neural network model are used as the input of delay unit group 1, and the output of each delay unit in delay unit group 1 is used as the input of the PSO-optimized DRNN neural network model. The output of the PSO-optimized DRNN neural network model is used as the input of delay unit group 2, and the output of each delay unit in delay unit group 2 is used as the input of the PSO-optimized DRNN neural network model. The output value of the PSO-optimized DRNN neural network model is used as the predicted value of the final input parameters.

[0049] Step 3: Based on the given temperature and humidity values, obtain the temperature prediction error and humidity prediction error values ​​respectively. Then, based on the temperature prediction error and humidity prediction error values, obtain the temperature decoupling error and humidity decoupling error values ​​respectively through a dynamic temperature and humidity decoupling strategy.

[0050] In this embodiment, the dynamic decoupling strategy for temperature and humidity is implemented using the PSO-optimized TS fuzzy neural network model 1. The PSO-optimized TS fuzzy neural network model 1 optimizes the parameters of the TS fuzzy neural network model through the particle swarm optimization algorithm to obtain the center and width values ​​of the weights and membership functions of the TS fuzzy neural network model. The optimized TS fuzzy neural network model is used to achieve dynamic decoupling of the temperature and humidity prediction error values ​​through fuzzy inference and neural network.

[0051] Step 4: Input the temperature prediction error value and error change rate, and the temperature decoupling error value and error change rate to the first environmental big data controller, and obtain the actual temperature control quantity after tracking and decoupling processing; input the humidity prediction error value and error change rate, and the humidity decoupling error value and error change rate to the second environmental big data controller, and obtain the actual humidity control quantity after tracking and decoupling processing.

[0052] In this embodiment, the first environmental big data controller and the second environmental big data controller have the same structure, both including a PSO-optimized DRNN neural network model-PID controller, a PSO-optimized TS fuzzy neural network model 2, and a pair of PSO-optimized DRNN neural network models (i.e., PSO-optimized DRNN neural network model 3 and PSO-optimized DRNN neural network model 4). See also Figure 1 The attached diagram shows the framework of the first environmental big data controller.

[0053] Temperature prediction error and its rate of change, or humidity prediction error and its rate of change, serve as inputs to the PSO-optimized DRNN neural network model-PID controller. The PSO-optimized DRNN neural network model calculates the error gradient of each neuron based on error backpropagation, adjusts the network weights and thresholds according to the error gradient, and finally outputs the PID incremental parameters Kp, Ki, and Kd. The PID controller controls the temperature or humidity based on the PID incremental parameters Kp, Ki, and Kd, and outputs the temperature or humidity tracking control quantity.

[0054] The temperature decoupling error value and its rate of change, or the humidity decoupling error value and its rate of change, are used as inputs to the PSO-optimized TS fuzzy neural network model 2. The output value of the PSO-optimized TS fuzzy neural network model 2 is used as input to the PSO-optimized DRNN neural network model 3. The combined recursive dynamic control of the PSO-optimized TS fuzzy neural network model 2 and the PSO-optimized DRNN neural network model 3 adjusts the temperature and humidity errors caused by coupling. The output values ​​of the PSO-optimized TS fuzzy neural network model 2 and the PSO-optimized DRNN neural network model 3 are dynamically weighted and used as the temperature or humidity decoupling control variable. During weighting, the sum of the weighting coefficients is 1.

[0055] The output value after superimposing the tracking control quantity and the decoupling control quantity and then passing it through a PSO-optimized DRNN neural network model 4 recursively dynamic control is weighted and summed with the accumulated tracking control quantity and the decoupling control quantity to obtain the final actual control quantity. Here, the sum of the weight coefficients is 1.

[0056] The structure of PSO-optimized TS fuzzy neural network model 2 is the same as that of PSO-optimized TS fuzzy neural network model 1 in step 3.

[0057] Step 5: The actual temperature control value and the actual humidity control value are used as the control inputs of the temperature regulation device and the humidity regulation device, respectively, to intelligently regulate the temperature and humidity based on big data of environmental data.

[0058] The model structure design is as follows:

[0059] I. PSO Optimization TS Fuzzy Neural Network Model Design

[0060] The TS fuzzy neural network model combines the logical reasoning ability of fuzzy systems with the powerful self-learning ability of neural networks. It can also adaptively process fuzzy information, exhibiting strong fuzzy reasoning and parallel processing capabilities, as well as anti-interference and fault tolerance. Its data processing capabilities are superior to simple neural network models such as BP and RBF. The design process of the TS fuzzy neural network model is as follows:

[0061] 1. Front-end network:

[0062] 1) Input layer, i.e., the input layer in the network. , In the formula, n is the input node.

[0063] 2) The fuzzification layer uses a membership function to calculate the membership degree of each input variable. :

[0064]

[0065] In the formula, The center of the membership function, is the width of the membership function, m is the input parameter, and n is the number of fuzzy segments for each input parameter.

[0066] 3) Fuzzy rule calculation layer: The fitness of each rule is calculated using a fuzzy multiplication formula. :

[0067]

[0068] 4) Output layer: Defuzzify the output of the preceding network and calculate the weights of each attribute. :

[0069]

[0070] 2. After-process network:

[0071] 1) Input layer, i.e., the input layer in the network. Its function is to pass input variables to the fully connected layer.

[0072] 2) The fully connected layer receives the input variables from the input layer and learns the input variables using the following formula to obtain the output result. , , For subsequent network weights.

[0073] 3) Output layer, which receives weights from the preceding network. and the output results of the consequent network The output of the TS fuzzy neural network model is calculated using the following formula: .

[0074] The training process of the TS fuzzy neural network model is a process of continuously correcting the weights and thresholds based on computational errors. Based on the working principle of the TS fuzzy neural network model, it is proposed to... The membership degree serves as the threshold of the neural network, acting as the weights. Therefore, the parameter that needs to be adjusted is the weight. and the center of the membership function and width In the initial training phase, the parameters requiring correction and the error requirements must first be initialized. Then, the calculated error is compared with the set error requirements. If the error does not meet the requirements, the center of the weights and membership functions needs to be adjusted. and width If the error meets the requirements, then parameter correction is not necessary, and the neural network model training is complete.

[0075] The structure of PSO-optimized TS fuzzy neural network model 2 is the same as that of PSO-optimized TS fuzzy neural network model 1 in step 3. That is, the parameters of the TS fuzzy neural network model are optimized by particle swarm optimization to obtain the center and width values ​​of the weights and membership functions of the TS fuzzy neural network model, thereby improving the accuracy of the output of the TS fuzzy neural network model.

[0076] The algorithm steps for PSO to optimize the TS fuzzy neural network model are as follows:

[0077] (1) Initialization; initialization of particle position and velocity, as well as center and width of weights and membership functions.

[0078] (2) Modeling: Based on PSO, a TS fuzzy neural network model is established, and the output error and training error are calculated.

[0079] (3) Update: Update the corresponding weights, membership function center and width, velocity and position according to the particle swarm algorithm.

[0080] (4) Judgment: If PSO reaches the maximum number of iterations, stop the iteration and input the obtained optimal solution into the TS fuzzy neural network model to calculate the output; otherwise, go to step (3).

[0081] (5) Verification: If the number of iterations of the TS fuzzy neural network model reaches the maximum or the error reaches the target value, then stop the iteration; otherwise, go to step (2).

[0082] Furthermore, the error calculation function is defined as follows during the training of the TS fuzzy neural network model:

[0083] ;

[0084] in, This is the expected output. It is the actual output, the weights of the neural network during the training process. The center of the membership function and width The parameters are corrected as follows:

[0085] ;

[0086] The corresponding change in parameter correction is:

[0087] ;

[0088] in, , , For the change in the corresponding parameter, , These represent the (k-1)th and (k-2)th times, respectively. , For learning rate, and and weight, and The value is:

[0089] ;

[0090] Where l is the disturbance factor, with a value of (0,1], e is the error, i.e. the temperature prediction error value and the humidity prediction error value, ec is the error change rate, i.e. the temperature prediction error change rate and the humidity prediction error change rate, E is the absolute value of the maximum error value, and EC is the absolute value of the maximum error change rate.

[0091] A parameter model for environmental big data was established using a TS fuzzy neural network model. The PSO algorithm was then used to optimize the weights and the center and width of the membership functions of the TS fuzzy neural network model. Simulation experiments were conducted using agricultural production environmental data. After PSO optimization, the average relative error of the TS fuzzy neural network model in predicting environmental big data parameters was minimized to 0.0048. Comparison with the prediction results of the TS fuzzy neural network model showed that the agricultural production prediction model established using the PSO-optimized TS fuzzy neural network model had higher accuracy.

[0092] II. PSO-optimized DRNN neural network model design

[0093] The DRNN neural network model is a diagonal regression neural network. Its structure consists of an input layer, hidden layers, and an output layer. The input layer is the same as a static network, using a linear function as the action function. The hidden layer contains N neurons, each with a self-feedback regressor, using the sigmoid function as the action function. The input to each neuron in the input layer is I(k), the regressor in the regression layer is X(k), the sum of the inputs to the regression neurons is S(k), and the output layer is O(k).

[0094]

[0095] in, For the i-th input of the input layer, For the output of the output layer, Let be the sum of the inputs to the j-th neuron in the regression layer. The output of the j-th neuron in the regression layer. For the weights of the input layer, and The weights for the output layer and the regression layer are respectively. For the activation function of the regression layer, this patent uses the double S-function.

[0096] The gradient descent method used for weight adjustment between layers in the DRNN neural network model has drawbacks such as slow convergence speed, low accuracy, and easy getting trapped in local minima. To overcome these drawbacks, an adaptive momentum term is introduced into the weight update algorithm. The weight learning law considers both the historical information of weight changes and the current gradient information. When the historical change direction of the weights is in the same direction as the negative gradient direction, the convergence speed is accelerated.

[0097] In the process of optimizing the DRNN neural network model using the PSO algorithm, the particle position with the best fitness is first used as the optimal solution, and then mapped to the weights of the DRNN neural network model and trained. When the prediction error of the DRNN neural network model meets the conditions or reaches the maximum number of iterations, the predicted value can be output.

[0098] The PSO-optimized DRNN neural network model optimizes the weight parameters of the network model. This requires dividing the environmental big data into training and testing sets. The training set is then used to train and improve the DRNN neural network model, resulting in predictions of environmental big data parameters. The average relative error of the PSO-optimized DRNN neural network model in predicting environmental big data parameters is as low as 0.0045. Comparison with the prediction results of the standard DRNN neural network model shows that the agricultural production prediction model established by the PSO-optimized DRNN neural network model has higher accuracy.

[0099] III. PSO Optimization DRNN Neural Network Model - PID Controller Design

[0100] The PSO-optimized DRNN neural network model continuously learns from samples to optimize and correct neurons and their interconnected weights, further obtaining the functional mapping relationship between sample data input and output. In environmental big data temperature control, the temperature error and error change rate are input into the PSO-optimized DRNN neural network model. The error gradient of each neuron is calculated using error backpropagation. Based on the error gradient, the network weights and thresholds are adjusted, outputting the incremental PID parameters Kp, Ki, and Kd. The output of the incremental PID controller is as follows:

[0101]

[0102] A PSO-optimized DRNN neural network model is introduced into the PID control algorithm to control environmental big data parameters. The PSO-optimized DRNN neural network model is trained multiple times with the following training parameters set: learning rate of 0.01, momentum factor of 0.7, training times of 400, and target error of 0.001. When the deviation is large and the rate of change of the deviation is rapid, the PSO-optimized DRNN neural network model may increase the values ​​of Kp and Kd after training, enhancing the controller's ability to respond quickly to and suppress deviations. At the same time, the Ki value is reasonably adjusted according to the accumulation of deviations to avoid integral saturation and ensure stable system operation and high-precision control.

[0103] Experimental results show that the overshoot of greenhouse ambient temperature controlled by traditional PID control is about 13%, and the settling time is as long as 0.3 hours. In contrast, the overshoot of the PSO-optimized DRNN neural network model-PID controller is controlled within 5%, and the settling time is shortened to 0.15 hours. The PSO-optimized DRNN neural network model-PID controller has good stability and high accuracy in temperature control of environmental big data, which is significantly better than traditional PID control. The results show that this method can effectively overcome the defects of traditional PID control. The dynamic and steady-state performance of temperature control of environmental big data is excellent in key indicators such as overshoot, settling time and anti-interference ability.

[0104] PSO optimizes the DRNN neural network model - compared with the PID controller, it has better robustness, shorter settling time and better accuracy.

[0105] IV. Experimental Verification:

[0106] To verify the practicality of the designed intelligent environmental big data monitoring method, an environmental big data regulation device was built. Using the intelligent environmental big data monitoring method of this patent, the PSO-optimized DRNN neural network model-PID controller and the PSO-optimized TS fuzzy neural network model 2 in the environmental big data controller were designated as A; the improved PSO-optimized DRNN neural network model 3 was designated as B; and the improved PSO-optimized DRNN neural network model 4 was designated as C. This system was applied to a tomato greenhouse in Huai'an City to achieve intelligent regulation of environmental temperature and humidity. The greenhouse was 40m long and 18m wide. During the experiment, the system first acquired the current temperature and humidity values ​​at each detection node through temperature and humidity sensors. Based on the aforementioned intelligent environmental big data monitoring method, the optimal actual control values ​​for temperature and humidity were provided to regulate the temperature and humidity regulation device. The most suitable relative humidity for tomato growth was 45%–50% RH and the optimal temperature was 20–30℃. The experiment lasted from 7:00 AM to 4:00 PM, with data collected every hour. The statistical results of temperature and humidity control are shown in Tables 1 and 2.

[0107] Table 1 Statistical Table of Temperature Control Results

[0108] Traditional PID control ±2.16 7.2 91 Difference 1（A） ±1.86 6.2 80 generally 2(A+B) ±1.26 4.2 73 better 3 (A+B+C) ±0.45 1.5 61 very good

[0109] For the four schemes mentioned above, the same experimental data, temperature and humidity data, and data acquisition time were used to collect and control temperature error, steady-state error, and corresponding time. During the experiment, the temperature inside the greenhouse was maintained between 19.82 and 29.58℃. At different sampling times, the error between the actual temperature and the corresponding optimal temperature was within ±0.45℃, the steady-state time was about 1 hour, and the maximum steady-state error was only 1.50%. The temperature control results in the table show that after adding the improved PSO optimized DRNN neural network model 3 (B and C) to the environmental big data controller, the accuracy, speed, steady-state error, and stability of temperature regulation improved significantly.

[0110] Table 2 Statistical Table of Humidity Control Results

[0111] Traditional PID control ±2.75 5.50 93 Difference 1（A） ±1.76 3.50 82 generally 2(A+B) ±1.15 2.30 72 better 3 (A+B+C) ±0.50 1.00 60 very good

[0112] For the four schemes mentioned above, the same experimental data, temperature and humidity data, and data collection time were used to collect and control humidity error, steady-state error, and corresponding time. Regarding humidity control, the relative humidity was set at 45%–50% RH, with data collected every hour. The humidity parameters in the greenhouse remained within the optimal range of 44.8%–50.3% RH, with the actual humidity error between the actual and optimal humidity being within ±0.5 RH. The steady-state time was approximately one hour, and the maximum steady-state error was only 1.00%. The humidity control results in the table show that after adding the improved PSO optimized DRNN neural network model 3 (B and C) to the environmental big data controller, the accuracy, speed, steady-state error, and stability of humidity control improved significantly.

[0113] When using traditional PID control for temperature and humidity in an environmental big data controller, the steady-state time is about 1.5 hours, with steady-state errors of 7.20% and 5.50% for temperature and humidity, respectively. The experimental results show that the intelligent environmental big data monitoring method of this patent has high accuracy, small steady-state error in temperature and humidity control, short adjustment time, and good practical application effect.

[0114] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent transformations or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent monitoring of environmental big data, characterized in that, include: Step 1: Obtain temperature and humidity measurements over a period of time, perform preprocessing, and obtain a temperature and humidity dataset; Step 2: Based on the temperature and humidity dataset, predict the temperature and humidity using the corresponding PSO-optimized DRNN neural network model to obtain the predicted values ​​of temperature and humidity; Step 3: Based on the given temperature and humidity values, obtain the temperature prediction error and humidity prediction error values ​​respectively. Then, based on the temperature prediction error and humidity prediction error values, obtain the temperature decoupling error and humidity decoupling error values ​​respectively through a dynamic temperature and humidity decoupling strategy. Step 4: Input the temperature prediction error value and error change rate, and the temperature decoupling error value and error change rate to the first environmental big data controller, and obtain the actual temperature control quantity after tracking and decoupling processing; input the humidity prediction error value and error change rate, and the humidity decoupling error value and error change rate to the second environmental big data controller, and obtain the actual humidity control quantity after tracking and decoupling processing; Step 5: The actual temperature control value and the actual humidity control value are used as the control inputs of the temperature regulation device and the humidity regulation device, respectively, to intelligently regulate the temperature and humidity based on the big data of environmental data.

2. The intelligent environmental big data monitoring method according to claim 1, characterized in that, The first environmental big data controller and the second environmental big data controller have the same structure.

3. The intelligent environmental big data monitoring method according to claim 2, characterized in that, Both the first and second environment big data controllers include a PSO-optimized DRNN neural network model-PID controller, a PSO-optimized TS fuzzy neural network model, and a pair of PSO-optimized DRNN neural network models. Temperature prediction error and its rate of change, or humidity prediction error and its rate of change, serve as inputs to the PSO-optimized DRNN neural network model-PID controller. The PSO-optimized DRNN neural network model calculates the error gradient of each neuron based on error backpropagation, adjusts the network weights and thresholds according to the error gradient, and finally outputs the PID incremental parameters Kp, Ki, and Kd. The PID controller controls the temperature or humidity based on the PID incremental parameters Kp, Ki, and Kd, and outputs the temperature or humidity tracking control quantity. Temperature decoupling error value and its rate of change, or humidity decoupling error value and its rate of change, are used as inputs to the PSO-optimized TS fuzzy neural network model. The output value of the PSO-optimized TS fuzzy neural network model is used as input to the PSO-optimized DRNN neural network model. The temperature and humidity errors caused by coupling are adjusted through recursive dynamic control of the combination of the PSO-optimized TS fuzzy neural network model and the PSO-optimized DRNN neural network model. The output values ​​of the PSO-optimized TS fuzzy neural network model and the PSO-optimized DRNN neural network model are dynamically weighted and used as the temperature or humidity decoupling control quantity. The output value of the superimposed tracking control quantity and decoupling control quantity, after being recursively dynamically controlled by a PSO-optimized DRNN neural network model, is weighted and summed with the accumulated tracking control quantity and decoupling control quantity to obtain the final actual control quantity.

4. The intelligent environmental big data monitoring method according to any one of claims 1 to 3, characterized in that, In step 3, the dynamic decoupling strategy for temperature and humidity is implemented using a PSO-optimized TS fuzzy neural network model. The PSO-optimized TS fuzzy neural network model optimizes the parameters of the TS fuzzy neural network model through a particle swarm optimization algorithm to obtain the center and width values ​​of the weights and membership functions of the TS fuzzy neural network model. The optimized TS fuzzy neural network model is then used to achieve dynamic decoupling of temperature and humidity prediction error values ​​through fuzzy inference and neural networks.

5. The intelligent environmental big data monitoring method according to claim 4, characterized in that, The error calculation function is defined as follows during the training of the TS fuzzy neural network model: ； in, This is the expected output. It is the actual output, the weights of the neural network during the training process. The center of the membership function and width The parameters are corrected as follows: ； The corresponding change in parameter correction is: ； in, , , This represents the change in the corresponding parameter. , These represent the (k-1)th and (k-2)th times, respectively. , For learning rate, and and weight, and The value is: ； Where l is the disturbance factor, with a value of (0,1], e is the error, i.e. the temperature prediction error or humidity prediction error, ec is the error change rate, i.e. the temperature prediction error change rate or humidity prediction error change rate, E is the absolute value of the maximum error, and EC is the absolute value of the maximum error change rate.

6. The intelligent environmental big data monitoring method according to any one of claims 1 to 3, characterized in that, Furthermore, the PSO-optimized DRNN neural network model is improved by setting two delay unit groups on the basis of the PSO-optimized DRNN neural network model. Each delay unit group consists of 1 to n delay units connected in pairs. The parameters of the PSO-optimized DRNN neural network model are used as the input of delay unit group 1, and the output of each delay unit in delay unit group 1 is used as the input of the PSO-optimized DRNN neural network model. The output of the PSO-optimized DRNN neural network model is used as the input of delay unit group 2, and the output of each delay unit in delay unit group 2 is used as the input of the PSO-optimized DRNN neural network model. The output value of the PSO-optimized DRNN neural network model is used as the predicted value of the final input parameters.

7. The intelligent environmental big data monitoring method according to claim 6, characterized in that, The PSO-optimized DRNN neural network model is a process of optimizing the parameters of the DRNN neural network model using the PSO algorithm. Specifically, the optimal solution is obtained by using the particle position with the best fitness output by the PSO algorithm, which is then mapped to the weights of the DRNN neural network model and trained. When the prediction error of the DRNN neural network model meets the conditions or reaches the maximum number of iterations, the predicted value is output.

8. The intelligent environmental big data monitoring method according to claim 3, characterized in that, When the output values ​​of the PSO-optimized TS fuzzy neural network model and the PSO-optimized DRNN neural network model are dynamically weighted, their weight coefficients sum to 1.

9. The intelligent monitoring method for environmental big data according to claim 3, characterized in that, The output value after the superposition of the tracking control quantity and the decoupling control quantity and then the recursive dynamic control of the PSO-optimized DRNN neural network model is weighted and summed with the cumulative sum of the tracking control quantity and the decoupling control quantity, and the weight coefficients sum to 1.

10. The intelligent environmental big data monitoring method according to claim 1, characterized in that, Temperature and humidity measurements over a period of time are obtained using multiple temperature and humidity sensors located in the environment.

Images