Pesticide production control method based on cloud network fusion technology

Abstract

The invention relates to the technical field of pesticide production, and discloses a pesticide production control method based on a cloud network fusion technology. According to the method, pesticide production data including raw materials, environmental parameters, production area images and equipment operation states are collected in real time through data collection equipment, and the data are transmitted to a cloud platform. The cloud platform carries out cleaning, normalization, dimensionality reduction, abnormal value detection and other processing on the data, and carries out data analysis by adopting a deep learning algorithm, including personnel detection, production monitoring and equipment fault monitoring. And according to an analysis result, the cloud platform outputs a pesticide production control strategy and automatically regulates and controls production equipment, such as adjusting equipment parameters, starting or stopping the equipment and the like. The pesticide production efficiency and quality can be improved, the production cost is reduced, and intelligent management and control are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of pesticide production, and in particular to a pesticide production management and control method based on cloud-network fusion technology. Background Art

[0002] Traditional pesticide production control mainly relies on manual monitoring and manual adjustment of production equipment. This method is not only inefficient, but also prone to human errors, affecting the production quality and safety of pesticides. With the continuous development of cloud-network integration technology, more and more industries are beginning to explore the application of cloud technology in production control. However, for the specific field of pesticide production, how to effectively use cloud-network integration technology to achieve real-time collection, transmission, processing and analysis of production data, as well as automatic adjustment of production equipment based on analysis results, is still a problem that needs to be solved. Summary of the invention

[0003] In view of the above-mentioned technical deficiencies, the purpose of the present invention is to provide a pesticide production control method based on cloud-network fusion technology to solve the problem of low efficiency of pesticide production control in the prior art.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a pesticide production control method based on cloud-network fusion technology, the method comprising: Step S100: using data collection equipment to collect pesticide production data in real time; Step S200: Transmitting pesticide production data to the cloud platform; Step S300: the cloud platform performs data processing and data analysis on the received pesticide production data; Step S400: outputting a pesticide production control strategy based on the data analysis results; Step S500: Automatically control pesticide production equipment according to the pesticide production control strategy.

[0005] Preferably, in a possible implementation manner of the first aspect, the pesticide production data includes pesticide production raw materials, pesticide production environmental parameters, pesticide production area images and operating status of pesticide production equipment.

[0006] Preferably, in a possible implementation manner of the first aspect, the data processing includes data cleaning, data normalization, data dimension reduction and outlier detection.

[0007] Preferably, in a possible implementation of the first aspect, the data analysis includes using a personnel detection model to detect the presence of personnel in the pesticide production area; using a pesticide production monitoring model to monitor the pesticide production process; and using a pesticide production equipment monitoring model to monitor failures of pesticide production equipment.

[0008] Preferably, in a possible implementation of the first aspect, the personnel detection model is trained using an improved Yolo v5 algorithm; the pesticide production monitoring model is trained using a deep learning algorithm based on a graph neural network; and the pesticide production equipment monitoring model is trained using a deep learning algorithm based on LSTM.

[0009] Preferably, in a possible implementation of the first aspect, the improved Yolo v5 algorithm specifically includes: a first input layer, used to receive image data of a pesticide production area; a feature extraction layer, which extracts image feature information through EfficientViT; a feature fusion layer, which fuses image feature information of different scales; a prediction layer, which predicts the confidence and position of a person based on the fused image feature information; and a first output layer, which is used to output the final detection result of the person detection model.

[0010] Preferably, in a possible implementation of the first aspect, the deep learning algorithm based on the graph neural network specifically includes: a second input layer, used to receive the input data required by the pesticide production monitoring model; a graph construction layer, used to construct graph structure data according to the input data; a graph convolution layer, used to perform convolution operations on the graph structure data to extract node features; a second fully connected layer, used to map the features extracted by the graph convolution layer to output categories; a second output layer, used to output the monitoring results of the pesticide production process.

[0011] Preferably, in a possible implementation of the first aspect, the LSTM-based deep learning algorithm specifically includes: a third input layer, used to receive the input data required by the pesticide production equipment monitoring model; an LSTM layer, used to perform sequence modeling on the input data and extract time series features; a second fully connected layer, used to map the features extracted by the LSTM layer to output categories; and a third output layer, used to output the fault prediction results of the pesticide production equipment.

[0012] Preferably, in a possible implementation of the first aspect, the pesticide production control strategy includes: When the personnel detection model detects the presence of personnel in the pesticide production area, the presence of personnel is recorded; When the pesticide production monitoring model detects abnormal parameters in the pesticide production process, it outputs a method for adjusting the pesticide production equipment; When the pesticide production equipment monitoring model detects that the pesticide production equipment has a fault, it outputs a fault handling method.

[0013] Preferably, in a possible implementation manner of the first aspect, step S500 specifically includes: According to the pesticide production control strategy, control instructions are sent to pesticide production equipment through the cloud platform; The pesticide production equipment receives and analyzes the control instructions, automatically adjusts the parameters of the pesticide production equipment, and starts or stops the pesticide production equipment according to the contents of the instructions; The cloud platform continuously monitors the performance of pesticide production equipment and real-time changes in production data.

[0014] The beneficial effects of the present invention are: by using data acquisition equipment to collect pesticide production data in real time, and transmitting it to the cloud platform for data processing and analysis, it is possible to promptly discover abnormal conditions in the pesticide production process and output corresponding production control strategies. According to these strategies, the cloud platform can automatically control the pesticide production equipment, thereby avoiding the shortcomings of manual monitoring and manual adjustment. In addition, the present invention also uses a variety of deep learning algorithms for data analysis, which improves the accuracy and efficiency of the analysis. Through the implementation of the present invention, pesticide production enterprises can realize intelligent control of the production process, improve production efficiency, reduce production costs, and ensure the production quality and safety of pesticides. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 A flow chart of a pesticide production control method based on cloud-network fusion technology is provided for this application. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] Embodiment 1: Figure 1 As shown, the present invention provides a pesticide production control method based on cloud network fusion technology, the method comprising: Step S100: using data collection equipment to collect pesticide production data in real time.

[0019] Specifically, the data acquisition equipment is connected to the cloud platform via a wireless network, and the data acquisition equipment includes industrial cameras, temperature sensors, humidity sensors, gas sensors, liquid level sensors, flow sensors, pressure sensors, and vibration sensors.

[0020] Industrial cameras are used to capture images of pesticide production areas and monitor whether there are people in the pesticide production area; temperature sensors are used to monitor the temperature in the pesticide production environment to ensure that production is carried out under appropriate temperature conditions; humidity sensors are used to monitor the humidity of the production environment to prevent excessively high or low humidity from having an adverse effect on pesticide production; gas sensors are used to detect harmful gases (including ammonia, sulfur dioxide, etc.) in the production environment to ensure the safety of the production environment; liquid level sensors are used to monitor the liquid level in the raw material tank or reactor to ensure accurate addition of raw materials and smooth reaction; flow sensors are used to monitor the flow of liquids or gases to ensure material ratios and flow control during the production process; pressure sensors are used to monitor the pressure of production equipment to prevent equipment from being damaged due to excessive pressure; vibration sensors are used to monitor the vibration of equipment, detect abnormal conditions of equipment in a timely manner, and avoid failures.

[0021] Pesticide production data includes pesticide production raw materials, pesticide production environmental parameters, pesticide production area images, and the operating status of pesticide production equipment. The data of pesticide production raw materials includes detailed information such as the type, quantity, and quality of raw materials; pesticide production environmental parameters cover parameters such as temperature, humidity, and harmful gas concentration collected by temperature sensors, humidity sensors, gas sensors, and other equipment, reflecting the environmental conditions during the pesticide production process; pesticide production area images are real-time images captured by industrial cameras and are used to monitor personnel activities in the production area; pesticide production equipment operating status data includes status information such as liquid level, flow, pressure, and vibration collected by liquid level sensors, flow sensors, pressure sensors, and vibration sensors.

[0022] In this embodiment, the industrial camera adopts Basler acA1600-60gc, the temperature sensor adopts Omega HH506R-2, the humidity sensor adopts Honeywell HIH-8130-003, the gas sensor adopts MembraporMGS-4, the liquid level sensor adopts VEGAFLEX 81, the flow sensor adopts Endress+Hauser Promag 55S, the pressure sensor adopts Honeywell MBS3000, and the vibration sensor adopts PCB Piezotronics 352C33.

[0023] Step S200: Transmitting pesticide production data to the cloud platform.

[0024] Specifically, during the pesticide production process, various data acquisition devices, such as industrial cameras, temperature sensors, humidity sensors, etc., continuously collect various types of data on pesticide production. In this embodiment, the sensor is connected to the mobile network through a built-in 4G / 5G communication module. Using the TCP / IP data transmission protocol, the sensor encapsulates the collected pesticide production data into a standard data packet format. These data packets contain pesticide production raw material data, environmental parameter data, regional image data, and equipment operation status data, etc. These data packets are then sent to the cloud platform via the mobile network. The cloud platform receives and stores data from various sensors, and verifies and parses the data packets when receiving the data.

[0025] Step S300: The cloud platform performs data processing and data analysis on the received pesticide production data.

[0026] Specifically, data processing includes data cleaning, data normalization, data dimension reduction and outlier detection.

[0027] Data analysis includes using a personnel detection model to detect the presence of personnel in the pesticide production area; using a pesticide production monitoring model to monitor the pesticide production process; and using a pesticide production equipment monitoring model to monitor pesticide production equipment failures.

[0028] The personnel detection model is trained using the improved Yolo v5 algorithm; the pesticide production monitoring model is trained using a deep learning algorithm based on a graph neural network; and the pesticide production equipment monitoring model is trained using a deep learning algorithm based on LSTM.

[0029] The improved Yolo v5 algorithm specifically includes: the first input layer, which is used to receive image data of pesticide production areas; the feature extraction layer, which extracts image feature information through EfficientViT; the feature fusion layer, which fuses image feature information of different scales; the prediction layer, which predicts the confidence and position of people based on the fused image feature information; the first output layer, which is used to output the final detection result of the people detection model.

[0030] The deep learning algorithm based on graph neural network specifically includes: a second input layer, used to receive the input data required by the pesticide production monitoring model; a graph construction layer, used to construct graph structure data based on the input data; a graph convolution layer, used to perform convolution operations on the graph structure data and extract node features; a first fully connected layer, used to map the features extracted by the graph convolution layer to output categories; and a second output layer, used to output the monitoring results of the pesticide production process.

[0031] The LSTM-based deep learning algorithm specifically includes: a third input layer, used to receive the input data required by the pesticide production equipment monitoring model; an LSTM layer, used to perform sequence modeling on the input data and extract time series features; a second fully connected layer, used to map the features extracted by the LSTM layer to output categories; and a third output layer, used to output the fault prediction results of the pesticide production equipment.

[0032] In this embodiment, the data processing process includes data cleaning, data normalization, data dimension reduction and outlier detection. First, data cleaning is performed, and the DataFrame (referred to as "df") containing pesticide production data is read. The missing values ​​in df are processed by the mean value filling method, that is, for each missing value in df, the average value of the column in which it is located is used to fill it, and then the duplicate data in df is removed to ensure that each row of data is unique. Then the data is normalized, and the MinMaxScaler method is used to normalize the columns to be normalized so that each value is scaled between 0 and 1, and the normalized data replaces the data of the corresponding column in the original df. Next, data dimension reduction is performed. First, the data of the numerical column to be reduced in dimension is extracted from the normalized df, and it is converted into a NumPy array X, and then the PCA method is used to reduce the dimension of X, and the high-dimensional data is converted into two-dimensional data. The reduced-dimensional data is stored in a new DataFrame df_pca, and finally df_pca is merged with the df after removing the column to be reduced in dimension to obtain the reduced-dimensional DataFrame df_reduced. Finally, outlier detection is performed. The IsolationForest method is used to detect outliers on the reduced-dimensional data X. First, the IsolationForest model is trained and used to predict X to obtain the predicted result yhat. Then, a new column "anomaly_flag" is created in df according to the value of yhat to mark whether each data point is an outlier. If the value of yhat is -1, the corresponding data point is marked as an outlier ("anomaly"), otherwise it is marked as a normal value ("normal").

[0033] The personnel detection model is trained using the improved Yolo v5 algorithm. The data set is acquired through industrial cameras and annotated using LabelImg. The annotated data set is divided into a training set and a validation set at a ratio of 8:2. In the images folder, the images of the training set and the validation set are stored in train and val. In the labels folder, the corresponding train and val files of these images are stored. The data / pesticide_production.yaml file is modified to specify the image paths of the training set and the validation set, set the number of categories to 1, and list the category name as ['person'].

[0034] First, the ImprovedYoloV5 class is constructed, which inherits from nn.Module. Then the EfficientViT model is used as the feature extraction layer. The parameters of the EfficientViT model include image size img_size=640, patch size patch_size=4, embedding dimension embed_dim=768, depth depth=6, number of heads num_heads=12, multi-layer perceptron dimension mlp_dim=3072, and dropout rate dropout=0.1. Then the head part of the Yolo v5 model is loaded, and the head of Yolov5 is constructed according to the feature dimension ch=self.backbone.num_features output by the backbone and the number of target categories nc=num_classes. In the forward propagation process of the model, the features of the input image are extracted through the backbone to obtain the feature map features, and finally the features are input into the yolo head for prediction to obtain the output outputs. The output contains the results of person detection, including information such as the location and confidence of the person.

[0035] The EfficientViT model consists of a Stem part and multiple stage feature extraction modules. The input image first passes through the Stem part to extract fine-grained low-level features, and then the features are gradually downsampled and optimized through multiple stages. In each stage, the MBConv module combined with PonitWiseConv+DepwiseConv is used to reduce model parameters, and the EfficientViT Module is added for linear self-attention feature optimization.

[0036] During the training process, the model model, data loader dataloader, loss function criterion, optimizer optimizer, device device and training round number num_epochs are received. First, the model is set to training mode and the training data set is traversed. For each batch of data, it is input into the model for forward propagation to obtain the predicted output, and then the loss function value is calculated, and back propagation and optimization are performed. .

[0037] The pesticide production monitoring model is trained using a deep learning algorithm based on a graph neural network. The data set consists of pesticide production raw materials, pesticide production environmental parameters collected by sensors, the operating status of pesticide production equipment, and pesticide production status. The data set is divided into a training set and a validation set in a ratio of 8:2.

[0038] First, a GCN model class is defined, which inherits from torch.nn.Module. In the initialization function of the GCN model class, two GCN convolutional layers, conv1 and conv2, are defined. The number of input channels of conv1 is in_channels, and the number of output channels is hidden_channels; the number of input channels of conv2 is hidden_channels, and the number of output channels is out_channels. These two convolutional layers are used to extract features from graph structure data. Next, the forward propagation function of the GCN model is defined. In the forward propagation function, the node feature x and edge index edge_index of the input data are first obtained, and then the node feature and edge index are input into the conv1 convolutional layer to obtain the convolutional features. Then, the ReLU activation function is used to perform nonlinear transformation on the convolutional features, and the dropout layer is used for regularization to prevent overfitting. Finally, the features after ReLU activation and dropout regularization are input into the conv2 convolutional layer to obtain the final output features. The output features are normalized using the log_softmax function to obtain the predicted probability of each category.

[0039] The training process instantiates the GCN model according to the number of node features and the number of categories in the dataset, and then defines an Adam optimizer to optimize the model's parameters. The training first sets the model to training mode, then clears the optimizer's gradient, then passes the input data to the model to get the predicted output, calculates the negative log-likelihood loss between the predicted output and the true label, and uses the back-propagation algorithm to calculate the gradient; finally, the optimizer is used to update the model's parameters. During the test process, the model is set to evaluation mode, and the trained model is used to predict the data. In addition, when abnormal parameters are monitored, the output of the model determines which adjustment method to output. Define a get_adjustment_method function that returns the corresponding adjustment method based on the abnormality type.

[0040] The pesticide production equipment monitoring model is trained using a LSTM-based deep learning algorithm. The data set consists of the operating status of the pesticide production equipment and the fault data of the pesticide production equipment. The data set is divided into a training set and a validation set in a ratio of 8:2.

[0041] First, the data in the training set and validation set are standardized. Create a standardizer scaler through the StandardScaler class, and use the fit_transform method to fit and transform the training set X_train while maintaining its original shape. For the validation set X_val, use the transform method to transform it, also maintaining its original shape. Next, convert the processed data into PyTorch tensors, X_train and X_val are converted to floating-point tensors, and y_train and y_val are converted to long integer tensors, representing the input features and labels respectively.

[0042] Then build the LSTM model. In the class initialization method, set the model's input size input_size, hidden layer size hidden_size, number of layers num_layers, and number of output categories num_classes. Use nn.LSTM to create an LSTM layer with an input size of input_size, a hidden layer size of hidden_size, and number of layers num_layers, and set batch_first=True. Then use nn.Linear to create a fully connected layer to map the output of the LSTM layer to the number of output categories. In the forward propagation method of the model, first initialize the hidden state and cell state to zero tensors, whose shape and number are determined by the number of layers and batch size of the model, then use the LSTM layer to process the input tensor, and take out the output of the last time step, and finally pass the output to the fully connected layer to get the final prediction result.

[0043] In the model training phase, the loss function is set to cross entropy loss nn.CrossEntropyLoss(), the optimizer is Adam optimizer optim.Adam(), and the learning rate is set to 0.001. Then 20 training cycles are iterated. In each cycle, model.train() is first called to set the model to training mode, then the output and loss of the model are calculated, and then optimizer.zero_grad() is used to clear the previous gradient, loss.backward() is called to calculate the current gradient, and optimizer.step() is used to update the model parameters.

[0044] In the model evaluation phase, call model.eval() to set the model to evaluation mode, and use the validation set data to calculate the model's predicted output. Then use the torch.max function to find the index of the maximum value in the predicted output as the predicted fault type. Finally, use the get_fault_handling_method function to obtain the corresponding fault handling method based on the predicted fault type.

[0045] Step S400: Outputting pesticide production control strategy based on data analysis results.

[0046] Specifically, after the cloud platform completes data analysis, the system will automatically generate a pesticide production control strategy based on the analysis results. When the personnel detection model identifies the presence of personnel in the pesticide production area based on the Yolo v5 algorithm, the system records the time, location and other detailed information of the personnel, and compares it with the preset safety specifications to determine whether the personnel activities meet the safety production requirements. If there is any violation, such as unauthorized personnel entering the production area, the system triggers the alarm mechanism and sends a notification to the management personnel through the cloud platform.

[0047] The pesticide production monitoring model uses graph neural networks to monitor and analyze various parameters in the production process in real time. Once it is detected that environmental parameters such as temperature, humidity, and harmful gas concentration or production parameters such as raw material addition and reactor liquid level are beyond the preset range, the system will judge it as an abnormal state. At this time, the system will automatically recommend or output the corresponding pesticide production equipment adjustment method based on the type and severity of the abnormal parameters. For example, if the temperature sensor detects that the production environment temperature is too high, the ventilation volume will be increased or the cooling system parameters will be adjusted to lower the temperature; if the liquid level sensor detects that the liquid level in the raw material tank is too low, the operator will be prompted to replenish the raw materials in time.

[0048] Finally, the pesticide production equipment monitoring model continuously monitors the operating status of the equipment through the LSTM algorithm. Once fault signs such as abnormal equipment vibration, pressure fluctuations, or unstable flow are detected, the system activates the fault warning mechanism and outputs specific fault handling methods. These handling methods include adjusting equipment operating parameters, suspending equipment operation for maintenance, or starting backup equipment. At the same time, the system will also send fault alarm information to relevant personnel based on the type and urgency of the fault to ensure that the fault can be handled promptly and effectively.

[0049] In this embodiment, the preset temperature range of the production environment is 15°C to 30°C. If the temperature sensor detects that the ambient temperature is lower than 15°C, the system starts the heating device to increase the temperature; when the temperature exceeds 30°C, the system increases the ventilation volume and adjusts the cooling system parameters to lower the temperature to ensure that the production environment remains within a suitable temperature range. The preset range for the humidity system is 40%RH to 70%RH. If the humidity sensor detects that the ambient humidity is lower than 40%RH, a humidifier is used to increase the moisture content in the air; and when the humidity exceeds 70%RH, a dehumidifier is used or the air humidity control system is adjusted to reduce the humidity of the production environment. The system sets the corresponding safety concentration threshold according to the types of harmful gases that may be generated during the production process. For example, for ammonia, the safety concentration threshold is set to no more than 25ppm. If the concentration of harmful gases exceeds the threshold, the system will immediately trigger the alarm mechanism and suggest that the operator immediately stop production, wear protective equipment, start air purification equipment or turn on the ventilation system to reduce the concentration of harmful gases and ensure the safety of production personnel.

[0050] In terms of the amount of raw materials added, the system sets the range of the amount of each raw material added. If the amount of raw materials added exceeds the preset range, the system will automatically issue a warning and suggest that the operator check the raw material ratio, adjust the amount of raw materials added or suspend production to prevent the quality of the product from being affected by improper raw material ratio. At the same time, the system will set a safe range for the liquid level according to the capacity of the reactor. If the liquid level sensor detects that the reactor level is too low, the system will prompt the operator to replenish the raw materials in time to prevent incomplete reaction or idling of the equipment; when the liquid level is too high, the system will stop adding raw materials, adjust the reaction conditions or start the emission system to prevent the reactor from overflowing or causing danger.

[0051] The pesticide production equipment failures in this embodiment include motor overheating failure, bearing wear failure, pump leakage failure, valve failure and electrical line short circuit failure.

[0052] Motor overheating failures include emergency level, with failure probability greater than 0.8, indicating that the motor is seriously overheated and needs to be stopped immediately for inspection and measures to be taken; important level, with failure probability between 0.5 and 0.8, indicating that the motor overheating is already quite obvious and a maintenance plan needs to be arranged as soon as possible; general level, with failure probability lower than 0.5, indicating that the motor overheating is slight and can be handled during production breaks or regular maintenance.

[0053] Bearing wear failures include emergency level, with failure probability greater than 0.75, indicating that the bearing is severely worn and has affected the stable operation of the equipment, and needs to be replaced or repaired immediately; important level, with failure probability between 0.4 and 0.75, indicating that the bearing has signs of wear and needs to be monitored more closely and repairs arranged as soon as possible; general level, with failure probability less than 0.4, indicating that the bearing is slightly worn and its status can be monitored during daily inspections.

[0054] Pump leakage failures include emergency level, with a failure probability greater than 0.65, indicating that the pump leakage is serious and may have caused pollution to the production environment, and immediate measures must be taken to repair it; important level, with a failure probability between 0.35 and 0.65, indicating that the pump leakage is present and monitoring must be strengthened and a maintenance plan must be arranged as soon as possible; general level, with a failure probability lower than 0.35, indicating that the pump leakage is minor and can be checked and handled during regular maintenance.

[0055] Valve failure includes emergency level, with failure probability greater than 0.7, indicating that the valve can no longer be opened or closed normally, causing serious impact on the production process and requiring immediate repair; important level, with failure probability between 0.4 and 0.7, indicating that the valve has signs of failure and requires enhanced monitoring and maintenance as soon as possible; general level, with failure probability lower than 0.4, indicating that the valve failure is minor and its status can be monitored during daily inspections.

[0056] Electrical line short-circuit failures include emergency level, with a failure probability greater than 0.85, indicating that there is a serious short circuit in the electrical line, which may cause serious consequences such as fire, and it is necessary to immediately cut off the power supply and take measures to repair it; important level, with a failure probability between 0.55 and 0.85, indicating that there are signs of short circuit in the electrical line, and it is necessary to strengthen monitoring and arrange a maintenance plan as soon as possible; general level, with a failure probability lower than 0.55, indicating that the short circuit in the electrical line is minor and can be checked and handled during regular maintenance.

[0057] Step S500: Automatically control pesticide production equipment according to the pesticide production control strategy.

[0058] Specifically, according to the pesticide production management and control strategy, control instructions are sent to the pesticide production equipment through the cloud platform; the pesticide production equipment receives and parses the control instructions, automatically adjusts the pesticide production equipment parameters, and starts or stops the pesticide production equipment according to the instruction content; the cloud platform continuously monitors the execution status of the pesticide production equipment and the real-time changes of production data.

[0059] In this embodiment, the cloud platform first generates specific control instructions based on the data analysis results, including adjusting the temperature, humidity, pressure and other parameters of the production equipment, starting or stopping specific production equipment, or adjusting the operation rhythm of the production line. These instructions are encapsulated into a data packet format and transmitted to the corresponding pesticide production equipment through the network.

[0060] After receiving the control instructions sent by the cloud platform, the pesticide production equipment will parse and process them. The control system inside the equipment will automatically adjust its operating parameters or perform corresponding actions according to the content of the instructions. For example, if the instruction requires lowering the temperature of the production environment, the cooling system inside the equipment will start and lower the temperature; if the instruction requires stopping a certain production link, the corresponding equipment will stop running.

[0061] At the same time, the cloud platform will continue to monitor the execution of pesticide production equipment and the real-time changes in production data. Through data collection equipment, the cloud platform obtains real-time information such as the operating status of production equipment, production parameters, and production data. In addition, the cloud platform will also record detailed information on each control operation, including control time, control content, and execution results.

[0062] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A pesticide production control method based on cloud-network fusion technology, characterized in that: The method comprises: Step S100: using data collection equipment to collect pesticide production data in real time; Step S200: Transmitting pesticide production data to the cloud platform; Step S300: the cloud platform performs data processing and data analysis on the received pesticide production data; Step S400: outputting a pesticide production control strategy based on the data analysis results; Step S500: Automatically control pesticide production equipment according to the pesticide production control strategy.

2. The pesticide production control method according to claim 1, characterized in that: The pesticide production data includes pesticide production raw materials, pesticide production environmental parameters, pesticide production area images and operating status of pesticide production equipment.

3. The pesticide production control method according to claim 1, characterized in that: The data processing includes data cleaning, data normalization, data dimension reduction and outlier detection.

4. The pesticide production control method according to claim 3, characterized in that: The data analysis includes using a personnel detection model to detect the presence of personnel in the pesticide production area; using a pesticide production monitoring model to monitor the pesticide production process; and using a pesticide production equipment monitoring model to monitor the failure of pesticide production equipment.

5. The pesticide production control method according to claim 4, characterized in that: The personnel detection model is trained using the improved Yolo v5 algorithm; the pesticide production monitoring model is trained using a deep learning algorithm based on a graph neural network; and the pesticide production equipment monitoring model is trained using a deep learning algorithm based on LSTM.

6. The pesticide production control method according to claim 5, characterized in that: The improved Yolo v5 algorithm specifically includes: The first input layer is used to receive image data of pesticide production areas; Feature extraction layer, extracting image feature information through EfficientViT; Feature fusion layer, which fuses image feature information of different scales; The prediction layer predicts the confidence and location of the person based on the fused image feature information; The first output layer is used to output the final detection results of the person detection model.

7. The pesticide production control method according to claim 5, characterized in that: The deep learning algorithm based on graph neural network specifically includes: The second input layer is used to receive the input data required by the pesticide production monitoring model; The graph construction layer is used to construct graph structure data based on input data; Graph convolution layer, used to perform convolution operations on graph structure data and extract node features; The first fully connected layer is used to map the features extracted by the graph convolutional layer to the output categories; The second output layer is used to output the monitoring results of the pesticide production process.

8. The pesticide production control method according to claim 5, characterized in that: The LSTM-based deep learning algorithm specifically includes: The third input layer is used to receive the input data required by the pesticide production equipment monitoring model; LSTM layer, used to perform sequence modeling on input data and extract time series features; The second fully connected layer is used to map the features extracted by the LSTM layer to the output categories; The third output layer is used to output the fault prediction results of pesticide production equipment.

9. The pesticide production control method according to claim 5, characterized in that: The pesticide production control strategy includes: When the personnel detection model detects the presence of personnel in the pesticide production area, the presence of personnel is recorded; When the pesticide production monitoring model detects abnormal parameters in the pesticide production process, it outputs a method for adjusting the pesticide production equipment; When the pesticide production equipment monitoring model detects that the pesticide production equipment has a fault, it outputs a fault handling method.

10. The pesticide production control method according to claim 1, characterized in that: The step S500 specifically includes: According to the pesticide production control strategy, control instructions are sent to pesticide production equipment through the cloud platform; The pesticide production equipment receives and analyzes the control instructions, automatically adjusts the parameters of the pesticide production equipment, and starts or stops the pesticide production equipment according to the instructions; The cloud platform continuously monitors the performance of pesticide production equipment and real-time changes in production data.

Images