A safety monitoring method and system for trichloroisocyanuric acid wastewater treatment

Abstract

The application discloses a kind of safety monitoring method and system for trichloroisocyanuric acid wastewater treatment, it is related to the technical field of data analysis.The vibration signal is converted into two-dimensional time-frequency image, and the two-dimensional time-frequency image is used as the input of the pre-trained multiple state recognition models, to obtain the prediction probability of multiple state categories;According to the preset integrated weight, the prediction probability of the target state category is weighted and summed to obtain the probability score of the target state category;The state category corresponding to the maximum probability score is output as the detection result.Through the pre-training model, the state of two-dimensional time-frequency image is recognized, and the prediction accuracy is improved by weighted integration, which can effectively identify early subtle faults.Compared with the traditional threshold method, the fault recognition time point is advanced from the obvious damage stage to the damage not obvious stage, which helps to early warning and reduces the risk of equipment sudden failure.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, and in particular to a safety monitoring method and system for trichloroisocyanuric acid wastewater treatment. BACKGROUND

[0002] As a highly efficient chlorine-containing disinfectant, the production of trichloroisocyanuric acid produces wastewater with high salt and high organic matter concentration, which poses a serious threat to the ecological environment. In order to realize the resource treatment and energy saving and emission reduction of wastewater, mechanical vapor recompression (MVR) process is widely used due to its high efficiency and energy saving advantage. The process realizes heat energy recycling by compressing and warming the secondary steam with a compressor. As the core equipment of the MVR system, the stable operation of the compressor directly determines the efficiency and energy consumption level of the entire wastewater treatment system. The performance degradation of the bearing as a key component of the compressor can easily cause equipment vibration abnormalities, efficiency decline, and even sudden stop, which not only interrupts the wastewater treatment process, but also may cause high-temperature and high-pressure steam leakage and other safety accidents.

[0003] At present, most enterprises still rely on manual inspection and regular maintenance to monitor the running state of the MVR compressor, which has problems such as long detection cycle. Some automatic monitoring schemes use threshold early warning, such as a monitoring method based on a temperature sensor. However, early subtle abnormalities do not cause obvious characteristics during machine operation. This type of threshold early warning method lacks sensitivity to early potential failures. The traditional monitoring method has a problem of reaction lag in detecting early fault hazards, which can cause the hazards to gradually intensify with the operation of the equipment, resulting in serious consequences such as unplanned shutdown, interruption of the wastewater treatment process, rising equipment repair costs, and even safety hazards such as high-temperature and high-pressure steam leakage. SUMMARY

[0004] The purpose of the present application is to solve the problem of reaction lag mentioned in the background, and a safety monitoring method and system for trichloroisocyanuric acid wastewater treatment are proposed.

[0005] In the first aspect of the present application, a safety monitoring method for trichloroisocyanuric acid wastewater treatment is provided, which uses a mechanical vapor recompression process to treat wastewater, and the method comprises:

[0006] Obtaining a vibration signal of the compressor and preprocessing the vibration signal to obtain a target signal;

[0007] Converting the target signal into a two-dimensional time-frequency image;

[0008] Taking the two-dimensional time-frequency image as an input of a plurality of pre-trained state recognition models to obtain prediction probabilities of a plurality of state categories;

[0009] According to a preset integrated weight, the prediction probabilities of the target state categories are weighted and summed to obtain probability scores of the target state categories; the target state categories are any one of the plurality of state categories;

[0010] The state category corresponding to the maximum probability score is output as a detection result.

[0011] Optionally, the target signal is obtained by preprocessing the vibration signal, and the preprocessing includes:

[0012] A first vibration signal of a preset length is extracted from the vibration signal;

[0013] The first vibration signal is subjected to variational modal decomposition to obtain a plurality of intrinsic modal components;

[0014] A Pearson correlation coefficient of each intrinsic modal component and the first vibration signal is calculated;

[0015] The intrinsic modal components with an absolute value of the correlation coefficient greater than a preset threshold are taken for signal reconstruction to obtain the target signal.

[0016] Optionally, the plurality of state recognition models have the same structure; and any one state recognition model includes:

[0017] An initial convolution module is configured to perform shallow feature extraction on the two-dimensional time-frequency image by using one convolution layer and perform spatial compression to obtain a first feature map;

[0018] A feature extraction module is configured to perform multi-scale feature extraction on the first feature map by using convolution layers with different kernel sizes to obtain a second feature map;

[0019] An attention enhancement module is configured to perform feature enhancement on the second feature map by using an attention mechanism to obtain a third feature map;

[0020] A reshaping module is configured to convert the third feature map into a feature vector;

[0021] A classification module is configured to map the feature vector to a probability vector of a preset scale by using a full connection network.

[0022] Optionally, the feature extraction module includes two cascaded multi-scale dense blocks; the two multi-scale dense blocks have the same structure, and the operation process of any one multi-scale dense block includes:

[0023] wherein, X inis the input of the multi-scale dense block; Conv represents a convolution operation, the subscript 3x3, 5x5, 7x7 and 1x1 represent the size of the convolution kernel, and the superscript 1 represents the convolution step length; concat represents concatenation in the channel dimension; MaxPool represents maximum pooling, and the subscript 2x2 represents the size of the pooling window, and the superscript 2 represents the pooling step length; X 11 , X 12 , X 13 , X 14 , X 21 , X 22 , X 23 and X 24 is a feature map generated during operation, X out is the output of the multi-scale dense block.

[0024] Optionally, the integrated weight is determined by the training effect of the plurality of state recognition models, specifically, after the plurality of state recognition models are trained, a particle swarm optimization algorithm is used to optimize and determine the integrated weight, comprising:

[0025] Step one, using a plurality of state recognition models to predict each sample of the verification set to obtain a probability vector corresponding to each sample;

[0026] Step two, initialize the population size, particle position and speed;

[0027] Step three, convert the particle position into an effective weight distribution using a softmax function;

[0028] Step four, according to the effective weight distribution, integrate the probability vector of each sample into a probability score to obtain the final prediction result, and calculate its accuracy as the fitness;

[0029] Step five, update the individual optimal position and the global optimal position according to the fitness value;

[0030] Step six, determine whether the preset stop condition is reached; if yes, end the optimization process; otherwise, go to step seven;

[0031] Step seven, update the particle position according to the individual optimal position and the global optimal position, and return to step three.

[0032] The second aspect of the embodiment of the application provides a safety monitoring system for treating trichloroisocyanuric acid wastewater, which is used for realizing the safety monitoring method for treating trichloroisocyanuric acid wastewater.

[0033] The data acquisition module is configured to acquire a vibration signal of the compressor.

[0034] The preprocessing module is configured to preprocess the vibration signal to obtain a target signal.

[0035] a two-dimensional representation module, configured to convert the target signal into a two-dimensional time-frequency image;

[0036] a base model decision module, configured to take the two-dimensional time-frequency image as an input of a plurality of pre-trained state recognition models, to obtain prediction probabilities of a plurality of state categories;

[0037] an integration module, configured to weight and sum the prediction probabilities of a target state category according to preset integration weights, to obtain a probability score of the target state category; the target state category is any one of the plurality of state categories;

[0038] a result output module, configured to output a state category corresponding to a maximum probability score as a detection result.

[0039] Optionally, the preprocessing module comprises:

[0040] a segment interception module, configured to extract a first vibration signal of a preset length from the vibration signal;

[0041] a modal decomposition module, configured to perform variational modal decomposition on the first vibration signal, to obtain a plurality of intrinsic modal components;

[0042] a correlation calculation module, configured to calculate a Pearson correlation coefficient of each intrinsic modal component and the first vibration signal;

[0043] a signal reconstruction module, configured to take an intrinsic modal component with an absolute value of the correlation coefficient greater than a preset threshold to perform signal reconstruction, to obtain a target signal.

[0044] Optionally, the plurality of state recognition models are of the same structure; any one of the state recognition models comprises:

[0045] an initial convolution module, configured to perform shallow feature extraction and spatial compression on the two-dimensional time-frequency image by using one convolution layer, to obtain a first feature map;

[0046] a feature extraction module, configured to perform multi-scale feature extraction on the first feature map by using convolution layers with different kernel sizes, to obtain a second feature map;

[0047] an attention enhancement module, configured to perform feature enhancement on the second feature map by using an attention mechanism, to obtain a third feature map;

[0048] a reshaping module, configured to convert the third feature map into a feature vector;

[0049] a classification module, configured to map the feature vector to a probability vector of a preset scale by using a fully connected network.

[0050] Optionally, the feature extraction module comprises two cascaded multi-scale dense blocks; the two multi-scale dense blocks are identical in structure, and the operation process of any one of the multi-scale dense blocks comprises:

[0051] wherein, X in is the input of the multi-scale dense block; Conv represents convolution operation, the subscript 3x3, 5x5, 7x7 and 1x1 represent the size of the convolution kernel, and the superscript 1 represents the convolution step length; concat represents concatenation in the channel dimension; MaxPool represents maximum pooling, the subscript 2x2 represents the size of the pooling window, and the superscript 2 represents the pooling step length; X 11 , X 12 , X 13 , X 14 , X 21 , X 22 , X 23 and X 24 are feature maps generated in the operation process, and X out is the output of the multi-scale dense block.

[0052] Optionally, the system further comprises an integrated optimization module for optimizing and determining the integrated weight by using a particle swarm optimization algorithm after the plurality of state recognition models are trained; the operation process of the integrated optimization module comprises:

[0053] Step one, predicting each sample of the verification set by using a plurality of state recognition models to obtain a probability vector corresponding to each sample;

[0054] Step two, initializing the population size, particle position and speed;

[0055] Step three, converting the particle position into an effective weight distribution by using a softmax function;

[0056] Step four, integrating the probability vector of each sample into a probability score according to the effective weight distribution to obtain a final prediction result, and calculating the accuracy rate as the fitness;

[0057] Step five, updating the individual optimal position and the global optimal position according to the fitness value;

[0058] Step six, judging whether the preset stop condition is reached; if yes, ending the optimization process; otherwise, entering step seven;

[0059] Step seven, updating the particle position according to the individual optimal position and the global optimal position, and returning to step three.

[0060] The beneficial effects of the present application are as follows:

[0061] The application provides a safety monitoring method for trichloroisocyanuric acid wastewater treatment.

[0062] By collecting the vibration signal of the compressor and converting it into a two-dimensional time-frequency image, using multiple pre-trained models for state recognition, and improving the prediction accuracy through weighted integration, early and subtle faults can be effectively identified. Compared with the traditional threshold method, this method advances the fault recognition time point from the obvious damage stage to the non-obvious damage stage, which helps to early warning and reduces the risk of equipment sudden failure. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 A flowchart of a safety monitoring method for trichloroisocyanuric acid wastewater treatment is provided for the embodiments of the application.

[0064] Figure 2 A network architecture diagram of a state recognition model is provided for the embodiments of the application.

[0065] Figure 3 A structural diagram of a multi-scale dense block is provided for the embodiments of the application.

[0066] Figure 4 A structural diagram of a safety monitoring system for trichloroisocyanuric acid wastewater treatment is provided for the embodiments of the application. DETAILED DESCRIPTION

[0067] To further illustrate the technical means and effects adopted by the application to achieve the predetermined invention purposes, the specific embodiments, structures, features and effects according to the application are described in detail below in combination with the drawings and preferred embodiments.

[0068] The embodiments of the application provide a safety monitoring method for trichloroisocyanuric acid wastewater treatment. Referring to Figure 1 , Figure 1 A flowchart of a safety monitoring method for trichloroisocyanuric acid wastewater treatment is provided for the embodiments of the application. The method includes the following steps:

[0069] S101, obtaining the vibration signal of the compressor.

[0070] S102, pre-processing the vibration signal to obtain a target signal.

[0071] S103, convert the target signal into a two-dimensional time-frequency image.

[0072] S104, input the two-dimensional time-frequency image into a plurality of pre-trained state recognition models to obtain prediction probabilities of a plurality of state categories.

[0073] S105, according to a preset integration weight, weight and sum the prediction probabilities of the target state category to obtain a probability score of the target state category.

[0074] S106, output the state category corresponding to the maximum probability score as a detection result.

[0075] The target state category is any one of the plurality of state categories.

[0076] The safety monitoring method for trichloroisocyanuric acid wastewater treatment provided by the embodiment of the application can effectively identify early subtle faults by collecting compressor vibration signals and converting them into two-dimensional time-frequency images, using a plurality of pre-trained models for state recognition, and improving prediction accuracy through weighted integration. Compared with the traditional threshold method, this method advances the fault identification time point from the obvious damage stage to the non-obvious damage stage, which helps to advance the early warning and reduce the risk of equipment sudden failure.

[0077] In an implementation manner, the continuous wavelet transform can be used to convert the target signal into a two-dimensional time-frequency image.

[0078] In an implementation manner, the results of the plurality of state recognition models are subjected to ensemble learning and probability weighting to improve the robustness and generalization ability of the overall judgment.

[0079] In one embodiment, step S102 includes:

[0080] Step one, from the vibration signal, a first vibration signal of a preset length is extracted.

[0081] Step two, the first vibration signal is subjected to variational modal decomposition to obtain a plurality of intrinsic modal components.

[0082] Step three, the Pearson correlation coefficient of each intrinsic modal component and the first vibration signal is calculated.

[0083] Step four, the intrinsic modal components with an absolute value of the correlation coefficient greater than a preset threshold value are taken for signal reconstruction to obtain a target signal.

[0084] This embodiment effectively removes noise and redundant information, extracts key features, and improves the quality of the target signal, thereby enhancing the accuracy and robustness of subsequent state recognition, by performing variational modal decomposition on the vibration signal and using the Pearson correlation coefficient to select the intrinsic modal components highly correlated with the original signal for reconstruction.

[0085] In one implementation, the preset length is 1024. The preset threshold is 0.2.

[0086] In one embodiment, multiple state recognition models have the same structure but different weights. See also Figure 2 , Figure 2 This invention provides a network architecture diagram for a state recognition model. As shown in the figure, any state recognition model includes an initial convolutional module, a feature extraction module (first multi-scale dense block + second multi-scale dense block), an attention enhancement module, a shaping module, and a classification module, wherein:

[0087] The initial convolutional module is used to perform shallow feature extraction and spatial compression on the two-dimensional time-frequency image using a single convolutional layer to obtain the first feature map. Specifically, a 3×3 convolutional layer is used for shallow feature extraction, and a max pooling layer with a stride of 2 is used for spatial compression.

[0088] The feature extraction module is used to perform multi-scale feature extraction on the first feature map using convolutional layers with different kernel sizes to obtain the second feature map. Specifically, the feature extraction module includes two cascaded multi-scale dense blocks, which have the same structure but different weights. See also... Figure 3 , Figure 3 This invention provides a schematic diagram of a multi-scale dense block structure. The computation process of any multi-scale dense block includes:

[0089] Among them, X in This is the input to the multi-scale dense block; Conv represents the convolution operation, with subscripts 3×3, 5×5, 7×7, and 1×1 indicating the kernel size and superscript 1 indicating the stride; concat indicates concatenation along the channel dimension; MaxPool represents max pooling, with subscript 2×2 indicating the pooling window size and superscript 2 indicating the stride; X 11 X 12 X 13 X 14 X 21 X 22 X 23 and X 24 It is a feature map generated during the operation, X out This is the output of the multi-scale dense block. The output of the first multi-scale dense block is the input of the second multi-scale dense block.

[0090] The attention enhancement module is used to enhance the features of the second feature map using an attention mechanism to obtain the third feature map. Specifically, it employs a channel attention mechanism for feature enhancement. The operation process of the attention enhancement module includes:

[0091] wherein x is the input of the attention enhancement module; GAP represents the global average pooling; FC represents the fully connected layer; ReLu and Sigmoid are two activation functions; T is the channel weight; and y is the output of the attention enhancement module.

[0092] a reshaping module, configured to convert the third feature map into a feature vector. Specifically, the third feature map is flattened into a one-dimensional feature vector.

[0093] a classification module, configured to map the feature vector to a probability vector of a preset scale by using a fully connected network. Specifically, three fully connected layers are used for classification, and the number of neurons is 128, 32 and 10 in turn. The third fully connected layer uses a Softmax function to output the probability of each class, including 10 classes, namely normal, mild inner ring fault, moderate inner ring fault, mild outer ring fault, moderate outer ring fault, mild rolling body fault, moderate rolling body fault, mild compound fault and moderate compound fault.

[0094] The embodiment proposes a three-branch network (multi-scale dense block) to perform multi-scale decomposition and feature extraction fusion on input data, so as to capture more comprehensive and strongly correlated fault features, thereby effectively enhancing the feature extraction and classification ability of the model. And the dense connection can also enhance the information flow and gradient propagation, which can alleviate the gradient vanishing and improve the training efficiency and convergence speed.

[0095] In an implementation manner, the channel attention mechanism is introduced to dynamically adjust the importance of each channel, suppress redundant information and highlight key fault features, thereby improving the model recognition effect.

[0096] In an implementation manner, the explicit fault level can help the operation and maintenance team to reasonably arrange the maintenance priority, save maintenance cost and time.

[0097] In an embodiment, one-dimensional vibration signals of the compressor under different loads, different speeds and different states are collected, and overlapping sampling is performed to obtain an original data set. The original data set is divided into a training set and a verification set. The original training set is randomly sampled with replacement to obtain a training subset of each state recognition model, allowing sample repetition. Different training subsets are used to train multiple state recognition models in parallel. The training process adopts cross-entropy loss.

[0098] In an embodiment, the integrated weight is determined by the training effect of the multiple state recognition models. Specifically, after the multiple state recognition models are trained, a particle swarm optimization algorithm is used to optimize and determine the integrated weight, including:

[0099] Step one, a plurality of state recognition models are used to predict each sample of the verification set to obtain the probability vector corresponding to each sample. For example, there are four state recognition models, each model outputs a 10-dimensional probability vector, and for each state category, the prediction probability has four, which is given by the four state recognition models.

[0100] Step two, the population size, particle position and speed are initialized. Specifically, the population size can be set to 30. Random numbers in (0, 1) are generated as particle positions, and the particle position is a four-dimensional vector, and the vector element represents the fusion weight of the state recognition model. The speed is initialized to 0.

[0101] Step three, the particle position is converted into an effective weight distribution by using the Softmax function. The elements are normalized by the Softmax function, so that the sum of the elements is 1.

[0102] Step four, according to the effective weight distribution, the probability vector of each sample is integrated into a probability score to obtain the final prediction result, and the accuracy is calculated as the fitness.

[0103] Step five, according to the fitness value, the individual optimal position and the global optimal position are updated.

[0104] Step six, it is judged whether the preset stop condition is reached; if yes, the optimization process is ended; otherwise, step seven is entered. Specifically, the stop conditions include reaching the maximum iteration number. After the optimization process is ended, the global optimal position is normalized by using the Softmax function to obtain the optimal integrated weight.

[0105] Step seven, according to the individual optimal position and the global optimal position, the particle position is updated, and step three is returned. Specifically, the speed and position are updated according to the rules of the particle swarm optimization algorithm, which will not be repeated here.

[0106] In an implementation manner, by weighting and fusing the prediction results of a plurality of state recognition models, the weight distribution is optimized in combination with the performance on the verification set, so that the advantages of each model can be effectively utilized, and finally a better recognition accuracy than a single model can be obtained.

[0107] In an implementation manner, the particle position is converted into a probability distribution by using the Softmax function, so that all weights are non-negative and the sum is 1, avoiding meaningless negative weights or a sum greater than 1, and improving the stability and interpretability of the fusion.

[0108] The embodiment of the present application provides a safety monitoring system for trichloroisocyanuric acid wastewater treatment. Figure 4 , Figure 4 The embodiment of the present application provides a safety monitoring system for trichloroisocyanuric acid wastewater treatment.

[0109] The data acquisition module is configured to acquire a vibration signal of the compressor.

[0110] The preprocessing module is configured to preprocess the vibration signal to obtain a target signal.

[0111] The two-dimensional characterization module is configured to convert the target signal into a two-dimensional time-frequency image.

[0112] The base model decision module is configured to take the two-dimensional time-frequency image as an input of a plurality of pre-trained state recognition models to obtain a prediction probability of a plurality of state categories.

[0113] The integration module is configured to weight and sum the prediction probabilities of the target state category according to a preset integration weight to obtain a probability score of the target state category.

[0114] The result output module is configured to output a state category corresponding to the maximum probability score as a detection result.

[0115] The target state category is any one of the plurality of state categories.

[0116] The safety monitoring system for trichloroisocyanuric acid wastewater treatment provided by the embodiment of the present application can effectively identify early subtle faults by acquiring a compressor vibration signal and converting it into a two-dimensional time-frequency image, using a plurality of pre-trained models for state recognition, and improving prediction accuracy through weighted integration. Compared with the traditional threshold method, this method advances the fault identification time point from the obvious damage stage to the non-obvious damage stage, which helps to provide early warning and reduce the risk of equipment sudden failure.

[0117] In one embodiment, the preprocessing module comprises:

[0118] The segment interception module is configured to extract a first vibration signal with a preset length from the vibration signal.

[0119] The modal decomposition module is configured to perform variational modal decomposition on the first vibration signal to obtain a plurality of intrinsic modal components.

[0120] The correlation calculation module is configured to calculate a Pearson correlation coefficient of each intrinsic modal component with the first vibration signal.

[0121] The signal reconstruction module is configured to take the intrinsic modal components with an absolute value of the correlation coefficient greater than a preset threshold for signal reconstruction to obtain the target signal.

[0122] In one embodiment, the system further comprises an integrated optimization module for optimizing and determining the integrated weight by using the particle swarm optimization algorithm after the plurality of state recognition model training is completed. The operation process of the integrated optimization module is consistent with the process (steps one to seven) of optimizing and determining the integrated weight by using the particle swarm optimization algorithm as recorded in the foregoing, and will not be described again.

[0123] The above is only a preferred embodiment of the present application, and does not limit the present application in any form. Although the present application has been disclosed as above with a preferred embodiment, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content to obtain equivalent embodiments with equivalent changes, without departing from the technical solution of the present application. Any brief modification, equivalent change and modification of the above embodiments made according to the technical essence of the present application, without departing from the technical solution of the present application, still belongs to the scope of the technical solution of the present application.

Claims

1. A safety monitoring method for the treatment of trichloroisocyanuric acid wastewater, characterized in that, Wastewater is treated using a mechanical steam recompression process, the method comprising: The vibration signal of the compressor is acquired, and the vibration signal is preprocessed to obtain the target signal; The target signal is converted into a two-dimensional time-frequency image; The two-dimensional time-frequency image is used as input to multiple pre-trained state recognition models to obtain the predicted probabilities of multiple state categories. Based on preset integration weights, the predicted probabilities of the target state category are weighted and summed to obtain the probability score of the target state category; the target state category is any one of multiple state categories. Output the state category corresponding to the highest probability score as the detection result; Multiple state recognition models have the same structure; any state recognition model includes: The initial convolutional module is used to perform shallow feature extraction and spatial compression on the two-dimensional time-frequency image using a single convolutional layer to obtain a first feature map. The feature extraction module is used to perform multi-scale feature extraction on the first feature map using convolutional layers with different kernel sizes to obtain the second feature map; An attention enhancement module is used to enhance the features of the second feature map using an attention mechanism to obtain a third feature map; The shaping module is used to convert the third feature map into a feature vector; The classification module is used to map the feature vector to a probability vector of a preset scale using a fully connected network. The feature extraction module includes two cascaded multi-scale dense blocks; the two multi-scale dense blocks have the same structure, and the operation process of any one of the multi-scale dense blocks includes: Among them, X in This is the input to the multi-scale dense block; Conv represents the convolution operation, with subscripts 3×3, 5×5, 7×7, and 1×1 indicating the kernel size and superscript 1 indicating the stride; concat indicates concatenation along the channel dimension; MaxPool represents max pooling, with subscript 2×2 indicating the pooling window size and superscript 2 indicating the stride; X 11 X 12 X 13 X 14 X 21 X 22 X 23 and X 24 It is a feature map generated during the operation, X out This is the output of the multi-scale dense block; The ensemble weights are determined by the training performance of multiple state recognition models. Specifically, after the multiple state recognition models are trained, the ensemble weights are optimized using a particle swarm optimization algorithm, including: Step 1: Use multiple state recognition models to predict each sample in the validation set to obtain the probability vector corresponding to each sample; Step 2: Initialize the population size, particle positions, and velocities; Step 3: Use the softmax function to convert the particle positions into an effective weight distribution; Step 4: Based on the effective weight distribution, integrate the probability vector of each sample into a probability score to obtain the final prediction result, and calculate its accuracy as the fitness. Step 5: Update the individual optimal position and the global optimal position based on the fitness value; Step six: Determine if the preset stopping condition has been met; if yes, end the optimization process; otherwise, proceed to step seven. Step 7: Update the particle position based on the individual optimal position and the global optimal position, and return to Step 3.

2. The safety monitoring method for trichloroisocyanuric acid wastewater treatment according to claim 1, characterized in that, The vibration signal is preprocessed to obtain the target signal, including: Extract a first vibration signal of a preset length from the vibration signal; The first vibration signal is subjected to variational mode decomposition to obtain multiple intrinsic mode components; Calculate the Pearson correlation coefficient between each intrinsic modal component and the first vibration signal; The intrinsic mode components whose absolute correlation coefficient is greater than a preset threshold are used for signal reconstruction to obtain the target signal.

3. A safety monitoring system for the treatment of trichloroisocyanuric acid wastewater, used to implement the safety monitoring method for the treatment of trichloroisocyanuric acid wastewater as described in claim 1, characterized in that, The system includes: The data acquisition module is used to acquire the vibration signal of the compressor; The preprocessing module is used to preprocess the vibration signal to obtain the target signal; A two-dimensional representation module is used to convert the target signal into a two-dimensional time-frequency image; The base model decision module is used to take the two-dimensional time-frequency image as input to multiple pre-trained state recognition models to obtain the predicted probabilities of multiple state categories. An integration module is used to sum the predicted probabilities of a target state category according to a preset integration weight to obtain a probability score for the target state category; the target state category is any one of multiple state categories. The results output module is used to output the state category corresponding to the highest probability score as the detection result.

4. A safety monitoring system for trichloroisocyanuric acid wastewater treatment according to claim 3, characterized in that, The preprocessing module includes: The segment extraction module is used to extract a first vibration signal of a preset length from the vibration signal; The mode decomposition module is used to perform variational mode decomposition on the first vibration signal to obtain multiple intrinsic mode components. The correlation calculation module is used to calculate the Pearson correlation coefficient between each intrinsic mode component and the first vibration signal; The signal reconstruction module is used to reconstruct the target signal by taking the intrinsic mode components whose absolute correlation coefficient is greater than a preset threshold.

5. A safety monitoring system for trichloroisocyanuric acid wastewater treatment according to claim 3, characterized in that, Multiple state recognition models have the same structure; Any state recognition model includes: The initial convolutional module is used to perform shallow feature extraction and spatial compression on the two-dimensional time-frequency image using a single convolutional layer to obtain a first feature map. The feature extraction module is used to perform multi-scale feature extraction on the first feature map using convolutional layers with different kernel sizes to obtain the second feature map; An attention enhancement module is used to enhance the features of the second feature map using an attention mechanism to obtain a third feature map; The shaping module is used to convert the third feature map into a feature vector; The classification module is used to map the feature vector to a probability vector of a preset scale using a fully connected network.

6. A safety monitoring system for trichloroisocyanuric acid wastewater treatment according to claim 5, characterized in that, The feature extraction module includes two cascaded multi-scale dense blocks; Two multi-scale dense blocks have the same structure. The computation process for any multi-scale dense block includes: Among them, X in This is the input to the multi-scale dense block; Conv represents the convolution operation, with subscripts 3×3, 5×5, 7×7, and 1×1 indicating the kernel size and superscript 1 indicating the stride; concat indicates concatenation along the channel dimension; MaxPool represents max pooling, with subscript 2×2 indicating the pooling window size and superscript 2 indicating the stride; X 11 X 12 X 13 X 14 X 21 X 22 X 23 and X 24 It is a feature map generated during the operation, X out This is the output of the multi-scale dense block.

7. A safety monitoring system for trichloroisocyanuric acid wastewater treatment according to claim 3, characterized in that, The system also includes an ensemble optimization module, used to optimize and determine the ensemble weights using a particle swarm optimization algorithm after multiple state recognition models have been trained; the operation process of the ensemble optimization module includes: Step 1: Use multiple state recognition models to predict each sample in the validation set to obtain the probability vector corresponding to each sample; Step 2: Initialize the population size, particle positions, and velocities; Step 3: Use the softmax function to convert the particle positions into an effective weight distribution; Step 4: Based on the effective weight distribution, integrate the probability vector of each sample into a probability score to obtain the final prediction result, and calculate its accuracy as the fitness. Step 5: Update the individual optimal position and the global optimal position based on the fitness value; Step six: Determine if the preset stopping condition has been met; if yes, end the optimization process; otherwise, proceed to step seven. Step 7: Update the particle position based on the individual optimal position and the global optimal position, and return to Step 3.

Images