Circuit breaker intelligent detection method and system and Internet of Things switch cabinet monitoring system

Abstract

The invention discloses a circuit breaker intelligent detection method and system and an Internet of Things switch cabinet monitoring system, and the method comprises the steps: forming a triangular fuzzy number according to input sensor parameters, and obtaining the safety level of a circuit breaker; the CAN field bus and the Internet of Things communication technology are combined, remote transmission of parameters of a short circuit breaker and a switch cabinet is achieved, working condition parameters of equipment in the circuit breaker and the switch cabinet are monitored in real time through various sensing equipment, the safety level of the circuit breaker is predicted, the monitoring intelligent level of the circuit breaker and the switch cabinet is effectively improved, and the safety of the circuit breaker and the switch cabinet is guaranteed. Accurate and comprehensive unmanned monitoring of the switch cabinet of the power supply system is realized.

Description

Technical Field

[0001] The present invention relates to automated equipment for circuit breaker detection and switch cabinet monitoring, in particular to a circuit breaker intelligent detection method and system and an Internet of Things switch cabinet monitoring system. Background Art

[0002] The operation level and safety and reliability of circuit breakers and high-voltage switchgear are crucial to ensuring the normal power supply of the power supply system. Since circuit breakers and high-voltage switchgear are in long-term operation and are affected by power supply load fluctuations and harsh working environment factors, the electrical connection parts of equipment such as circuit breakers and disconnectors in the switchgear are very likely to age and be damaged due to excessive temperature rise, collision and extrusion, resulting in equipment leakage, short circuit and circuit breaker failure to open and close normally, which seriously affects the safety and efficiency of power supply production. At present, the monitoring reliability of circuit breakers and high-voltage switchgear is low, the accuracy of circuit breaker and switchgear parameter processing is low, and there is no accurate warning of circuit breaker status level, which cannot effectively ensure their normal and safe operation. Summary of the invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a circuit breaker intelligent detection method, system and Internet of Things switch cabinet monitoring system, which adopts the combination of CAN field bus and Internet of Things communication technology to realize the remote transmission of circuit breaker and switch cabinet parameters, monitor the operating parameters of circuit breakers and equipment in the switch cabinet in real time through a variety of sensor devices, and apply circuit breaker safety intelligent detection methods to realize circuit breaker safety level warning, effectively improve the intelligence level of circuit breaker and switch cabinet monitoring, and realize accurate and comprehensive unmanned monitoring of switch cabinets in power supply systems.

[0004] Technical solution: The circuit breaker intelligent detection method of the present invention includes a circuit breaker state detection module, which obtains the safety level of the circuit breaker according to the input sensor parameters; the circuit breaker state detection module includes a circuit breaker parameter processing module, a first model to a twelfth model;

[0005] The circuit breaker parameter processing module obtains parameter prediction values ​​according to the input sensor parameters through the neural network model, including the temperature prediction value, displacement prediction value and current prediction value of the circuit breaker;

[0006] The temperature prediction value, the displacement prediction value and the current prediction value are input into the first model, the second model and the third model respectively;

[0007] The output of the first model is used as the input of the fourth model, the fifth model and the sixth model; the output of the fifth model is used as the input of the sixth model; the output of the sixth model is used as the input of the first model and the seventh model;

[0008] The output of the second model is used as the input of the fourth model, the seventh model and the eighth model, the output of the seventh model is used as the input of the eighth model; the output of the eighth model is used as the input of the second model and the ninth model;

[0009] The output of the third model is used as the input of the fourth model, the ninth model and the tenth model, the output of the ninth model is used as the input of the tenth model; the output of the tenth model is used as the input of the third model and the fifth model;

[0010] The output of the fourth model is used as the input of the eleventh model and the twelfth model; the output of the eleventh model is used as the input of the twelfth model, the fifth model, the seventh model and the ninth model; the output of the twelfth model is used as the input of the fourth model, the sixth model, the eighth model and the tenth model;

[0011] The outputs of the sixth model, the eighth model and the tenth model constitute triangular fuzzy numbers, and the safety level of the circuit breaker is obtained according to the triangular fuzzy numbers;

[0012] Among them, the first model is a BiTCN neural network model and a Transformer neural network model connected in series, the second model is a BiGRU neural network model and a Transformer neural network model connected in series; the third model is a GCN graph neural network model and a Transformer neural network model connected in series; the fourth model is a NARX neural network model and a Transformer neural network model connected in series; the fifth model, the seventh model, the ninth model and the eleventh model are all NARX neural network models and SARIMA models connected in series; the sixth model, the eighth model, the tenth model and the twelfth model are all the integration of several GCN graph neural network models using the Adaboost model.

[0013] Further, the circuit breaker parameter processing module includes a thirteenth model to a twenty-first model;

[0014] The sensor parameters are input into the thirteenth model and the sixteenth model; the output of the thirteenth model is used as the input into the fifteenth model, the seventeenth model and the eighteenth model; the output of the fifteenth model is used as the input into the fourteenth model, the seventeenth model and the eighteenth model; the output of the fourteenth model is used as the input into the sixteenth model, the seventeenth model and the eighteenth model; the output of the sixteenth model is used as the input into the seventeenth model and the eighteenth model;

[0015] The output of the seventeenth model is used as the input of the nineteenth model, the twentieth model and the twenty-first model, the output of the eighteenth model is used as the input of the nineteenth model, the twentieth model and the twenty-first model, and the output of the nineteenth model and the twenty-first model are both used as the input of the twenty-first model;

[0016] The output of the twenty-first model is the parameter prediction value corresponding to the input sensor parameter;

[0017] Among them, the thirteenth model and the fourteenth model are both the BiTCN neural network model and the Transformer neural network model in series; the fifteenth model and the twentieth model are both the NARX neural network model and the Transformer neural network model in series; the sixteenth model and the nineteenth model are both the NARX neural network model and the SARIMA model in series; the seventeenth model is the BiGRU neural network model and the Transformer neural network model in series; the eighteenth model is the GCN graph neural network model and the Transformer neural network model in series; the twenty-first model is the integration of several GCN graph neural network models using the Adaboost model.

[0018] Furthermore, the outputs of the sixth model, the eighth model and the tenth model constitute triangular fuzzy numbers, and obtaining the safety level of the circuit breaker according to the triangular fuzzy numbers includes: dividing the safety level of the circuit breaker into very safe, relatively safe, safe, unsafe and very unsafe according to the triangular fuzzy numbers.

[0019] Furthermore, the outputs of the sixth model, the eighth model and the tenth model constitute a triangular fuzzy number<x,y,z> , the corresponding relationship between the triangular fuzzy number and the safety level of the circuit breaker is:

[0020] The triangular fuzzy number <0.7, 0.8, 0.9>, the safety level of the circuit breaker is very safe;

[0021] The triangular fuzzy number <0.6, 0.7, 0.8>, the safety level of the circuit breaker is relatively safe;

[0022] For triangular fuzzy numbers <0.5, 0.6, 0.7>, the safety level of the circuit breaker is safe;

[0023] The triangular fuzzy number <0.4, 0.5, 0.6>, the safety level of the circuit breaker is unsafe;

[0024] The triangular fuzzy number is <0.3, 0.4, 0.5>, and the safety level of the circuit breaker is very unsafe.

[0025] The circuit breaker intelligent detection system of the present invention comprises a circuit breaker state detection module, which is used to obtain the safety level of the circuit breaker according to the input sensor parameters; the circuit breaker state detection module comprises a circuit breaker parameter processing module, a first model to a twelfth model;

[0026] The circuit breaker parameter processing module obtains parameter prediction values ​​according to the input sensor parameters through the neural network model, including the temperature prediction value, displacement prediction value and current prediction value of the circuit breaker;

[0027] The temperature prediction value, the displacement prediction value and the current prediction value are input into the first model, the second model and the third model respectively;

[0028] The output of the first model is used as the input of the fourth model, the fifth model and the sixth model; the output of the fifth model is used as the input of the sixth model; the output of the sixth model is used as the input of the first model and the seventh model;

[0029] The output of the second model is used as the input of the fourth model, the seventh model and the eighth model, the output of the seventh model is used as the input of the eighth model; the output of the eighth model is used as the input of the second model and the ninth model;

[0030] The output of the third model is used as the input of the fourth model, the ninth model and the tenth model, the output of the ninth model is used as the input of the tenth model; the output of the tenth model is used as the input of the third model and the fifth model;

[0031] The output of the fourth model is used as the input of the eleventh model and the twelfth model; the output of the eleventh model is used as the input of the twelfth model, the fifth model, the seventh model and the ninth model; the output of the twelfth model is used as the input of the fourth model, the sixth model, the eighth model and the tenth model;

[0032] The outputs of the sixth model, the eighth model and the tenth model constitute triangular fuzzy numbers, and the safety level of the circuit breaker is obtained according to the triangular fuzzy numbers;

[0033] Among them, the first model is a BiTCN neural network model and a Transformer neural network model connected in series, the second model is a BiGRU neural network model and a Transformer neural network model connected in series; the third model is a GCN graph neural network model and a Transformer neural network model connected in series; the fourth model is a NARX neural network model and a Transformer neural network model connected in series; the fifth model, the seventh model, the ninth model and the eleventh model are all NARX neural network models and SARIMA models connected in series; the sixth model, the eighth model, the tenth model and the twelfth model are all the integration of several GCN graph neural network models using the Adaboost model.

[0034] Further, the circuit breaker parameter processing module includes a thirteenth model to a twenty-first model;

[0035] The sensor parameters are input into the thirteenth model and the sixteenth model; the output of the thirteenth model is used as the input into the fifteenth model, the seventeenth model and the eighteenth model; the output of the fifteenth model is used as the input into the fourteenth model, the seventeenth model and the eighteenth model; the output of the fourteenth model is used as the input into the sixteenth model, the seventeenth model and the eighteenth model; the output of the sixteenth model is used as the input into the seventeenth model and the eighteenth model;

[0036] The output of the seventeenth model is used as the input of the nineteenth model, the twentieth model and the twenty-first model, the output of the eighteenth model is used as the input of the nineteenth model, the twentieth model and the twenty-first model, and the output of the nineteenth model and the twenty-first model are both used as the input of the twenty-first model;

[0037] The output of the twenty-first model is the parameter prediction value corresponding to the input sensor parameter;

[0038] Among them, the thirteenth model and the fourteenth model are both the BiTCN neural network model and the Transformer neural network model in series; the fifteenth model and the twentieth model are both the NARX neural network model and the Transformer neural network model in series; the sixteenth model and the nineteenth model are both the NARX neural network model and the SARIMA model in series; the seventeenth model is the BiGRU neural network model and the Transformer neural network model in series; the eighteenth model is the GCN graph neural network model and the Transformer neural network model in series; the twenty-first model is the integration of several GCN graph neural network models using the Adaboost model.

[0039] Furthermore, the outputs of the sixth model, the eighth model and the tenth model constitute triangular fuzzy numbers, and obtaining the safety level of the circuit breaker according to the triangular fuzzy numbers includes: dividing the safety level of the circuit breaker into very safe, relatively safe, safe, unsafe and very unsafe according to the triangular fuzzy numbers.

[0040] Furthermore, the outputs of the sixth model, the eighth model and the tenth model constitute a triangular fuzzy number<x,y,z> , the corresponding relationship between the triangular fuzzy number and the safety level of the circuit breaker is:

[0041] The triangular fuzzy number <0.7, 0.8, 0.9>, the safety level of the circuit breaker is very safe;

[0042] The triangular fuzzy number <0.6, 0.7, 0.8>, the safety level of the circuit breaker is relatively safe;

[0043] For triangular fuzzy numbers <0.5, 0.6, 0.7>, the safety level of the circuit breaker is safe;

[0044] The triangular fuzzy number <0.4, 0.5, 0.6>, the safety level of the circuit breaker is unsafe;

[0045] The triangular fuzzy number is <0.3, 0.4, 0.5>, and the safety level of the circuit breaker is very unsafe.

[0046] The present invention stores a computer program, characterized in that the computer program implements the circuit breaker intelligent detection method when executed by a processor.

[0047] The IoT switch cabinet monitoring system of the present invention comprises a field acquisition layer, a communication network layer and a remote monitoring layer;

[0048] The field acquisition layer includes a switch cabinet information acquisition module, which includes sensors arranged in the switch cabinet and their corresponding signal conditioning circuits, STM32 microprocessors and communication interfaces, and is used to collect switch cabinet equipment parameters, machine vision and micro-meteorological information parameters in real time, and convert sensor analog signals into digital signals for output;

[0049] The communication network layer is composed of the CAN field bus interface of the switch cabinet information acquisition module and the switch cabinet information concentrator, which is used for parameter acquisition of the switch cabinet and control command transmission of the remote monitoring layer; the switch cabinet information concentrator includes a communication module, an STM32 microcontroller and a CAN communication interface;

[0050] The remote monitoring layer includes the switch cabinet cloud platform, the switch cabinet remote monitoring terminal and the switch cabinet mobile APP terminal. The switch cabinet cloud platform realizes the collection, storage and management of switch cabinet equipment parameters. The switch cabinet remote monitoring terminal realizes the detection of switch cabinet parameters and switch cabinet safety monitoring. The switch cabinet mobile APP terminal monitors and remotely controls the switch cabinet equipment parameters.

[0051] Beneficial effects: Compared with the prior art, the advantages of the present invention are:

[0052] 1. The BiTCN neural network model is a bidirectional temporal convolutional neural network model, which realizes the effective extraction of bidirectional features of the parameters of the detected circuit breaker. The BiTCN neural network model can more effectively extract the deeper and more abstract temporal features of the circuit breaker parameters of the detected circuit breaker through stacked TCN residual blocks. Comparative experiments show that the BiTCN neural network model achieves better recognition performance and robustness of the detected circuit breaker.

[0053] 2. The BiTCN neural network model is used based on the structural foundation of the TCN neural network model. The TCN neural network model is constructed into a forward network and a reverse network. The data time series characteristics of the detected circuit breaker can be extracted from both the forward sequence and the reverse sequence. The fused data are added in an additive manner to obtain the fused bidirectional time series characteristics. The fused bidirectional time series characteristics can maximize the expression of the deep and global feature information of the data set of the detected circuit breaker. The time series information of the bidirectional sequence of the circuit breaker parameters is extracted by using the BiTCN stacked residual block for feature fusion to obtain the deep circuit breaker parameter characteristics, which effectively improves the performance of detecting circuit breaker parameters.

[0054] 3. The Transformer neural network model is a neural network structure that can effectively integrate the long sequence information of circuit breaker parameters. It encodes the correlation between the sequence information of circuit breaker parameters through the self-attention mechanism. The self-attention mechanism is a mechanism that can effectively integrate the detected circuit breaker information over a long period of time. It does not require the introduction of a recurrent neural network structure, but uses the self-attention calculation of the circuit breaker data to obtain the relationship between the sequences. The self-attention mechanism abandons the recursive structure in RNN and LSTM, directly connects the input and output of the circuit breaker parameters, and effectively solves the problems of gradient explosion and gradient disappearance, thus having a more significant effect in processing the long-distance dependency of the time series of circuit breaker parameters. The performance of the self-attention mechanism is significantly improved compared to the traditional recurrent neural network and long short-term memory model.

[0055] 4. The NARX neural network model includes a three-layer neural network model with 1 input layer, 1 hidden layer and 1 output layer, among which the time delay layer (TimeDelayLayer, TDL) is the input layer, which means that the input circuit breaker data is delayed for several sampling cycles before participating in the operation of the NARX neural network model. The NARX neural network model is in an open-loop state during training, and in actual use, the NARX neural network model is in a closed-loop state. The circuit breaker parameter prediction value output by the NARX neural network model at time k is fed back to the input end of the network NARX neural network model at time k+1 as the corresponding TDL input. At this moment, the output of the NARX neural network model can be used as the input of the NARX neural network model at the next moment, thereby enhancing the NARX neural network model's ability to remember the historical parameters of the circuit breaker. Compared with traditional neural networks, it can remember the historical and real-time status information of the circuit breaker parameters for a longer period of time.

[0056] 5. The BiGRU neural network model can simultaneously consider the historical and future data of the circuit breaker through the bidirectional GRU neural network model. The BiGRU neural network model is used to effectively process the timing of the circuit breaker parameters, while considering the forward and reverse information of the input circuit breaker parameter time series. It usually has stronger generalization ability for circuit breaker parameter time series data of different lengths and structures, considering the past and future circuit breaker time series information, so as to more comprehensively analyze the changes in short-term circuit breaker parameters and improve the accuracy of predicting circuit breaker parameters.

[0057] 6. The GCN graph neural network model uses multi-layer graph convolution operations to enable each node of the circuit breaker parameters to aggregate more extensive neighboring circuit breaker parameter node information, thereby realizing the hierarchical representation of the node features of the circuit breaker parameters and learning of complex relationships. The GCN graph neural network model can perform tasks such as node link prediction and graph representation learning of circuit breaker parameters through the node feature vectors of the circuit breaker parameters, thereby improving the accuracy, robustness and reliability of predicting circuit breaker parameters.

[0058] 7. The GCN graph neural network model-AdaBoost model is an integrated learning method that integrates the AdaBoost model with the GCN graph neural network model. A strong GCN graph neural network model is formed by combining multiple weak GCN graph neural network models. In each round of iteration, the Adaboost model adjusts the sample weights of each GCN graph neural network model output according to the output results of the previous GCN graph neural network model, thereby improving the overall performance of the GCN graph neural network model-AdaBoost model in predicting circuit breaker parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a structural diagram of a circuit breaker parameter processing module of the present invention;

[0060] Figure 2 It is a structural diagram of a circuit breaker state detection module of the present invention;

[0061] Figure 3 This is the structure diagram of the Internet of Things switch cabinet monitoring system of the present invention

[0062] Figure 4 This is a structural diagram of the switch cabinet information acquisition module of the present invention;

[0063] Figure 5 It is a structural diagram of the switch cabinet information concentrator of the present invention;

[0064] Figure 6 This is a functional diagram of the switch cabinet remote monitoring terminal software of the present invention. DETAILED DESCRIPTION

[0065] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0066] 1. The circuit breaker intelligent detection method described in the present invention inputs sensor parameters into the circuit breaker state detection module to obtain the circuit breaker safety level. The circuit breaker state detection module design includes the following contents.

[0067] The circuit breaker status detection module includes circuit breaker parameter processing modules 1-3, BiTCN neural network model-Transformer neural network model 3, BiGRU neural network model-Transformer neural network model 2, GCN graph neural network model-Transformer neural network model 2, NARX neural network model-Transformer neural network model 3, NARX neural network model-SARIMA model 3-6 and GCN graph neural network model-AdaBoost model 2-5.

[0068] Among them, Figure 1 As shown, the circuit breaker parameter processing module includes BiTCN neural network model-Transformer neural network model 1-2, NARX neural network model-Transformer neural network model 1-2, BiGRU neural network model-Transformer neural network model 1, GCN graph neural network model-Transformer neural network model 1, NARX neural network model-SARIMA model 1-2 and GCN graph neural network model-AdaBoost model 1.

[0069] The parameter sensor outputs are respectively used as inputs of the BiTCN neural network model-Transformer neural network model 1-2, the NARX neural network model-Transformer neural network model 1 and the NARX neural network model-SARIMA model 1; the output of the BiTCN neural network model-Transformer neural network model 1 is used as the input of the NARX neural network model-Transformer neural network model 1, the output of the NARX neural network model-Transformer neural network model 1 is used as the input of the BiTCN neural network model-Transformer neural network model 2, and the output of the BiTCN neural network model-Transformer neural network model 2 is used as the input of the NARX neural network model-SARIMA model 1.

[0070] The outputs of the BiTCN neural network model-Transformer neural network model 1-2, the NARX neural network model-Transformer neural network model 1 and the NARX neural network model-SARIMA model 1 are respectively used as inputs of the BiGRU neural network model-Transformer neural network model 1 and the GCN graph neural network model-Transformer neural network model 1.

[0071] The outputs of the BiGRU neural network model-Transformer neural network model 1 and the GCN graph neural network model-Transformer neural network model 1 are respectively used as the inputs of the NARX neural network model-SARIMA model 2, the NARX neural network model-Transformer neural network model 2 and the GCN graph neural network model-AdaBoost model 1, and the outputs of the NARX neural network model-SARIMA model 2 and the NARX neural network model-Transformer neural network model 2 are used as the inputs of the GCN graph neural network model-AdaBoost model 1.

[0072] The output of GCN graph neural network model-AdaBoost model 1 is used as the parameter sensor prediction value.

[0073] like Figure 2 As shown, the outputs of the temperature sensor, displacement sensor and current sensor are respectively used as inputs of the circuit breaker parameter processing modules 1-3, and the outputs of the circuit breaker parameter processing modules 1-3 are respectively used as corresponding inputs of the BiTCN neural network model-Transformer neural network model 3, the BiGRU neural network model-Transformer neural network model 2 and the GCN graph neural network model-Transformer neural network model 2.

[0074] The outputs of the BiTCN neural network model-Transformer neural network model 3 are respectively used as the inputs of the NARX neural network model-SARIMA model 3, the NARX neural network model-Transformer neural network model 3 and the GCN graph neural network model-AdaBoost model 2; the outputs of the NARX neural network model-SARIMA model 3 are used as the inputs of the GCN graph neural network model-AdaBoost model 2, and the outputs of the GCN graph neural network model-AdaBoost model 2 are respectively used as the inputs of the BiTCN neural network model-Transformer neural network model 3 and the NARX neural network model-SARIMA model 4.

[0075] The outputs of the BiGRU neural network model-Transformer neural network model 2 are respectively used as the inputs of the NARX neural network model-SARIMA model 4, the NARX neural network model-Transformer neural network model 3 and the GCN graph neural network model-AdaBoost model 3; the outputs of the NARX neural network model-SARIMA model 4 are used as the inputs of the GCN graph neural network model-AdaBoost model 3; and the outputs of the GCN graph neural network model-AdaBoost model 3 are respectively used as the inputs of the BiGRU neural network model-Transformer neural network model 2 and the NARX neural network model-SARIMA model 5.

[0076] The outputs of the GCN graph neural network model-Transformer neural network model 2 are respectively used as the input of the NARX neural network model-SARIMA model 5, the NARX neural network model-Transformer neural network model 3 and the GCN graph neural network model-AdaBoost model 4, the outputs of the NARX neural network model-SARIMA model 5 are used as the input of the GCN graph neural network model-AdaBoost model 4, and the outputs of the GCN graph neural network model-AdaBoost model 4 are respectively used as the input of the GCN graph neural network model-Transformer neural network model 2 and the NARX neural network model-SARIMA model 3.

[0077] The outputs of NARX neural network model-Transformer neural network model 3 are used as inputs of NARX neural network model-SARIMA model 6 and GCN graph neural network model-AdaBoost model 5 respectively, and the outputs of NARX neural network model-SARIMA model 6 are used as inputs of NARX neural network model-SARIMA model 3, NARX neural network model-SARIMA model 4, NARX neural network model-SARIMA model 5 and GCN graph neural network model-AdaBoost model 5.

[0078] The outputs of GCN graph neural network model-AdaBoost model 5 are respectively used as inputs of GCN graph neural network model-AdaBoost model 2, GCN graph neural network model-AdaBoost model 3, GCN graph neural network model-AdaBoost model 4 and NARX neural network model-Transformer neural network model 3;

[0079] GCN graph neural network model - AdaBoost model 2-4 outputs X, Y and Z constitute three parameters,<X,Y,Z> The safety levels of the tested circuit breakers are divided into five safety levels: very safe, safe, relatively safe, unsafe and very unsafe, as shown in Table 1.

[0080] Table 1 Correspondence between circuit breaker safety level and triangular fuzzy number

[0081] Serial number Circuit breaker safety level Triangular fuzzy numbers 1 Very safe <0.7，0.8，0.9> 2 Safer <0.6，0.7，0.8> 3 Safety <0.5，0.6，0.7> 4 Unsafe <0.4，0.5，0.6> 5 Very unsafe <0.3，0.4，0.5>

[0082] The design methods of each model in the circuit breaker status detection module and the circuit breaker parameter processing module are as follows:

[0083] 1. BiTCN neural network model-Transformer neural network model design.

[0084] The BiTCN neural network model-Transformer neural network model is a series connection of the BiTCN neural network model and the Transformer neural network model.

[0085] The BiTCN neural network model mainly consists of three parts: causal convolution, dilated convolution and residual connection. It is a process of dilated causal convolution of the detected circuit breaker information. The framework layer of the TCN neural network model is composed of residual blocks stacked with exponentially increasing dilation coefficients. Each layer of the residual block retains the important characteristics of the circuit breaker time series information, significantly improving the convolution effect of the circuit breaker time series. The TCN neural network model has the ability to extract the local and global time series information of the detected circuit breaker by introducing causal convolution and dilated convolution; dilated convolution is a one-dimensional convolution. By sampling the parameters of the detected circuit breaker at intervals, the expansion of its receptive field increases exponentially with the dilation coefficient, that is, it can use fewer network layers to obtain a larger receptive field of the detected circuit breaker information; using the dilation coefficient to sample the parameters of the detected circuit breaker at intervals can significantly reduce the complexity of the BiTCN neural network model structure and the amount of calculation of the model. The introduction of dilated convolution makes the TCN neural network model more flexible in receptive field and can learn more comprehensive time series information characteristics of the detected circuit breaker parameters. The BiTCN neural network model introduces residual connections to add the original features to the features extracted by the model, and then performs short-circuit connections. This not only retains the important original features in the process of circuit breaker information extraction, avoids the occurrence of gradient vanishing or overfitting problems, but also improves the stability and depth of the BiTCN neural network model in detecting circuit breaker parameters. The BiTCN neural network model is a bidirectional time convolutional neural network that obtains bidirectional information on the time series changes of the detected circuit breaker, realizes the complementary advantages of the forward and reverse features of the circuit breaker information, obtains circuit breaker information fusion features with better separability and robustness, and improves the effectiveness of the characteristic parameters of the detected circuit breaker information through bidirectional fusion.

[0086] The Transformer neural network model adopts an encoding-decoding structure and uses an attention mechanism to extract the features of the detected circuit breaker parameters. It supports the parallel processing of the time series feature parameters of the input circuit breaker parameters, which brings faster training speed. At the same time, it has a stronger long-term dependency modeling capability and can capture the dependency between different positions in the long time series of the detected circuit breaker parameters, thus improving the prediction effect of the long time series of circuit breaker parameters. The Transformer neural network model is a neural network structure that can effectively integrate the long sequence information of circuit breaker parameters. It encodes the correlation between the circuit breaker parameter sequence information through the self-attention mechanism. The self-attention mechanism is a mechanism that can effectively integrate the detected circuit breaker information over a long period of time. It does not need to introduce a recurrent neural network structure, but uses the self-attention calculation of the circuit breaker data to obtain the relationship between the sequences. The self-attention mechanism abandons the recursive structure in RNN and LSTM, directly connects the input and output of the circuit breaker parameters, effectively solves the gradient explosion and gradient vanishing problems of the circuit breaker parameters, and thus has a more significant effect in processing the long-distance dependency of the time series of circuit breaker parameters. The self-attention mechanism has significantly improved the performance of traditional recurrent neural networks and long short-term memory models. The dot product attention mechanism is the core mechanism of the Transformer neural network model. Suppose the input vector is linearly transformed to obtain the query matrix Q, key matrix K and value matrix V, where L is the sequence length and d is the embedding dimension. The attention score is calculated as:

[0087]

[0088] In the Transformer neural network model, the multi-head self-attention mechanism is actually used, which uses a linear layer to map queries, keys, and values ​​to multiple independent attention operations, and finally connects the results before outputting. Multi-head attention enables the Transformer neural network model to have the ability to learn the correlation of various types of parameters of the circuit breaker parameters. The Transformer neural network model is composed of multiple stacked layers, each of which consists of two modules: a multi-head self-attention layer and a multi-layer perceptron. The output of the l-th self-attention layer is denoted as Y l , and the multi-head self-attention operation is MHA, and the layer normalization operation is LN. Then the calculation of the multi-head self-attention submodule is:

[0089] Y l =LN(E l-1 +MHA(E l-1 )) (2)

[0090] The multilayer perceptron submodule actually applies a convolutional network with a kernel size and a step size of l to each step of the input circuit breaker parameter time series to produce a new embedding vector E l , let f l is the calculation of the multi-layer perceptron, then the calculation of the multi-layer perceptron submodule is:

[0091] E l =LN(Y l +f l (Y l )) (3)

[0092] 2. NARX neural network model-Transformer neural network model design.

[0093] The NARX neural network model-Transformer neural network model is a series connection of the NARX neural network model and the Transformer neural network model.

[0094] The NARX neural network model is a recurrent neural network, which adds the input delay and output feedback mechanism of the circuit breaker parameters, and enhances the memory capacity of the circuit breaker historical data. Therefore, the NARX neural network model is suitable for the time series prediction of the circuit breaker parameters. In view of the nonlinear characteristics and time-varying characteristics of the secondary channel, the NARX neural network model consists of three parts: input layer, hidden layer and output layer. The neurons between two adjacent layers are connected to each other through different weights. Considering the real-time and feasibility of the NARX neural network model, the present invention only considers a neural network containing one hidden layer. The NARX neural network model includes a three-layer neural network model of one input layer, one hidden layer and one output layer, wherein the time delay layer (TimeDelayLayer, TDL) is the input layer, indicating that the input circuit breaker data is delayed for several sampling cycles before participating in the operation of the NARX neural network model. The NARX neural network model is in an open-loop state during training, and the gradient descent method is used to train the parameters of the NARX neural network model, and its training method is the same as that of the BP neural network. In actual use, the NARX neural network model is in a closed-loop state. The predicted value of the circuit breaker parameters output by the NARX neural network model at time k is fed back to the input of the network NARX neural network model at time k+1 as the corresponding TDL input. The output of the NARX neural network model at this moment can be used as the input of the NARX neural network model at the next moment, thereby enhancing the NARX neural network model's ability to remember the historical parameters of the circuit breaker. Compared with traditional neural networks, it can remember the historical and real-time status information of the circuit breaker parameters for a longer period of time. The nonlinear autoregressive (NARX) neural network model with exogenous input is often used to describe nonlinear discrete systems. u(t) and y(t) represent the input and output of the circuit breaker parameters of the NARX neural network model at time t. D y represents the maximum time delay order of the output of the circuit breaker parameter prediction value, and f(·) represents the nonlinear function fitted by the NARX neural network model:

[0095] y(t)=f{u(t),y(tD y ),…,y(t-1)} (4)

[0096] The design method of the Transformer neural network model refers to the BiTCN neural network model-Transformer neural network model design.

[0097] 3. BiGRU neural network model-Transformer neural network model design.

[0098] The BiGRU neural network model-Transformer neural network model is a series connection of the BiGRU neural network model and the Transformer neural network model.

[0099] Although GRU performs well in processing time series circuit breaker data, it may be difficult to accurately capture the key factors of the current circuit breaker parameter changes in some cases. In order to overcome this limitation, the BiGRU neural network model can simultaneously consider the forward and reverse information of the input circuit breaker parameter time series. It usually has stronger generalization ability for circuit breaker parameter time series data of different lengths and structures, which helps to improve the accuracy of feature vector extraction of circuit breaker parameters. The bidirectional GRU neural network model simultaneously considers the historical and future data of the circuit breaker, so as to more comprehensively analyze the changes in short-term circuit breaker parameters and improve the accuracy of predicting circuit breaker parameters. The BiGRU neural network model is used to effectively process the time series of circuit breaker parameters, consider the past and future circuit breaker time series information, and improve the accuracy of the BiGRU neural network model in predicting circuit breaker parameters. Bidirectional Gated Recurrent Unit Structure (BiGRU Neural Network Model) In the traditional unidirectional neural network structure, the state is output sequentially from front to back. The Bidirectional Gated Recurrent Unit (BiGRU Neural Network Model) is an improved recurrent neural network (RNN) structure, which consists of two independent GRU neural network models, one for processing the circuit breaker parameter time series in the forward direction, and the other for processing the circuit breaker parameter time series in the reverse direction. The output of the circuit breaker parameter prediction value is jointly determined by the states of these two GRU neural network models. At each moment, the input of the circuit breaker parameter will simultaneously provide two GRU neural network models in opposite directions, and the circuit breaker parameter prediction output is jointly determined by these two unidirectional GRU neural network models. This bidirectional structure enables the BiGRU neural network model to capture the long-distance dependencies in the time series of the circuit breaker parameters, thereby improving the prediction accuracy of the BiGRU neural network model.

[0100] The design method of the Transformer neural network model refers to the BiTCN neural network model-Transformer neural network model design.

[0101] 4. GCN graph neural network model-Transformer neural network model design.

[0102] The GCN graph neural network model-Transformer neural network model is a series connection of the GCN graph neural network model and the Transformer neural network model.

[0103] The GCN graph neural network model is a graph neural network model used to process graph structure data of circuit breaker parameters. The GCN graph neural network model performs information propagation and feature learning on the graph data of circuit breakers through convolution operations, so that each node can aggregate the circuit breaker parameter information of its neighboring nodes, thereby realizing tasks such as learning the graph data features of circuit breakers and learning complex relationships. The core idea of ​​the GCN graph neural network model is based on graph convolution operations. For a graph of circuit breaker parameters containing N nodes, the GCN graph neural network model introduces a learnable circuit breaker parameter matrix to linearly transform the features of the circuit breaker parameter nodes. By weighted summation of the features of the neighboring nodes of the circuit breaker parameters, each circuit breaker parameter node obtains an updated feature representation. Through multi-layer graph convolution operations, the GCN graph neural network model enables each circuit breaker parameter node to aggregate a wider range of neighboring circuit breaker parameter node information, thereby realizing the hierarchical representation of the node features of the circuit breaker parameters and learning of complex relationships. The GCN graph neural network model can perform tasks such as node link prediction and graph representation learning of circuit breaker parameters through the node feature vectors of circuit breaker parameters, thereby improving the accuracy, robustness and reliability of predicting circuit breaker parameters.

[0104] The design method of the Transformer neural network model refers to the BiTCN neural network model-Transformer neural network model design.

[0105] 5. NARX neural network model-SARIMA model design

[0106] The NARX neural network model-SARIMA model is a series connection of the NARX neural network model and the SARIMA model.

[0107] The SARIMA model is an extension and improvement of the ARIMA model, which includes consideration of seasonal or cyclical factors. Within a cycle, the SARIMA model extracts the correlation between the current data and the previous data; between cycles, it extracts the correlation between the current data and the data at the same time in the previous cycles. Therefore, the SARIMA model is more accurate in fitting the changes in the sequence. The SARIMA model can be expressed as SARIMA(p,d,q)(P,D,Q)s, where (p,d,q) represents the non-seasonal part and (P,D,Q)s represents the seasonal part. The SARIMA model is a traditional statistical forecasting method. For the time series of circuit breaker parameters, the periodicity exists in the seasonal changes (quarterly, monthly, weekly or other inherent factors). This type of time series will show obvious periodicity and is called a seasonal series. As a classic model for seasonal circuit breaker parameter time series analysis, the SARIMA model has a simple analysis process and is therefore widely used.

[0108] The SARIMA model analysis process can be summarized into the following steps: ① Generate the time series of circuit breaker parameters, and draw a time series diagram to intuitively present the changing characteristics of the time series of circuit breaker parameters. ② Data verification, use the time series correlation diagram of the circuit breaker parameters to test whether the time series is a white noise series, and combine the unit root test to determine the stationarity of the time series of circuit breaker parameters. If the data is white noise, it is not suitable for establishing a SARIMA model. When the circuit breaker data is not stable, the circuit breaker data can be stabilized by differential transformation. ③ Model order determination, determine the appropriate order according to the autocorrelation coefficient (ACF) and partial autocorrelation coefficient (PACF) of the model, and estimate the model parameters. ④ Model test, test whether the residual of the model obeys the normal distribution and whether it passes the parameter test. ⑤ Model evaluation, judge the effectiveness of the model by comparing the AIC and BIC values.

[0109] The design method of NARX neural network model refers to NARX neural network model-Transformer neural network model design.

[0110] 6. GCN graph neural network model-AdaBoost model design.

[0111] The GCN graph neural network model-AdaBoost model is an integrated learning method that integrates the AdaBoost model with the GCN graph neural network model. A strong GCN graph neural network model is formed by combining multiple weak GCN graph neural network models. In each round of iteration, the Adaboost model adjusts the sample weights of each GCN graph neural network model output according to the output results of the previous GCN graph neural network model, thereby improving the overall performance of the GCN graph neural network model-AdaBoost model in predicting circuit breaker parameters.

[0112] The basic idea of ​​integrating the Adaboost model with the GCN graph neural network model is to first use the GCN graph neural network model as the basic weak learner prediction model, and then use the Adaboost model algorithm to integrate multiple GCN graph neural network models. In each round of iteration, the Adaboost model adjusts the weight of each GCN graph neural network model sample according to the previous prediction results of the GCN graph neural network model, so that the regression model pays more attention to the GCN graph neural network model samples with large prediction errors, thereby improving the regression performance of the overall GCN graph neural network model.

[0113] Steps of Adaboost integrated GCN graph neural network model: (1) Initialize sample weights. Randomly select m groups of training data in the sample space and initialize the weights of all samples to equal values; (2) Iteratively train the GCN graph neural network model. Train the GCN graph neural network model and predict the output of the tth weak learner to obtain the sum of the prediction errors of the circuit breaker parameter prediction sequence values; (3) Calculate model weights. Determine the output weight of each GCN graph neural network model based on the error sum of the prediction results; (4) Construct the GCN graph neural network model of the strong predictor - the AdaBoost model.

[0114] The design method of the GCN graph neural network model refers to the GCN graph neural network model-Transformer neural network model design.

[0115] 2. The circuit breaker intelligent detection system of the present invention includes a circuit breaker state detection module, which is used to obtain the safety level of the circuit breaker according to the input sensor parameters; and includes the following contents.

[0116] The circuit breaker state detection module includes a circuit breaker parameter processing module and a first model to a twelfth model;

[0117] The circuit breaker parameter processing module obtains parameter prediction values ​​according to the input sensor parameters through the neural network model, including the temperature prediction value, displacement prediction value and current prediction value of the circuit breaker;

[0118] The temperature prediction value, the displacement prediction value and the current prediction value are input into the first model, the second model and the third model respectively;

[0119] The output of the first model is used as the input of the fourth model, the fifth model and the sixth model; the output of the fifth model is used as the input of the sixth model; the output of the sixth model is used as the input of the first model and the seventh model;

[0120] The output of the second model is used as the input of the fourth model, the seventh model and the eighth model, the output of the seventh model is used as the input of the eighth model; the output of the eighth model is used as the input of the second model and the ninth model;

[0121] The output of the third model is used as the input of the fourth model, the ninth model and the tenth model, the output of the ninth model is used as the input of the tenth model; the output of the tenth model is used as the input of the third model and the fifth model;

[0122] The output of the fourth model is used as the input of the eleventh model and the twelfth model; the output of the eleventh model is used as the input of the twelfth model, the fifth model, the seventh model and the ninth model; the output of the twelfth model is used as the input of the fourth model, the sixth model, the eighth model and the tenth model;

[0123] The outputs of the sixth model, the eighth model and the tenth model constitute triangular fuzzy numbers, and the safety level of the circuit breaker is obtained according to the triangular fuzzy numbers;

[0124] Among them, the first model is a BiTCN neural network model and a Transformer neural network model connected in series, the second model is a BiGRU neural network model and a Transformer neural network model connected in series; the third model is a GCN graph neural network model and a Transformer neural network model connected in series; the fourth model is a NARX neural network model and a Transformer neural network model connected in series; the fifth model, the seventh model, the ninth model and the eleventh model are all NARX neural network models and SARIMA models connected in series; the sixth model, the eighth model, the tenth model and the twelfth model are all the integration of several GCN graph neural network models using the Adaboost model.

[0125] The circuit breaker parameter processing module includes a thirteenth model to a twenty-first model;

[0126] The sensor parameters are input into the thirteenth model and the sixteenth model; the output of the thirteenth model is used as the input into the fifteenth model, the seventeenth model and the eighteenth model; the output of the fifteenth model is used as the input into the fourteenth model, the seventeenth model and the eighteenth model; the output of the fourteenth model is used as the input into the sixteenth model, the seventeenth model and the eighteenth model; the output of the sixteenth model is used as the input into the seventeenth model and the eighteenth model;

[0127] The output of the seventeenth model is used as the input of the nineteenth model, the twentieth model and the twenty-first model, the output of the eighteenth model is used as the input of the nineteenth model, the twentieth model and the twenty-first model, and the output of the nineteenth model and the twenty-first model are both used as the input of the twenty-first model;

[0128] The output of the twenty-first model is the parameter prediction value corresponding to the input sensor parameter;

[0129] Among them, the thirteenth model and the fourteenth model are both the BiTCN neural network model and the Transformer neural network model in series; the fifteenth model and the twentieth model are both the NARX neural network model and the Transformer neural network model in series; the sixteenth model and the nineteenth model are both the NARX neural network model and the SARIMA model in series; the seventeenth model is the BiGRU neural network model and the Transformer neural network model in series; the eighteenth model is the GCN graph neural network model and the Transformer neural network model in series; the twenty-first model is the integration of several GCN graph neural network models using the Adaboost model.

[0130] 3. The Internet of Things switch cabinet monitoring system described in the present invention includes the following contents.

[0131] like Figure 3 As shown, the IoT switch cabinet monitoring system includes a field acquisition layer, a communication network layer, and a remote monitoring layer;

[0132] The field acquisition layer includes a switch cabinet information acquisition module, which includes sensors arranged in the switch cabinet and their corresponding signal conditioning circuits, STM32 microprocessors and communication interfaces, and is used to collect switch cabinet equipment parameters, machine vision and micro-meteorological information parameters in real time, and convert sensor analog signals into digital signals for output;

[0133] The communication network layer is composed of the CAN field bus interface of the switch cabinet information acquisition module and the switch cabinet information concentrator, which is used for switch cabinet parameter acquisition and control command transmission of the remote monitoring layer;

[0134] The remote monitoring layer includes the switch cabinet cloud platform, the switch cabinet remote monitoring terminal and the switch cabinet mobile APP terminal. The switch cabinet cloud platform realizes the collection, storage and management of switch cabinet equipment parameters. The switch cabinet remote monitoring terminal realizes the detection of switch cabinet parameters and switch cabinet safety monitoring. The switch cabinet mobile APP terminal monitors and remotely controls the switch cabinet equipment parameters.

[0135] 1. Design of switch cabinet information collection module.

[0136] like Figure 4 As shown, the switch cabinet information acquisition module based on the CAN communication interface serves as a perception terminal for switch cabinet equipment parameters, machine vision and micro-meteorological information parameters, and realizes information interaction between multiple switch cabinet information acquisition modules and switch cabinet information concentrators through the CAN field bus communication network.

[0137] The switch cabinet information acquisition module includes sensors for collecting voltage, current, temperature, displacement, micro-meteorology, vibration and video images of switch cabinet equipment and corresponding signal conditioning circuits, STM32 microprocessors and CAN communication interfaces; multiple temperature sensors and multiple current sensors are used to measure the temperature and current of busbars, disconnector contacts and circuit breakers respectively; micro-meteorological sensors collect temperature, humidity, flame and gas concentration inside the switch cabinet; displacement sensors collect angular displacement of the main shaft of the circuit breaker; vibration sensors collect vibration amplitude of the switch cabinet; voltage sensors collect input and output voltage of the switch cabinet; the software of the switch cabinet information acquisition module mainly realizes CAN field bus communication and acquisition and preprocessing of switch cabinet equipment parameters.

[0138] The software is designed in C language with high compatibility, which greatly improves the work efficiency of software design and development and enhances the reliability, readability and portability of program codes.

[0139] 2. Design of switch cabinet information concentrator.

[0140] like Figure 5 As shown, the switch cabinet information concentrator includes a 4G communication module, an STM32 microcontroller and a CAN communication interface. The CAN communication interface realizes CAN field bus communication with the switch cabinet information acquisition module, and the 4G communication module realizes two-way data interaction between the switch cabinet information concentrator and the switch cabinet cloud platform server.

[0141] 3. Software function design for remote monitoring of switch cabinet.

[0142] like Figure 6 As shown, the switch cabinet remote monitoring terminal is an industrial control computer, which realizes the interaction of switch cabinet equipment parameters and switch cabinet monitoring instructions with the switch cabinet cloud platform through the 5G network. The switch cabinet remote monitoring terminal mainly realizes the collection and management of switch cabinet equipment parameters. The main functions of the switch cabinet remote monitoring terminal are communication parameter setting, switch cabinet equipment data analysis and management, and circuit breaker status detection module.

Claims

1. A circuit breaker intelligent detection method, characterized in that: It includes a circuit breaker state detection module, which obtains the safety level of the circuit breaker according to the input sensor parameters; the circuit breaker state detection module includes a circuit breaker parameter processing module, a first model to a twelfth model; The circuit breaker parameter processing module obtains parameter prediction values ​​according to the input sensor parameters through the neural network model, including the temperature prediction value, displacement prediction value and current prediction value of the circuit breaker; The temperature prediction value, the displacement prediction value and the current prediction value are input into the first model, the second model and the third model respectively; The output of the first model is used as the input of the fourth model, the fifth model and the sixth model; the output of the fifth model is used as the input of the sixth model; the output of the sixth model is used as the input of the first model and the seventh model; The output of the second model is used as the input of the fourth model, the seventh model and the eighth model, the output of the seventh model is used as the input of the eighth model; the output of the eighth model is used as the input of the second model and the ninth model; The output of the third model is used as the input of the fourth model, the ninth model and the tenth model, the output of the ninth model is used as the input of the tenth model; the output of the tenth model is used as the input of the third model and the fifth model; The output of the fourth model is used as the input of the eleventh model and the twelfth model; the output of the eleventh model is used as the input of the twelfth model, the fifth model, the seventh model and the ninth model; the output of the twelfth model is used as the input of the fourth model, the sixth model, the eighth model and the tenth model; The outputs of the sixth model, the eighth model and the tenth model constitute triangular fuzzy numbers, and the safety level of the circuit breaker is obtained according to the triangular fuzzy numbers; Among them, the first model is a BiTCN neural network model and a Transformer neural network model connected in series, the second model is a BiGRU neural network model and a Transformer neural network model connected in series; the third model is a GCN graph neural network model and a Transformer neural network model connected in series; the fourth model is a NARX neural network model and a Transformer neural network model connected in series; the fifth model, the seventh model, the ninth model and the eleventh model are all NARX neural network models and SARIMA models connected in series; the sixth model, the eighth model, the tenth model and the twelfth model are all the integration of several GCN graph neural network models using the Adaboost model.

2. The circuit breaker intelligent detection method according to claim 1, characterized in that: The circuit breaker parameter processing module includes a thirteenth model to a twenty-first model; The sensor parameters are input into the thirteenth model and the sixteenth model; the output of the thirteenth model is used as the input into the fifteenth model, the seventeenth model and the eighteenth model; the output of the fifteenth model is used as the input into the fourteenth model, the seventeenth model and the eighteenth model; the output of the fourteenth model is used as the input into the sixteenth model, the seventeenth model and the eighteenth model; The output of the sixteenth model serves as the input of the seventeenth model and the eighteenth model; The output of the seventeenth model is used as the input of the nineteenth model, the twentieth model and the twenty-first model, the output of the eighteenth model is used as the input of the nineteenth model, the twentieth model and the twenty-first model, and the output of the nineteenth model and the twenty-first model are both used as the input of the twenty-first model; The output of the twenty-first model is the parameter prediction value corresponding to the input sensor parameter; Among them, the thirteenth model and the fourteenth model are both the BiTCN neural network model and the Transformer neural network model in series; the fifteenth model and the twentieth model are both the NARX neural network model and the Transformer neural network model in series; the sixteenth model and the nineteenth model are both the NARX neural network model and the SARIMA model in series; the seventeenth model is the BiGRU neural network model and the Transformer neural network model in series; the eighteenth model is the GCN graph neural network model and the Transformer neural network model in series; the twenty-first model is the integration of several GCN graph neural network models using the Adaboost model.

3. The circuit breaker intelligent detection method according to claim 1, characterized in that: The outputs of the sixth model, the eighth model and the tenth model constitute triangular fuzzy numbers, and obtaining the safety level of the circuit breaker according to the triangular fuzzy numbers includes: dividing the safety level of the circuit breaker into very safe, relatively safe, safe, unsafe and very unsafe according to the triangular fuzzy numbers.

4. The circuit breaker intelligent detection method according to claim 3, characterized in that: The outputs of the sixth model, the eighth model and the tenth model constitute triangular fuzzy numbers.<x,y,z> , the corresponding relationship between the triangular fuzzy number and the safety level of the circuit breaker is: The triangular fuzzy number <0.7, 0.8, 0.9>, the safety level of the circuit breaker is very safe; The triangular fuzzy number <0.6, 0.7, 0.8>, the safety level of the circuit breaker is relatively safe; For triangular fuzzy numbers <0.5, 0.6, 0.7>, the safety level of the circuit breaker is safe; The triangular fuzzy number <0.4, 0.5, 0.6>, the safety level of the circuit breaker is unsafe; The triangular fuzzy number is <0.3, 0.4, 0.5>, and the safety level of the circuit breaker is very unsafe.

5. A circuit breaker intelligent detection system, characterized in that: It includes a circuit breaker state detection module, which is used to obtain the safety level of the circuit breaker according to the input sensor parameters; the circuit breaker state detection module includes a circuit breaker parameter processing module, a first model to a twelfth model; The circuit breaker parameter processing module obtains parameter prediction values ​​according to the input sensor parameters through the neural network model, including the temperature prediction value, displacement prediction value and current prediction value of the circuit breaker; The temperature prediction value, the displacement prediction value and the current prediction value are input into the first model, the second model and the third model respectively; The output of the first model is used as the input of the fourth model, the fifth model and the sixth model; the output of the fifth model is used as the input of the sixth model; the output of the sixth model is used as the input of the first model and the seventh model; The output of the second model is used as the input of the fourth model, the seventh model and the eighth model, the output of the seventh model is used as the input of the eighth model; the output of the eighth model is used as the input of the second model and the ninth model; The output of the third model is used as the input of the fourth model, the ninth model and the tenth model, the output of the ninth model is used as the input of the tenth model; the output of the tenth model is used as the input of the third model and the fifth model; The output of the fourth model is used as the input of the eleventh model and the twelfth model; the output of the eleventh model is used as the input of the twelfth model, the fifth model, the seventh model and the ninth model; the output of the twelfth model is used as the input of the fourth model, the sixth model, the eighth model and the tenth model; The outputs of the sixth model, the eighth model and the tenth model constitute triangular fuzzy numbers, and the safety level of the circuit breaker is obtained according to the triangular fuzzy numbers; Among them, the first model is a BiTCN neural network model and a Transformer neural network model connected in series, the second model is a BiGRU neural network model and a Transformer neural network model connected in series; the third model is a GCN graph neural network model and a Transformer neural network model connected in series; the fourth model is a NARX neural network model and a Transformer neural network model connected in series; the fifth model, the seventh model, the ninth model and the eleventh model are all NARX neural network models and SARIMA models connected in series; the sixth model, the eighth model, the tenth model and the twelfth model are all the integration of several GCN graph neural network models using the Adaboost model.

6. The circuit breaker intelligent detection system according to claim 5, characterized in that: The circuit breaker parameter processing module includes a thirteenth model to a twenty-first model; The sensor parameters are input into the thirteenth model and the sixteenth model; the output of the thirteenth model is used as the input into the fifteenth model, the seventeenth model and the eighteenth model; the output of the fifteenth model is used as the input into the fourteenth model, the seventeenth model and the eighteenth model; the output of the fourteenth model is used as the input into the sixteenth model, the seventeenth model and the eighteenth model; The output of the sixteenth model serves as the input of the seventeenth model and the eighteenth model; The output of the seventeenth model is used as the input of the nineteenth model, the twentieth model and the twenty-first model, the output of the eighteenth model is used as the input of the nineteenth model, the twentieth model and the twenty-first model, and the output of the nineteenth model and the twenty-first model are both used as the input of the twenty-first model; The output of the twenty-first model is the parameter prediction value corresponding to the input sensor parameter; Among them, the thirteenth model and the fourteenth model are both the BiTCN neural network model and the Transformer neural network model in series; the fifteenth model and the twentieth model are both the NARX neural network model and the Transformer neural network model in series; the sixteenth model and the nineteenth model are both the NARX neural network model and the SARIMA model in series; the seventeenth model is the BiGRU neural network model and the Transformer neural network model in series; the eighteenth model is the GCN graph neural network model and the Transformer neural network model in series; the twenty-first model is the integration of several GCN graph neural network models using the Adaboost model.

7. The circuit breaker intelligent detection system according to claim 5, characterized in that: The outputs of the sixth model, the eighth model and the tenth model constitute triangular fuzzy numbers, and obtaining the safety level of the circuit breaker according to the triangular fuzzy numbers includes: dividing the safety level of the circuit breaker into very safe, relatively safe, safe, unsafe and very unsafe according to the triangular fuzzy numbers.

8. The circuit breaker intelligent detection system according to claim 7, characterized in that: The outputs of the sixth model, the eighth model and the tenth model constitute triangular fuzzy numbers.<x,y,z> , the corresponding relationship between the triangular fuzzy number and the safety level of the circuit breaker is: The triangular fuzzy number <0.7, 0.8, 0.9>, the safety level of the circuit breaker is very safe; The triangular fuzzy number <0.6, 0.7, 0.8>, the safety level of the circuit breaker is relatively safe; For triangular fuzzy numbers <0.5, 0.6, 0.7>, the safety level of the circuit breaker is safe; The triangular fuzzy number <0.4, 0.5, 0.6>, the safety level of the circuit breaker is unsafe; The triangular fuzzy number is <0.3, 0.4, 0.5>, and the safety level of the circuit breaker is very unsafe.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the circuit breaker intelligent detection method according to any one of claims 1 to 4 is implemented.

10. An Internet of Things switch cabinet monitoring system, characterized in that: It includes field collection layer, communication network layer and remote monitoring layer; The field acquisition layer includes a switch cabinet information acquisition module, which includes sensors arranged in the switch cabinet and their corresponding signal conditioning circuits, STM32 microprocessors and communication interfaces, and is used to collect switch cabinet equipment parameters, machine vision and micro-meteorological information parameters in real time, and convert sensor analog signals into digital signals for output; The communication network layer is composed of the CAN field bus interface of the switch cabinet information acquisition module and the switch cabinet information concentrator, which is used for parameter acquisition of the switch cabinet and control command transmission of the remote monitoring layer; the switch cabinet information concentrator includes a communication module, an STM32 microcontroller and a CAN communication interface; The remote monitoring layer includes the switch cabinet cloud platform, the switch cabinet remote monitoring terminal and the switch cabinet mobile APP terminal. The switch cabinet cloud platform realizes the collection, storage and management of switch cabinet equipment parameters. The switch cabinet remote monitoring terminal realizes the detection of switch cabinet parameters and switch cabinet safety monitoring. The switch cabinet mobile APP terminal monitors and remotely controls the switch cabinet equipment parameters.

Images