Short circuit type identification system based on time convolutional neural network

Abstract

The invention relates to the technical field of electrical safety, in particular to a short circuit type identification system based on a time convolutional neural network. The TCN effectively captures local and global modes in sequence data through causal convolution and expansion convolution technologies, and is suitable for processing electrical line data with long-term dependence characteristics. The model construction comprises data preprocessing (cleaning, normalization and splitting), data set division (a training set, a verification set and a test set) and model training and optimization. In the training process, a cross entropy loss function and an Adam optimization algorithm are adopted, and accurate identification of the normal state, carbonization path type short circuit and metal short circuit type short circuit of the electrical circuit is achieved. The accuracy rate of the optimized model is close to 100%, and the classification performance is remarkably improved. In addition, the invention further provides an electrical line short circuit detection method based on the TCN, the importance of a real-time monitoring and early warning system is emphasized, and technical support is provided for electrical fire prevention.

Description

Technical Field

[0001] The present invention relates to the field of electrical safety technology, and in particular to a short circuit type recognition system based on a temporal convolutional neural network. Background Art

[0002] In electrical systems, electrical line short circuit is one of the main causes of electrical fires, which seriously threatens people's lives and property. Traditional electrical line short circuit detection methods mainly rely on manual inspections and empirical judgments, which are not only inefficient, but also difficult to accurately detect potential short circuit faults in real time. With the development of smart grid technology, the demand for automated and intelligent electrical line short circuit detection is becoming increasingly urgent.

[0003] In recent years, deep learning technology has made significant progress in the field of time series analysis and pattern recognition, providing a new solution for electrical line short circuit detection. As a neural network model designed specifically for sequence data, the Temporal Convolutional Network (TCN) can effectively capture local and global features in sequence data through causal convolution and dilated convolution techniques, and is suitable for processing electrical line data with long-term dependency characteristics.

[0004] However, there are still many challenges in the practical application of electrical line short circuit detection. First, electrical line voltage data often contains a lot of noise and outliers, which places high demands on the accuracy and robustness of the model. Secondly, different short circuit fault types (such as carbonization path short circuit and metal short circuit) show different characteristics in the voltage waveform, and the model needs to have the ability to distinguish these subtle differences. In addition, the establishment of a real-time monitoring and early warning system also imposes strict restrictions on the computational efficiency and response time of the model.

[0005] In order to solve the above problems, the applicant proposes a short circuit type recognition system based on a temporal convolutional neural network. Summary of the invention

[0006] The purpose of the present invention is to provide a short circuit type identification system based on a temporal convolutional neural network to solve the problems in the prior art.

[0007] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a short circuit type identification system based on a temporal convolutional neural network, comprising a data acquisition module, a data preprocessing module, a model training module, a short circuit detection module and an alarm module. The system can monitor the voltage data in the electrical circuit in real time, and use the temporal convolutional neural network model to automatically identify and distinguish the short circuit type.

[0008] Optionally, the data acquisition module is used to collect voltage data in the electrical circuit in real time to form continuous time series data.

[0009] Optionally, the data preprocessing module includes data cleaning, normalization and data splitting steps, the data cleaning step is used to eliminate obviously unreasonable voltage data; the normalization step is used to normalize the voltage data to a uniform range; the data splitting step is used to split the voltage data into time series samples of equal length, and divide them into training set, validation set and test set.

[0010] Optionally, the model training module adopts a temporal convolutional neural network model, which includes an input layer, a causal dilated convolution layer, a fully connected layer and an output layer. Through the combination of causal convolution and dilated convolution, the receptive field of the model is effectively expanded and the recognition accuracy of the model is improved.

[0011] Optionally, the model training module uses a cross entropy loss function to measure the difference between the model prediction result and the true label, and optimizes the model parameters through the Adam optimization algorithm to improve the generalization ability and stability of the model.

[0012] Optionally, the short circuit detection module uses a trained temporal convolutional neural network model to perform short circuit detection on voltage data collected in real time, and can automatically identify and distinguish between three situations: normal state, carbonization path type short circuit and metal short circuit type short circuit.

[0013] Optionally, the alarm module automatically triggers a corresponding alarm mechanism according to the detection result of the short-circuit detection module, including multiple alarm modes such as sound and light alarm and SMS notification, so as to remind relevant personnel to deal with the short-circuit fault in time.

[0014] Beneficial effects: 1. High-precision detection capability: By applying the causal convolution and dilated convolution techniques of the temporal convolutional network, the model can accurately capture local and global patterns in the time series data of electrical circuits, effectively improving the accuracy of short circuit detection. The accuracy of the optimized model on the test set is close to 100%, achieving accurate identification of the normal state of electrical circuits, carbonization path short circuits, and metal short circuits.

[0015] 2. Powerful long-term dependency modeling capabilities: The TCN model effectively expands the receptive field of the model through multi-scale convolution and residual connection, and can capture long-term dependencies in electrical line data. It is suitable for processing sequence data with long-term memory characteristics, and provides strong support for the prediction of early short-circuit faults.

[0016] 3. Flexibility and robustness of data processing: The present invention performs preprocessing steps such as cleaning, normalization and splitting on the data to ensure the quality of the model input data and improve the robustness of the model. Through normalization processing, the differences in voltage data values ​​under different conditions are eliminated, making the model more focused on the waveform shape characteristics and enhancing the generalization ability of the model.

[0017] 4. Efficiency of optimization algorithm: Using cross entropy loss function and Adam optimization algorithm for model training not only speeds up the convergence speed, but also effectively prevents gradient vanishing and gradient exploding problems through adaptive learning rate and momentum term, thus improving the stability and performance of the model.

[0018] 5. Integration of real-time monitoring and early warning systems: The detection method of the present invention can be seamlessly integrated into the electrical fire early warning system. By real-time monitoring of the electrical line status and timely warning of potential short-circuit faults, the probability of electrical fires is greatly reduced, ensuring the safety of personnel and property. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the process of an embodiment of the present invention; DETAILED DESCRIPTION

[0020] The following describes the preferred embodiments of the present invention with reference to the drawings in the specification, so that the technical content is clearer and easier to understand. The present invention can be embodied in many different forms of embodiments, and the protection scope of the present invention is not limited to the embodiments mentioned in the text.

[0021] Example 1

[0022] System Overview

[0023] The short circuit type identification system is mainly composed of data acquisition module, data preprocessing module, model training module, short circuit detection module and alarm module.

[0024] Data acquisition module

[0025] The data acquisition module is the input end of the entire system and is responsible for capturing the voltage signal in the electrical circuit in real time. In order to ensure the accuracy and integrity of the data, the module uses a high-precision voltage sensor to convert the analog voltage signal in the electrical circuit into a digital signal. The digital signal output by the sensor is transmitted to the data preprocessing module through the data bus for processing.

[0026] During the data collection process, a reasonable sampling frequency needs to be set according to the actual characteristics of the electrical circuit. The selection of the sampling frequency needs to comprehensively consider factors such as the frequency component of the electrical signal, the system response time requirements, and the hardware processing capabilities. At the same time, in order to reduce the impact of external interference on the collected data, the data acquisition module should have anti-electromagnetic interference capabilities to ensure the accuracy of the collected data.

[0027] Data preprocessing module

[0028] The data preprocessing module is the key link between data acquisition and model training. Its main task is to clean, normalize and split the collected raw voltage data to provide a high-quality data set for subsequent model training.

[0029] Data cleaning

[0030] The raw voltage data may contain abnormal values ​​due to sensor failure, external interference, etc. These abnormal values ​​will have an adverse effect on model training, so data cleaning is required to remove these abnormal data. The specific steps are as follows:

[0031] Set threshold value: According to the normal working voltage range of the electrical circuit, set a reasonable voltage threshold value (such as ±400V). Voltage values ​​outside this range are considered abnormal values.

[0032] Outlier detection: Traverse the raw voltage data, detect and record all voltage values ​​that exceed the threshold.

[0033] Outlier removal: Remove detected outliers from the data set to ensure that the data processed subsequently is within the normal range.

[0034] Data Normalization

[0035] Since the voltage value range of electrical lines in different working conditions varies greatly, in order to eliminate the impact of this dimension difference on model training, the cleaned voltage data needs to be normalized. The purpose of normalization is to map the voltage value to a uniform range (such as [-1,1]) so that the model can better learn the intrinsic characteristics of the data.

[0036] The normalization process is performed using the mapminmax function. The specific formula is as follows:

[0037] x_n=\frac{2(x-x_{\min})}{x_{\max}-x_{\min}}-1\]where \(x\) is the original voltage value, \(x_{\min}\) and \(x_{\max}\) are the minimum and maximum values ​​in the voltage data set, respectively, and \(x_n\) is the normalized voltage value. Data splitting Since the short-circuit detection model has certain requirements on the format and length of the input data, it is necessary to split the normalized continuous time series data into time series samples of equal length. Considering the periodicity of voltage data, it is a reasonable choice to split it based on half-power frequency cycle as the basic unit. The specific splitting steps are as follows:

[0038] Detect sign change: traverse the normalized voltage data and detect the change point of the data sign (i.e., the voltage value changes from positive to negative or from negative to positive).

[0039] Generate samples: Take the first sign change point detected as the starting point, and when the sign change is detected again, intercept the data as a sample. Repeat this process until all data has been processed.

[0040] Labeling: According to the working state of the electrical circuit corresponding to the sample (normal state, carbonization path short circuit, metal short circuit short circuit), label each sample with a corresponding label (0, 1, 2).

[0041] Model training module The model training module is the core part of the system, which is responsible for training the TCN-based short-circuit detection model using the preprocessed data set. The TCN model has achieved remarkable results in many fields with its powerful time series processing capabilities, and is particularly suitable for processing voltage time series data in electrical lines.

[0042] TCN model construction The TCN model is mainly composed of input layer, causal expansion convolution layer, fully connected layer and output layer. The specific structure is as follows:

[0043] Input layer: receives time series samples from the data preprocessing module as input.

[0044] Causal dilated convolution layer: As a key part of the model, the causal dilated convolution layer effectively expands the model's receptive field and improves the ability to understand time series data by combining causal convolution and dilated convolution operations. Causal convolution ensures that the model only uses past information when predicting the current time step to avoid leakage of future information; dilated convolution increases the receptive field of the convolution kernel by introducing a dilation rate parameter while keeping the number of parameters relatively small, which helps reduce the amount of computation and the risk of overfitting.

[0045] Fully connected layer: After flattening the output of the causal dilation convolutional layer, the deep features are further extracted through the fully connected layer.

[0046] Output layer: Use the Softmax function to map the output of the fully connected layer to a probability distribution, indicating the probability that the sample belongs to different short-circuit types. Loss function and optimization algorithm During the model training process, it is necessary to select appropriate loss function and optimization algorithm to guide the parameter update of the model.

[0047] Loss function: Since this task is a multi-classification problem, the cross entropy loss function (Categorical Cross-Entropy Loss) is selected as the loss function. This function can measure the difference between the probability distribution of the model output and the true label, and is a common choice in classification tasks.

[0048] Optimization algorithm: The Adam optimization algorithm is used to optimize the model parameters. The Adam algorithm combines the advantages of the momentum method and the adaptive learning rate, and can converge to the optimal solution more quickly and stably. During the training process, it is necessary to set reasonable hyperparameters such as the initial learning rate, the decay rate of the first-order moment estimate (the default is 0.9), and the decay rate of the second-order moment estimate (the default is 0.999).

[0049] The model training process mainly includes the following steps:

[0050] 1. Initialize model parameters: including convolution kernel size, expansion rate, learning rate, etc.

[0051] 2 Forward propagation: Input the preprocessed training set data into the model, and obtain the model's prediction results through calculations at each layer of the network.

[0052] 3. Calculate the loss: Use the cross entropy loss function to calculate the difference between the model prediction result and the true label.

[0053] 4 Back propagation: Back propagation is performed according to the loss value, and the model parameters are updated to reduce the loss value.

[0054] 5 Iterative training: Repeat steps 2 to 4 until the performance of the model on the validation set becomes stable or the preset number of iterations is reached. During the training process, the training effect of the model can be evaluated by observing the accuracy, loss value and other indicators on the training set and validation set. If overfitting occurs (i.e., the accuracy of the validation set decreases), it can be improved through methods such as data enhancement, regularization, and adjustment of the model structure.

[0055] Short circuit detection module

[0056] The short-circuit detection module uses the trained TCN model to perform short-circuit detection on the voltage data collected in real time. The specific steps are as follows: Data acquisition: The voltage data in the electrical circuit is captured in real time through the data acquisition module.

[0057] Data preprocessing: Clean, normalize and split the collected voltage data according to the method of the data preprocessing module.

[0058] Model prediction: The preprocessed time series samples are input into the trained TCN model for prediction. The model outputs the probability distribution of each sample belonging to the normal state, carbonization path short circuit and metal short circuit.

[0059] Result determination: Compare the predicted probabilities of each short circuit type and select the category with the highest probability as the final detection result.

[0060] The alarm module automatically triggers the corresponding alarm mechanism according to the detection results of the short-circuit detection module. When a short-circuit fault is detected, the system will remind relevant personnel to deal with it in time through various means such as sound and light alarms and text message notifications. The alarm information should include detailed information such as the fault type, fault location and time in order to quickly locate and troubleshoot the fault. The design of the alarm module should fully consider the user's usage habits and needs to ensure the accuracy and timeliness of the alarm information. At the same time, in order to improve the availability and reliability of the system, the alarm module should have self-diagnosis and maintenance functions in order to promptly discover and repair potential fault points. Through the above-mentioned specific implementation methods, this short-circuit type identification system realizes the accurate identification and early warning functions of different types of short circuits in electrical lines. The system not only improves the monitoring level of electrical safety, but also reduces the workload and cost of manual inspections, providing a strong guarantee for the stable operation of the power system.

[0061] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the attached claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any figure mark in the claims should not be regarded as limiting the claims involved.

[0062] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A short circuit type recognition system based on temporal convolutional neural network, characterized in that: The system includes a data acquisition module, a data preprocessing module, a model training module, a short circuit detection module and an alarm module. The system can monitor the voltage data in the electrical circuit in real time and use a time convolutional neural network model to automatically identify and distinguish short circuit types.

2. The system according to claim 1, characterized in that The data acquisition module is used to collect voltage data in the electrical circuit in real time to form continuous time series data.

3. The system according to claim 1, characterized in that The data preprocessing module includes data cleaning, normalization and data splitting steps. The data cleaning step is used to eliminate obviously unreasonable voltage data; the normalization step is used to normalize the voltage data to a uniform range; the data splitting step is used to split the voltage data into time series samples of equal length and divide them into training set, verification set and test set.

4. The system according to claim 1, characterized in that The model training module adopts a temporal convolutional neural network model, which includes an input layer, a causal dilated convolution layer, a fully connected layer and an output layer. Through the combination of causal convolution and dilated convolution, the receptive field of the model is effectively expanded and the recognition accuracy of the model is improved.

5. The system according to claim 4, characterized in that The model training module uses the cross entropy loss function to measure the difference between the model prediction results and the true labels, and optimizes the model parameters through the Adam optimization algorithm to improve the generalization ability and stability of the model.

6. The system according to claim 1, characterized in that The short circuit detection module uses a trained temporal convolutional neural network model to perform short circuit detection on voltage data collected in real time, and can automatically identify and distinguish between three situations: normal state, carbonization path type short circuit and metal short circuit type short circuit.

7. The system according to claim 1, characterized in that The alarm module automatically triggers the corresponding alarm mechanism according to the detection result of the short-circuit detection module, including multiple alarm modes such as sound and light alarm and SMS notification, so as to remind relevant personnel to deal with the short-circuit fault in time.

Images