Low-voltage electrical equipment fault prediction system based on artificial intelligence

Abstract

The invention relates to the technical field of equipment fault prediction, and discloses a low-voltage electrical equipment fault prediction system based on artificial intelligence, and the system comprises a data collection and preprocessing module, a feature engineering module, a prediction module, and an optimization hyper-parameter module. The system extracts time sequence features, frequency domain features and fault mode features by collecting electrical sensor and monitoring data, and based on an enhanced TCN-Learar-BiLSTM model, realizes accurate prediction of equipment health status and fault risk by combining alternate expansion convolution, trend and seasonal decomposition and linear modeling technologies; meanwhile, through joint optimization of an improved faecium beetle algorithm and an immune algorithm, hyper-parameter configuration of the model is dynamically adjusted, and prediction accuracy and stability are remarkably improved; the system is suitable for the field of electrical equipment operation and maintenance and intelligent power management, and can effectively improve the equipment fault early warning capability and the operation reliability.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment fault prediction, and particularly to a fault prediction system for low-voltage electrical equipment based on artificial intelligence. Background Art

[0002] With the rapid development of technology, low-voltage electrical equipment plays an important role in modern industry and daily life. However, long-term operation and complex usage environments often lead to equipment failures, seriously affecting the stability and safety of the system. Therefore, more and more enterprises adopt artificial intelligence technology to assist in equipment condition monitoring and fault prediction to improve the safety and reliability of system operation. However, although the artificial intelligence methods in traditional low-voltage equipment fault prediction systems have made some progress in fault diagnosis accuracy, there are still significant deficiencies. On the one hand, traditional fault prediction systems mostly rely on statistical analysis or methods based on specific signal patterns, which often ignore the complex characteristics of electrical equipment operation data, such as time series dependence, frequency domain characteristics, and potential fault mode characteristics, resulting in insufficient sensitivity of traditional systems in processing diverse operation data and difficulty in capturing key trends and abnormal signals, thus affecting the accuracy and reliability of fault prediction. On the other hand, although deep learning technology has been introduced into traditional systems, the complex interaction relationships between time series, frequency domain, and fault mode characteristics have not been fully modeled, and the feature fusion effect is poor. In addition, the hyperparameter optimization strategy of traditional systems is single and easily falls into local optimal solutions, restricting the performance and reliability of fault risk prediction. Therefore, there is an urgent need for a fault prediction system for low-voltage electrical equipment that can comprehensively integrate various characteristics of equipment operation, optimize hyperparameter configuration, and improve prediction accuracy to meet the requirements of modern equipment operation safety and stability. Summary of the Invention

[0003] The present invention proposes a fault prediction system for low-voltage electrical equipment based on artificial intelligence. This system combines various data analysis and processing means to achieve comprehensive monitoring and efficient prediction of the equipment status in the complex scenarios of low-voltage electrical equipment operation. First, in terms of health status prediction and fault risk judgment, the present invention optimizes the feature extraction of the system by introducing an alternating extended convolution mechanism, and combines the decomposition methods of trend, seasonality and abnormal features to perform multi-dimensional analysis on the equipment operation data in the system, accurately capturing the multi-scale dependence relationship of the equipment operation status, thereby significantly improving the accuracy of the system for health status classification and fault risk prediction, realizing the refined monitoring of the equipment operation status by the system, and providing strong support for the real-time maintenance of the equipment and the early warning of faults. Second, in terms of hyperparameter optimization, the present invention first introduces a dynamic scaling factor, a position memory guidance mechanism and a sine and cosine hybrid mechanism, and combines the golden sine strategy to optimize the system hyperparameter adjustment process. On this basis, the system further integrates the immune optimization technology, effectively solving the problems of slow convergence speed and easy to fall into local optimum of the traditional system, enabling the system to quickly and accurately determine the optimal hyperparameter configuration, and significantly enhancing the adaptability to the complex operation environment of low-voltage electrical equipment.

[0004] The present invention provides a fault prediction system for low-voltage electrical equipment based on artificial intelligence. The system includes a data acquisition and preprocessing module, a feature engineering module, a prediction module and a hyperparameter optimization module;

[0005] The data acquisition and preprocessing module collects electrical sensor data and electrical real-time monitoring data to form raw electrical data; performs outlier processing, missing value processing and normalization on the raw electrical data to generate preprocessed electrical data;

[0006] The feature engineering module extracts the time series features, frequency domain features and fault mode features of the preprocessed electrical data;

[0007] The prediction module establishes a TCN model, optimizes the residual block structure in the TCN model through an alternating extended convolution mechanism, uses the linear modeling and fusion of trend, seasonality and abnormal feature decomposition to optimize the output of the TCN model, constructs an enhanced TCN-Linear-BiLSTM model, and inputs the time series features, frequency domain features and fault mode features into the enhanced TCN-Linear-BiLSTM model to predict the health status and fault risk of low-voltage electrical equipment;

[0008] Optimize the hyperparameter module, introduce a dynamic scaling factor, a position memory guidance mechanism, and a sine and cosine hybrid mechanism, and optimize the rolling update mechanism of the dung beetle algorithm in combination with the golden sine strategy; optimize the population quality of the dung beetle algorithm in combination with an improved immune algorithm to construct an improved dung beetle algorithm; find the optimal hyperparameter configuration of the enhanced TCN-Linear-BiLSTM model through the improved dung beetle algorithm.

[0009] Furthermore, for the prediction module, the process of inputting time series features, frequency domain features, and fault mode features into the enhanced TCN-Linear-BiLSTM model to predict the health status and fault risk of low-voltage electrical equipment specifically includes the following steps:

[0010] Step M1: Normalization processing: Perform normalization processing on time series features, frequency domain features, and fault mode features to generate normalized time series feature data;

[0011] Step M2: Residual block design: Process the normalized time series feature data using an alternating extended convolutional residual block to generate comprehensive time series feature data;

[0012] Step M3: Trend, seasonality, and anomaly feature decomposition: Decompose the comprehensive time series feature data into trend and seasonality to extract the trend component and generate long-term change feature data; extract the periodic component from the comprehensive time series feature data to generate periodic change feature data; extract the anomaly signal component from the comprehensive time series feature data to generate anomaly change feature data; the comprehensive change feature data includes long-term change feature data, periodic change feature data, and anomaly change feature data;

[0013] Step M4: Linear modeling and fusion: Independently model the long-term change feature data, periodic change feature data, and anomaly change feature data using linear layers respectively, and perform weighted fusion to generate TCN feature data;

[0014] Step M5: BiLSTM modeling: Further model the TCN feature data using BiLSTM. BiLSTM captures the bidirectional dependency relationships in the TCN feature data through two LSTM networks, a forward one and a backward one, to generate comprehensive time-dependent feature data;

[0015] Step M6: Prediction and output: Input the comprehensive time-dependent feature data into the fully connected layer, perform non-linear mapping through the activation function, and output the health status and fault risk of the low-voltage electrical equipment; the health status prediction includes the classification of the equipment operation status; the fault risk prediction includes the probability of a fault occurring within a future time period and the judgment of the fault type.

[0016] Furthermore, step M2 specifically includes the following steps:

[0017] Step M21: Standard extended convolution: Perform three-layer standard extended convolution on the normalized time series feature data to generate multi-scale standard extended convolution feature data;

[0018] Step M22: Standard regularization: Perform weight normalization, ReLU activation, and Dropout regularization on the multi-scale standard extended convolution feature data to obtain multi-scale standard regularized time feature data;

[0019] Step M23: Alternating extended convolution: Introduce an alternating extended convolution mechanism to generate alternating extended convolution feature data;

[0020] Step M24: Standard regularization: Perform weight normalization, ReLU activation, and Dropout regularization on the alternating extended convolution feature data to obtain multi-scale alternating regularized time feature data;

[0021] Step M25: Residual connection: Superimpose the normalized time series feature data and the multi-scale alternating regularized time feature data through residual connection to generate comprehensive time series feature data.

[0022] Furthermore, for the hyperparameter optimization module, the process of finding the optimal hyperparameter configuration of the enhanced TCN-Linear-BiLSTM through the improved dung beetle algorithm specifically includes the following steps:

[0023] Step B1: Population initialization: Use the Latin hypercube sampling method to generate an initial dung beetle population, ensuring that the initial dung beetle population is evenly distributed within the search space. Each dung beetle position represents a set of hyperparameter configurations of the enhanced TCN-Linear-BiLSTM model;

[0024] Step B2: Immune memory bank construction: Construct an immune memory bank, calculate the fitness value for each dung beetle individual in the initial dung beetle population, and select the dung beetle individual with the optimal fitness value as the initial global optimal solution, which is stored in the immune memory bank;

[0025] Step B3: Rolling update: Based on the initial global optimal solution, introduce a dynamic scaling factor, a position memory guiding mechanism, and a sine and cosine hybrid mechanism, and update the dung beetle individuals in combination with the golden sine strategy. Check whether the updated dung beetle individuals exceed the search space, and perform boundary adjustment on the dung beetle individuals that exceed the search space to generate a rolling population set. Recalculate the fitness value of the dung beetle population for the rolling population set, select the dung beetle individual with the optimal fitness value as the rolling global optimal solution, and store the rolling global optimal solution in the immune memory bank;

[0026] Step B4: Immune algorithm optimization: According to the rolling population set, use the concentration selection mechanism, cloning operation, mutation operation, and immune memory bank update of the immune algorithm to further optimize the quality of the dung beetle population and update the immune memory bank;

[0027] Step B5: Determine convergence and output: Set the maximum number of iterations. When the maximum number of iterations is reached, terminate the optimization; otherwise, return to Step B2 to continue the optimization. After the optimization is terminated, extract the stored final global optimal solution from the immune memory bank and use the final global optimal solution as the optimal hyperparameter configuration.

[0028] Furthermore, Step B4 specifically includes the following steps:

[0029] Step B41: Set the similarity threshold. According to the rolling population set, calculate the similarity between each pair of dung beetle individuals, and suppress the dung beetle individuals in the rolling population set that exceed the similarity threshold to avoid the homogenization of dung beetle individuals in the rolling population set, and generate a concentration selection population set;

[0030] Step B42: Calculate the fitness values of all dung beetle individuals in the concentration selection population set. According to the ranking of the fitness values of the dung beetle individuals, select the top 10% of the dung beetle individuals with the highest fitness values as high-fitness individuals, clone the high-fitness individuals through the cloning operation, and jointly form a post-cloning population set with the original dung beetle individuals in the concentration selection population set; The cloning operation formula is as follows:

[0031] Step B43: Calculate the fitness values of all dung beetle individuals in the post-cloning population set. According to the fitness value ranking, select the bottom 20% of the dung beetle individuals with the lowest fitness values as the mutation operation target individuals, introduce Cauchy distribution perturbation and population center guidance to perform mutation operations on the mutation operation target individuals, and jointly form a post-mutation population set with the original dung beetle individuals in the post-cloning population set;

[0032] Step B44: Calculate the fitness values of all dung beetle individuals in the post-mutation population set, select the dung beetle individual with the best fitness value as the current global optimal solution, compare the current global optimal solution with the historical optimal solution in the immune memory bank, and update the immune memory bank.

[0033] Adopting the above solution, the beneficial effects obtained by the present invention are as follows:

[0034] First, the present invention optimizes the residual block structure in the system through an alternating expansion convolution mechanism, significantly enhancing the feature extraction ability of the system; on this basis, by combining the decomposition method of trend, seasonality, and anomaly features, as well as linear modeling and fusion means, the precise separation and weighted fusion of multi-dimensional data features are realized, effectively solving the problem of insufficient capture of complex time-series signal dependence in traditional systems, significantly enhancing the accuracy of the system in classifying the health status of equipment and predicting fault risks, and providing reliable technical support for real-time fault warning and equipment maintenance plans;

[0035] Second, the present invention optimizes the rolling update process of the dung beetle algorithm by introducing a dynamic scaling factor, a position memory guidance mechanism, and a sine-cosine hybrid mechanism, and combining the golden sine strategy, thereby realizing the dynamic adjustment of system parameters. On this basis, by combining an improved immune algorithm, through concentration selection, cloning operation, and mutation operation, the problems of slow convergence speed and easy entrapment in local optimal solutions in traditional systems are effectively solved; this hyperparameter optimization strategy significantly improves the adaptability and operation efficiency of the system in different environments, and provides fast and accurate parameter configuration support for the system to predict the health status of low-voltage electrical equipment and judge fault risks;

[0036] Through the above optimization technologies, the present invention realizes the precise monitoring and real-time prediction of the operating state of low-voltage electrical equipment, and improves the reliability of judging the health status of equipment and fault risks; the system effectively solves the problems of insufficient prediction accuracy and low operation efficiency in traditional systems, and enhances the adaptability of the system to complex operating environments of equipment; at the same time, through an efficient hyperparameter optimization mechanism, the consumption of computing resources of the system is significantly reduced, the practicability and economy of the system are improved, and an advanced technical solution is provided for the operation and maintenance of low-voltage electrical equipment in industrial and civil scenarios. Brief Description of the Drawings

[0037] Figure 1 It is a schematic diagram of the modules of a low-voltage electrical equipment fault prediction system based on artificial intelligence proposed by the present invention;

[0038] Figure 2 It is a schematic diagram of the process of the enhanced TCN-Linear-BiLSTM model proposed in Embodiment 2;

[0039] Figure 3 It is a schematic diagram of the structure of step M2 proposed in Embodiment 4. Detailed Embodiments

[0040] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0041] Embodiment 1. According to Figure 1 , the present invention provides a low-voltage electrical equipment fault prediction system based on artificial intelligence. The system includes a data collection and preprocessing module, a feature engineering module, a prediction module, and an optimized hyperparameter module;

[0042] The data collection and preprocessing module collects electrical sensor data and electrical real-time monitoring data to form original electrical data; performs outlier processing, missing value processing, and normalization on the original electrical data to generate preprocessed electrical data;

[0043] The feature engineering module extracts the time series features, frequency domain features, and fault mode features of the preprocessed electrical data;

[0044] The prediction module establishes a TCN model, optimizes the residual block structure in the TCN model through an alternating expansion convolution mechanism, and optimizes the output of the TCN model by using linear modeling and fusion of trend, seasonal, and anomaly feature decomposition to construct an enhanced TCN-Linear-BiLSTM model. Input the time series features, frequency domain features, and fault mode features into the enhanced TCN-Linear-BiLSTM model to predict the health status and fault risk of low-voltage electrical equipment;

[0045] The optimized hyperparameter module introduces a dynamic scaling factor, a position memory guidance mechanism, and a sine and cosine hybrid mechanism, and combines the golden sine strategy to optimize the rolling update mechanism of the dung beetle algorithm; combines an improved immune algorithm to optimize the population quality of the dung beetle algorithm to construct an improved dung beetle algorithm; finds the optimal hyperparameter configuration of the enhanced TCN-Linear-BiLSTM model through the improved dung beetle algorithm.

[0046] Embodiment 2. According to Figure 2 , based on Embodiment 1, the process of the prediction module inputting the time series features, frequency domain features, and fault mode features into the enhanced TCN-Linear-BiLSTM model to predict the health status and fault risk of low-voltage electrical equipment specifically includes the following steps:

[0047] Step M1: Normalization processing: Perform normalization processing on the time series features, frequency domain features, and fault mode features to generate normalized time series feature data;

[0048] Step M2: Residual block design: Use the alternating extended convolutional residual block to process the normalized time series feature data to generate comprehensive time series feature data;

[0049] The alternating extended convolutional residual block includes standard extended convolution, standard regularization, alternating extended convolution, standard regularization, and residual connection;

[0050] The number of residual blocks is: 6;

[0051] Step M3: Trend, seasonality, and anomaly feature decomposition: Perform trend and seasonality decomposition on the comprehensive time series feature data to extract the trend component and generate long-term change feature data; extract the periodic component from the comprehensive time series feature data to generate periodic change feature data; extract the anomaly signal component from the comprehensive time series feature data to generate anomaly change feature data; the comprehensive change feature data includes long-term change feature data, periodic change feature data, and anomaly change feature data;

[0052] Step M4: Linear modeling and fusion: Independently model the long-term change feature data, periodic change feature data, and anomaly change feature data using linear layers and perform weighted fusion to generate TCN feature data;

[0053] Step M5: BiLSTM modeling: Use BiLSTM to further model the TCN feature data. BiLSTM captures the bidirectional dependencies in the TCN feature data through two LSTM networks, forward and backward, to generate comprehensive time-dependent feature data;

[0054] Step M6: Prediction and output: Input the comprehensive time-dependent feature data into the fully connected layer, perform nonlinear mapping through the activation function, and output the health status and fault risk of the low-voltage electrical equipment; the health status prediction includes the classification of the equipment operation status; the fault risk prediction includes the probability of failure and the judgment of the fault type within a future time period.

[0055] Embodiment 3, based on Embodiment 1, the process of the prediction module inputting the time series features, frequency domain features, and fault mode features into the enhanced TCN-Linear-BiLSTM model to predict the health status and fault risk of the low-voltage electrical equipment specifically includes the following steps:

[0056] Step R1: Normalization processing: Perform normalization processing on the time series features, frequency domain features, and fault mode features to generate normalized time series feature data;

[0057] Step R2: Residual block design: Use the alternating extended convolutional residual block to process the normalized time series feature data to generate TCN feature data;

[0058] Step R3: BiLSTM Modeling: Use BiLSTM to further model the TCN feature data. BiLSTM captures the bidirectional dependencies in the TCN feature data through two LSTM networks, a forward one and a backward one, and generates comprehensive time-dependent feature data.

[0059] Step R4: Prediction and Output: Input the comprehensive time-dependent feature data into the fully connected layer, perform non-linear mapping through the activation function, and output the health status and fault risk of the low-voltage electrical equipment; the health status prediction includes the classification of the equipment operation status; the fault risk prediction includes the probability of failure and the judgment of the fault type within the future time period.

[0060] Example 4, according to Figure 3 , based on Example 2, Step M2 specifically includes the following steps:

[0061] Step M21: Standard Dilated Convolution: Perform three layers of standard dilated convolution on the normalized time series feature data, with dilation factors being in sequence; the first layer extracts local time-dependent features, the second layer extracts medium-range time-dependent features, and the third layer extracts long-term time-dependent features, generating multi-scale standard dilated convolution feature data;

[0062] Step M22: Standard Regularization: Perform weight normalization, ReLU activation, and Dropout regularization on the multi-scale standard dilated convolution feature data to obtain multi-scale standard regularized time feature data;

[0063] Step M23: Alternating Dilated Convolution: Introduce the alternating dilated convolution mechanism, with dilation factors being in sequence; the first layer extracts long-term time-dependent features to enhance global pattern perception; the second layer focuses on medium-range time-dependent features; the third layer refines local time features, generating alternating dilated convolution feature data;

[0064] Step M24: Standard Regularization: Perform weight normalization, ReLU activation, and Dropout regularization on the alternating dilated convolution feature data to obtain multi-scale alternating regularized time feature data;

[0065] Step M25: Residual Connection: Superimpose the normalized time series feature data and the multi-scale alternating regularized time feature data through residual connection to generate comprehensive time series feature data.

[0066] Example 5, based on Example 2, Step M2 specifically includes the following steps:

[0067] Step Y1: Standard Dilated Convolution: Perform three layers of standard dilated convolution on the normalized time series feature data, with dilation factors being ; The first layer extracts local time-dependent features, the second layer extracts medium-range time-dependent features, and the third layer extracts long-term time-dependent features to generate multi-scale standard extended convolution feature data;

[0068] Step Y2: Standard regularization: Perform weight normalization, ReLU activation, and Dropout regularization on the multi-scale standard extended convolution feature data to obtain multi-scale standard regularized time feature data;

[0069] Step Y3: Residual connection: Superimpose the normalized time series feature data and the multi-scale standard regularized time feature data through residual connection to generate comprehensive time series feature data.

[0070] Example 6, based on Example 4, optimizes the hyperparameter module. The process of finding the optimal hyperparameter configuration of the enhanced TCN-Linear-BiLSTM through the improved dung beetle algorithm specifically includes the following steps:

[0071] Step B1: Population initialization: Use the Latin hypercube sampling method to generate an initial dung beetle population, ensuring that the initial dung beetle population is evenly distributed within the search space. Each dung beetle position represents a set of hyperparameter configurations of the enhanced TCN-Linear-BiLSTM model;

[0072] Step B2: Immune memory bank construction: Construct an immune memory bank, calculate the fitness value for each dung beetle individual in the initial dung beetle population, and select the dung beetle individual with the optimal fitness value as the initial global optimal solution, which is stored in the immune memory bank;

[0073] Step B3: Rolling update: According to the initial global optimal solution, introduce a dynamic scaling factor, a position memory guiding mechanism, and a sine and cosine hybrid mechanism, and update the dung beetle individuals in combination with the golden sine strategy. Check whether the updated dung beetle individuals exceed the search space, and perform boundary adjustment on the dung beetle individuals that exceed the search space to generate a rolling population set. Recalculate the fitness value of the dung beetle population for the rolling population set, select the dung beetle individual with the optimal fitness value as the rolling global optimal solution, and store the rolling global optimal solution in the immune memory bank; The rolling update formula that introduces a dynamic scaling factor, a position memory guiding mechanism, and a sine and cosine hybrid mechanism, and combines the golden sine strategy is as follows:

[0074] ;

[0075] Among them, represents the dung beetle individual index, represents the iteration number index, represents the dung beetle individual at the position in the th iteration, At the position of the iteration, represents a random number, and represents the sine and cosine components of the random number ; represents the maximum number of iterations, represents the control factor, represents the dynamic scaling factor, represents the step size adjustment factor, represents best, represents the current global optimal individual position, represents the fitness value of, represents the fitness value of; represents the normalization term; and represents the weight coefficient;

[0076] Step B4: Immune algorithm optimization: According to the rolling population set, use the concentration selection mechanism, cloning operation, mutation operation and immune memory bank update of the immune algorithm to further optimize the quality of the dung beetle population and update the immune memory bank;

[0077] Step B5: Determine convergence and output: Set the maximum number of iterations. If the maximum number of iterations is reached, terminate the optimization; otherwise, return to Step B2 to continue optimization. After the optimization is terminated, extract the stored final global optimal solution from the immune memory bank and use the final global optimal solution as the optimal hyperparameter configuration.

[0078] Example 7. Based on Example 4, the process of optimizing the hyperparameter module to find the optimal hyperparameter configuration of the enhanced TCN-Linear-BiLSTM by the dung beetle algorithm specifically includes the following steps:

[0079] Step E1: Population initialization: Use the Latin hypercube sampling method to generate the initial dung beetle population to ensure that the initial dung beetle population is evenly distributed in the search space, and each dung beetle position represents a set of hyperparameter configurations of the enhanced TCN-Linear-BiLSTM model;

[0080] Step E2: Immune memory bank construction: Construct an immune memory bank, calculate the fitness value of each dung beetle individual in the initial dung beetle population, and select the dung beetle individual with the best fitness value as the initial global optimal solution and store it in the immune memory bank;

[0081] Step E3: Rolling update: Update the dung beetle individuals according to the initial global optimal solution, check whether the updated dung beetle individuals exceed the search space, adjust the boundaries of the dung beetle individuals that exceed the search space, generate a rolling population set, recalculate the fitness values of the dung beetle population for the rolling population set, select the dung beetle individual with the optimal fitness value as the rolling global optimal solution, and store the rolling global optimal solution in the immune memory bank; The rolling update formula is as follows:

[0082] ;

[0083] where, represents randomly selecting the dung beetle individual index, randomly select a dung beetle individual 's position;

[0084] Step E4: Immune algorithm optimization: According to the rolling population set, use the concentration selection mechanism, cloning operation, mutation operation and immune memory bank update of the immune algorithm to further optimize the quality of the dung beetle population and update the immune memory bank;

[0085] Step E5: Judge convergence and output: Set the maximum number of iterations. If the maximum number of iterations is reached, terminate the optimization; otherwise, return to step B2 to continue the optimization. After the optimization is terminated, extract the stored final global optimal solution from the immune memory bank and use the final global optimal solution as the optimal hyperparameter configuration.

[0086] Example VIII, this example is based on the above example, step B4, specifically includes the following steps:

[0087] Step B41: Set the similarity threshold. According to the rolling population set, calculate the similarity between each pair of dung beetle individuals, suppress the dung beetle individuals in the rolling population set that exceed the similarity threshold, avoid the homogenization of the dung beetle individuals in the rolling population set, and generate a concentration selection population set;

[0088] Step B42: Calculate the fitness values of all dung beetle individuals in the concentration selection population set. According to the ranking of the dung beetle individual fitness values, select the top 10% of the dung beetle individuals with the optimal fitness values as the high-fitness individuals, clone the high-fitness individuals through the cloning operation, and jointly form a post-cloning population set with the original dung beetle individuals in the concentration selection population set; The cloning operation formula is as follows:

[0089] ;

[0090] where, represents the new dung beetle individual generated by the cloning operation, represents the high-fitness individual, represents the mutation intensity factor in the cloning operation; represents a random number;

[0091] Step B43: Calculate the fitness values of all dung beetle individuals in the cloned population set. According to the fitness value ranking, select the dung beetle individuals in the last 20% of the fitness values as the target individuals for the mutation operation. Introduce Cauchy distribution perturbation and population center guidance to perform the mutation operation on the target individuals for the mutation operation. The new dung beetle individuals generated by the mutation operation and the original dung beetle individuals in the cloned population set together constitute the mutated population set. The mutation operation formula is as follows:

[0092] ;

[0093] where, represents the index of the target individual for the mutation operation, represents the new dung beetle individual generated by the mutation operation, represents the th dung beetle individual in the last 20% of the fitness values represents the Cauchy distribution perturbation factor, represents the weight factor, controlling the intensity of the population center guidance; represents the geometric center of the cloned population set;

[0094] Step B44: Calculate the fitness values of all dung beetle individuals in the mutated population set. Select the dung beetle individual with the best fitness value as the current global optimal solution. Compare the current global optimal solution with the historical optimal solution in the immune memory bank and update the immune memory bank.

[0095] The above describes the present invention and its implementation manners. This description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto; in short, if those of ordinary skill in the art are inspired by it and without departing from the gist of the present invention creation, they design similar structural manners and embodiments to this technical solution without creative efforts, which shall fall within the protection scope of the present invention.

Claims

1. A low-voltage electrical equipment fault prediction system based on artificial intelligence, including a data acquisition and preprocessing module, generating preprocessed electrical data, characterized in that: The system also includes a feature engineering module, a prediction module, and an optimization hyperparameter module; The feature engineering module extracts comprehensive feature data of preprocessed electrical data; The prediction module constructs an enhanced TCN-Linear-BiLSTM model, inputs the comprehensive feature data into the enhanced TCN-Linear-BiLSTM model, and predicts the health status and fault risk of low-voltage electrical equipment; The hyperparameter optimization module constructs an improved dung beetle algorithm; the optimal hyperparameter configuration of the enhanced TCN-Linear-BiLSTM model is found through the improved dung beetle algorithm.

2. The artificial intelligence-based low-voltage electrical equipment fault prediction system according to claim 1, characterized in that: The comprehensive characteristic data includes time series characteristic data, frequency domain characteristic data and failure mode characteristic data.

3. The artificial intelligence-based low-voltage electrical equipment fault prediction system according to claim 1, characterized in that: The prediction module predicts the health status and failure risk of low-voltage electrical equipment, which specifically includes the following steps: Step M1: normalize the comprehensive feature data to generate normalized time series feature data; Step M2: Process the normalized time series feature data using the alternating extended convolution residual block to generate comprehensive time series feature data; Step M3: extracting trend components, periodic components and abnormal signal components from the comprehensive time series feature data to generate comprehensive change feature data; Step M4: weighted fusion of comprehensive change feature data to generate TCN feature data; Step M5: Use BiLSTM to capture the bidirectional dependency in TCN feature data and generate comprehensive time-dependent feature data; Step M6: Generate predicted health status and failure risk of low-voltage electrical equipment.

4. The artificial intelligence-based low-voltage electrical equipment fault prediction system according to claim 2, characterized in that: Step M2 specifically includes the following steps: Step M21: performing three-layer standard dilated convolution on the normalized time series feature data to generate multi-scale standard dilated convolution feature data; Step M22: normalizing the multi-scale standard extended convolution feature data to obtain multi-scale standard regularized time feature data; Step M23: Introduce an alternating extended convolution mechanism to perform a convolution operation on the multi-scale standard regularized time feature data to generate alternating extended convolution feature data; Step M24: normalizing the alternating extended convolution feature data to obtain multi-scale alternating regularized time feature data; Step M25: superimpose the normalized time series feature data and the multi-scale alternating regularized time feature data through residual connection to generate comprehensive time series feature data.

5. The artificial intelligence-based low-voltage electrical equipment fault prediction system according to claim 1, characterized in that: The process of optimizing the hyperparameter module and finding the optimal hyperparameter configuration of the enhanced TCN-Linear-BiLSTM model specifically includes the following steps: Step B1: Generate the initial dung beetle population using the Latin hypercube sampling method, where each dung beetle position represents a set of hyperparameter configurations for the enhanced TCN-Linear-BiLSTM model; Step B2: construct an immune memory bank, calculate the fitness value of each dung beetle individual in the initial dung beetle population, select the dung beetle individual with the best fitness value as the initial global optimal solution, and store it in the immune memory bank; Step B3: According to the initial global optimal solution, a dynamic scaling factor, a position memory guidance mechanism, and a sine and cosine hybrid mechanism are introduced, and the golden sine strategy is combined to update the dung beetle individuals to generate a rolling population set, and the fitness value of the dung beetle population is recalculated for the rolling population set. The dung beetle individual with the best fitness value is selected as the rolling global optimal solution, and the rolling global optimal solution is stored in the immune memory bank; Step B4: Based on the rolling population set, the concentration selection mechanism, cloning operation, mutation operation and immune memory bank update of the immune algorithm are used to further optimize the dung beetle population quality and update the immune memory bank; Step B5: Iterate step B2-step B4 to obtain the optimal hyperparameter configuration.

Images