Electric wire and cable protection plastic pipe fault prediction method based on deep learning

Abstract

The invention discloses an electric wire and cable protection plastic pipe fault prediction method based on deep learning, and the method comprises the steps: S1, installing a plurality of sensors, and collecting all data in a cable operation process; s2, preprocessing the collected sensor data; s3, performing feature learning on the input data by adopting a transformation type auto-encoder, and realizing anomaly detection and reconstruction of fault data through a sequence generation model; s4, constructing a fault prediction model, and performing comprehensive analysis on fault features; s5, generating normal and abnormal fault data samples by using the improved generative adversarial network, and performing enhanced training on the fault prediction model through the extended training set; and S6, performing fault prediction on the state of the electric wire and cable protection plastic pipe, and pushing alarm information in real time. According to the method, an efficient and scientific optimization scheme can be provided in fault prediction of the electric wire and cable protection plastic pipe, and remarkable technical value and economic benefits are brought to practical application.

Description

Technical Field

[0001] The present invention relates to the technical field of wire and cable, and particularly relates to a method for predicting faults of a wire and cable protection plastic pipe based on deep learning. Background Art

[0002] With the acceleration of the industrialization process, the scope of use and the number of wire and cable are constantly expanding, and the fault problems of wire and cable are becoming increasingly serious. As an important part of infrastructure such as power and communication, the faults of wire and cable will not only affect the normal work process, but may also trigger major safety accidents such as fires and explosions. Therefore, timely detection and prediction of wire and cable faults, especially the faults occurring in the wire and cable protection plastic pipe, play a crucial role in ensuring the normal operation of the system and avoiding accidents.

[0003] At present, the fault detection methods of wire and cable mainly include physical inspection, sensor monitoring, and intelligent detection based on machine learning. The physical inspection method mainly relies on manual inspection and regular maintenance. This method is not only inefficient but also unable to detect faults in real time, and often only discovered when serious faults occur. The sensor monitoring technology monitors the state of the cable in real time by installing sensors such as temperature, humidity, vibration, and current at key parts of the cable. The sensor monitoring system can effectively collect various parameters during the operation of the cable. Especially for the monitoring of the wire and cable protection plastic pipe, it can provide timely data support. However, the existing sensor monitoring technology has certain defects, such as unreasonable sensor layout and too low data acquisition frequency, resulting in untimely discovery of fault information and inability to accurately predict faults.

[0004] The intelligent detection method based on machine learning has been widely used in recent years. Especially the introduction of deep learning technology provides a more efficient and accurate solution for the fault detection of wire and cable. The deep learning model can extract complex features from a large amount of historical data through training, and then predict and classify cable faults. Traditional deep learning methods usually require a large amount of labeled data, which puts forward higher requirements for the acquisition and annotation of wire and cable fault data. Moreover, the existing cable fault prediction methods based on deep learning often rely on fixed data sets and lack effective data augmentation strategies, resulting in poor generalization ability of the model and inability to cope with diverse fault patterns. In addition, the existing fault prediction methods often only focus on single monitoring data and ignore the complexity of the internal and external environments of the cable protection plastic pipe. The faults of the cable protection plastic pipe may be caused by various factors, such as temperature changes, humidity changes, mechanical vibrations, and fluctuations in electrical loads. Traditional fault prediction methods generally analyze through single-dimensional monitoring data and fail to effectively fuse multiple sensor data, resulting in weak prediction ability for multi-dimensional faults.

[0005] In summary, the existing cable fault prediction technologies still have many deficiencies in terms of the accuracy, timeliness, and data diversity of fault detection. Traditional monitoring technologies and deep learning methods are difficult to solve the problem of identifying complex fault patterns, and there is a lack of effective data enhancement mechanisms to improve the generalization ability of the model. In view of these defects, the present invention proposes a fault prediction method for wire and cable protection plastic pipes based on deep learning. Through technological innovations such as multi-modal data fusion, the introduction of self-attention mechanisms, and generative adversarial network data enhancement, it can effectively improve the accuracy and robustness of fault prediction and overcome the limitations in the existing technologies. Summary of the Invention

[0006] An object of the present invention is to propose a fault prediction method for wire and cable protection plastic pipes based on deep learning. The present invention can provide an efficient and scientific optimization scheme in the fault prediction of wire and cable protection plastic pipes, bringing significant technical value and economic benefits to practical applications.

[0007] A fault prediction method for wire and cable protection plastic pipes based on deep learning according to an embodiment of the present invention includes the following steps:

[0008] S1. Install a variety of sensors on the wire and cable protection plastic pipes to collect various data during the operation of the cables in real time;

[0009] S2. Clean the collected sensor data and smooth the data using the wavelet transform method;

[0010] S3. Use a variational autoencoder to perform feature learning on the input data, and implement anomaly detection and reconstruction of fault data through a sequence generation model;

[0011] S4. Construct a fault prediction model, perform fault feature modeling on the sensor data by combining a convolutional neural network, a long short-term memory network, and a graph neural network, realize the deep joint learning of the sensor data, and comprehensively analyze the fault features;

[0012] S5. Use an improved generative adversarial network to generate normal and abnormal fault data samples. The generator of the improved generative adversarial network adopts a structure combining a convolutional neural network and a self-attention mechanism to enhance the generation ability. By expanding the training set, and using the newly generated training set to perform enhanced training on the fault prediction model;

[0013] S6. Apply the trained multi-fault prediction model to the real-time monitoring data to perform fault prediction on the state of the wire and cable protection plastic pipes. According to the prediction results output by the model, determine whether the cable has a fault. If a fault occurs, trigger an alarm mechanism and push alarm information to relevant personnel in real time.

[0014] Optionally, the S1 includes the following steps:

[0015] S11. Install multiple sensors on the wire and cable protection plastic pipe, including: temperature sensor, humidity sensor, vibration sensor, current sensor, voltage sensor, for real-time collection of temperature data, humidity data, vibration signal data, current signal, voltage signal of the wire and cable protection pipe;

[0016] S12, the sensor is connected to a data acquisition system via a wired network, and the data acquisition system collects temperature, humidity, vibration, current and voltage data in real time and uploads them to a central control unit;

[0017] S13. The time interval for sensor data collection is T. The time interval T is set according to the characteristics of cable operation and fault detection requirements. T is an adjustable parameter, and the value range of T is 1 second to 3600 seconds.

[0018] Optionally, S2 includes the following steps:

[0019] S21, performing missing value processing on the collected temperature, humidity, vibration, current and voltage data, and using linear interpolation to fill in the missing values ​​in the data, wherein the linear interpolation method is to infer the missing values ​​based on the linear relationship between adjacent data points;

[0020] S22, performing outlier detection on the collected sensor data, by setting the upper and lower thresholds of the data, using the Z-score method to identify and remove outliers, the Z value calculation formula in the Z-score method is:

[0021]

[0022] Among them, X i is the i-th data point, μ is the mean of the data, σ is the standard deviation of the data, and if the Z value is greater than the set threshold, it is judged as an outlier;

[0023] S23, using wavelet transform method to smooth the collected sensor data;

[0024] S24, normalizing data from different sensors.

[0025] Optionally, S3 includes the following steps:

[0026] S31, using a transformative autoencoder to perform feature learning on input data, where the input data is cleaned sensor data, including temperature, humidity, vibration, current and voltage data, and the autoencoder performs low-dimensional representation compression on the input data through an encoder, reconstructs the input data through a decoder, and performs self-supervised learning;

[0027] S32. In the variational autoencoder, the encoder part adopts a multi-layer self-attention mechanism. By calculating the attention weights of each element of the input data to other elements, a weighted feature representation is generated. The calculation formula is:

[0028]

[0029] where Q is the query vector, K is the key vector, V is the value vector, and d k is the dimension of the key vector, and softmax is the normalization function;

[0030] S33. The decoder part adopts a reverse self-attention mechanism to restore the low-dimensional representation generated by the encoder to the original high-dimensional data. The reconstruction loss function is:

[0031]

[0032] where X i is the original data of the i-th sample, is the data reconstructed by the decoder, N is the total number of samples, and optimization is performed by minimizing the reconstruction error;

[0033] S34. The variational autoencoder of the sequence generation model is used to perform anomaly detection and reconstruction on the fault data. The variational autoencoder uses a probability model to model the data, and the distribution of the latent variables adopts a Gaussian distribution. The inference formula of the latent space is:

[0034] q(z|x) = N(z; μ(x), σ 2 (x));

[0035] where q(z|x) is the posterior distribution of the latent variable z of the input data x, μ(x) is the mean, and σ 2 (x) is the variance. The variational autoencoder is trained by maximizing the lower bound;

[0036] S35. By optimizing the learned low-dimensional representation and the reconstruction process, the representation ability of the fault data is improved.

[0037] Optionally, the S4 includes the following steps:

[0038] S41. Construct a fault prediction model. Process the image and time series data in the sensor data through a convolutional neural network to extract low-level local features, and gradually extract deeper abstract features through multiple convolutional layers and pooling layers;

[0039] S42. Use a long short-term memory network to model the cleaned time series data, learn the long-term dependencies between each time step in the data, and represent the hidden state sequence of each time step as:

[0040] h t = LSTM(x t , h t-1 );

[0041] Among them, h t is the hidden state at the current time step t, x t is the data input at the current time step, and h t-1 is the hidden state of the previous time step;

[0042] S43. Introduce a graph neural network to model the relationships between sensors, establish a graph structure among sensor data, and use a graph convolutional network to aggregate the graph node features to generate a comprehensive feature representation for each graph node. The graph node feature update formula is:

[0043]

[0044] Among them, is the feature representation of node i at the l+1 layer, N(i) is the set of neighbor nodes of node i, c ij is the normalization coefficient, W (l) is the weight matrix at the l layer, and σ is the sigmoid activation function;

[0045] S44. Perform multi-layer fusion on the output features of the convolutional neural network, long short-term memory network, and graph neural network, and use a multi-layer perceptron to learn and classify each feature to finally generate a fault prediction result;

[0046] S45. Use the backpropagation algorithm to calculate the loss function and update the network parameters. The loss function is:

[0047]

[0048] Among them, N is the number of samples, y i is the true label of the i-th sample, is the model prediction value, and finally adjust the weights through the Adam optimization algorithm.

[0049] Optionally, the S5 includes the following steps:

[0050] S51. Use an improved generative adversarial network to generate normal and abnormal cable fault data samples. The generator adopts a structure combining a convolutional neural network and a self-attention mechanism to enhance the generation ability of the generator. The generation process of the generator is:

[0051]

[0052] Among them, G(z) is the output of the generator, is the generator network, θG \(G\) is the network parameter of the generator, \(z\) is the input vector of the latent space, and the generator maps the latent vector \(z\) through a deep convolutional neural network to generate cable fault data samples close to the real ones;

[0053] S52. Conditionally constrain the noise input in the generator. Combine the fault type label and sensor data as conditional inputs, and use a conditional generative adversarial network to generate cable fault data of specific types. The generator generates sample data with consistency and accuracy according to different fault types. The combination process of the conditional input and the noise input of the generator is as follows:

[0054]

[0055] Among them, \(G(z, c)\) is the output of the generator, and \(c\) is the conditional variable. \(G\) is the generator network, \(z\) is the input vector of the latent space, and \(\theta\) G is the network parameter of the generator. The generator generates fault data of specific types according to the given conditional variable \(c\) and random noise \(z\);

[0056] S53. Perform multi-dimensional perturbation processing on the generated fault data by introducing a perturbation mechanism to simulate the operating characteristics of the cable under different fault scenarios. The perturbation operations include adding random noise or scaling to the eigenvalue of temperature, humidity, vibration, current, and voltage to simulate various changes in the environment. The perturbation process is as follows:

[0057] X perturbed \( = X+\epsilon\cdot\Delta X\);

[0058] Among them, \(X\) perturbed is the data sample after perturbation, \(X\) is the original generated data, \(\epsilon\) is the perturbation coefficient, and \(\Delta X\) is the perturbation amount generated by random noise, simulating different fault modes and environmental changes;

[0059] S54. Mix the generated abnormal data samples with the original data samples and use a weighted fusion strategy to enhance the diversity of the data set:

[0060] X mixed \( = \alpha\cdot X\) real \(+(1 - \alpha)\cdot X\) generated ;

[0061] Among them, \(X\) mixed is the data sample after mixing, \(X\) real is the original data sample, \(X\) generated is the generated data sample, \(\alpha\) is the weight coefficient, \(\alpha\in[0, 1]\), and it controls the proportion of the original data and the generated data in the mixed sample;

[0062] S55. Combine the generated data samples with the original data to form a new training set, and train the fault prediction model on the new training set.

[0063] Optionally, S6 includes the following steps:

[0064] S61. Apply the trained fault prediction model to the sensor data collected in real time, and perform fault prediction according to the output of the fault prediction model;

[0065] S62. According to the predicted value, judge the state of the wire and cable protection plastic pipe. If the predicted value exceeds the set threshold, it is determined to be in a fault state;

[0066] S63. If the prediction result is a fault state, trigger an alarm mechanism and generate a fault alarm message:

[0067] Alert = {fault_type, fault_location, severity_level, predicted_time};

[0068] Among them, fault_type represents the fault type, fault_location represents the fault location, severity_level represents the severity of the fault, and predicted_time represents the predicted time of the fault occurrence;

[0069] S64. Sort the alarm information by priority. The formula for calculating the priority is:

[0070] Priority = f(severity_level, predicted_time);

[0071] Among them, severity_level represents the severity of the fault, predicted_time represents the predicted time of the fault occurrence, f(.) is the priority function calculated according to the severity and predicted time, and the alarm information with a higher priority is processed first;

[0072] S65. Transmit the alarm information to the maintenance personnel or system administrator through the communication system. The communication system includes SMS, email, and mobile application push, and different transmission methods are selected for notification according to the urgency of the alarm information;

[0073] S66. Determine the response strategy according to the alarm priority. The response strategy includes immediate response to emergency faults, regular inspection of non-emergency faults, and formulation of fault location and repair plans.

[0074] The beneficial effects of the present invention are:

[0075] (1) By introducing deep learning technologies, especially multi-modal data fusion, convolutional neural networks, long short-term memory networks, graph neural networks, and self-attention mechanisms, the present invention effectively improves the prediction accuracy and timeliness of cable faults. Different from traditional methods that rely on a single data source, the present invention makes full use of various sensor data, such as temperature, humidity, vibration, current, and voltage, etc. By comprehensively analyzing these data through a deep learning model, it can identify the occurrence of cable faults more comprehensively and accurately. This multi-dimensional data fusion and processing method can effectively extract the fault characteristics of cable protection plastic pipes, reduce the errors caused by single-sensor information, and thus improve the accuracy of fault prediction.

[0076] (2) The present invention can overcome the problem of insufficient data. By generating virtual data samples close to real data and mixing them with the actual data of sensors, the present invention effectively expands the training dataset and enhances the robustness of the deep learning model. The generated data can not only simulate various fault types but also cover a wider range of fault modes, enabling the model to maintain a high prediction ability when facing different cable fault scenarios. In addition, by combining the self-attention mechanism, the generator can capture the local features of the data during the generation process and optimize the quality of the generated data, further enhancing the effect of the data augmentation process.

[0077] (3) Through the improved adversarial training mechanism of the generator and discriminator, the present invention minimizes the difference between the generated data samples and the original data. The generated data has higher credibility and can help the model better learn and identify different patterns of cable faults. This innovative method of generating data not only expands the scale of the dataset but also ensures a high degree of consistency between the generated data and the actual fault modes, enhancing the accuracy and reliability of the model in practical applications. Description of the Drawings

[0078] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0079] Figure 1 is a flowchart of a method for predicting faults in cable protection plastic pipes based on deep learning proposed by the present invention;

[0080] Figure 2 is a flowchart of generating a new training set using an improved generative adversarial network in a method for predicting faults in cable protection plastic pipes based on deep learning proposed by the present invention. Detailed Embodiments

[0081] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only schematically showing the basic structure of the present invention, and therefore only showing the components related to the present invention.

[0082] Reference Figure 1 - Figure 2 , a fault prediction method for wire and cable protection plastic pipes based on deep learning, comprising the following steps:

[0083] S1. Install a variety of sensors on the wire and cable protection plastic pipes to collect various data during the operation of the cables in real time;

[0084] S2. Clean the collected sensor data and smooth the data using the wavelet transform method;

[0085] S3. Use a variational autoencoder to perform feature learning on the input data, and implement anomaly detection and reconstruction of fault data through a sequence generation model;

[0086] S4. Build a fault prediction model, perform fault feature modeling on the sensor data by combining a convolutional neural network, a long short-term memory network, and a graph neural network, achieve deep joint learning of the sensor data, and comprehensively analyze the fault features;

[0087] S5. Use an improved generative adversarial network to generate normal and abnormal fault data samples. The generator of the improved generative adversarial network adopts a structure combining a convolutional neural network and a self-attention mechanism to enhance the generation ability. By expanding the training set and using the newly generated training set to perform enhanced training on the fault prediction model;

[0088] S6. Apply the trained multi-fault prediction model to real-time monitoring data, perform fault prediction on the state of the wire and cable protection plastic pipes, judge whether the cable has a fault according to the prediction result output by the model. If a fault occurs, trigger an alarm mechanism and push alarm information to relevant personnel in real time.

[0089] In this embodiment, S1 includes the following steps:

[0090] S11. Install a plurality of sensors on the wire and cable protection plastic pipes, including: a temperature sensor, a humidity sensor, a vibration sensor, a current sensor, and a voltage sensor, for collecting temperature data, humidity data, vibration signal data, current signal, and voltage signal of the wire and cable protection pipeline in real time;

[0091] S12. The sensors are connected to the data acquisition system through a wired network, and the data acquisition system collects and uploads the temperature, humidity, vibration, current, and voltage data to the central control unit in real time;

[0092] S13. The time interval for sensor data collection is T. The time interval T is set according to the characteristics of cable operation and fault detection requirements. T is an adjustable parameter, and the value range of T is 1 second to 3600 seconds.

[0093] In this implementation, S2 includes the following steps:

[0094] S21, performing missing value processing on the collected temperature, humidity, vibration, current and voltage data, and using linear interpolation to fill in the missing values ​​in the data, wherein the linear interpolation method is to infer the missing values ​​based on the linear relationship between adjacent data points;

[0095] S22, performing outlier detection on the collected sensor data, by setting the upper and lower thresholds of the data, using the Z-score method to identify and remove outliers, the Z value calculation formula in the Z-score method is:

[0096]

[0097] Among them, X i is the i-th data point, μ is the mean of the data, σ is the standard deviation of the data, and if the Z value is greater than the set threshold, it is judged as an outlier;

[0098] S23, using wavelet transform method to smooth the collected sensor data;

[0099] S24, normalizing data from different sensors.

[0100] In this implementation, S3 includes the following steps:

[0101] S31, using a transformative autoencoder to perform feature learning on input data, where the input data is cleaned sensor data, including temperature, humidity, vibration, current and voltage data, and the autoencoder performs low-dimensional representation compression on the input data through an encoder, reconstructs the input data through a decoder, and performs self-supervised learning;

[0102] S32. In the transformative autoencoder, the encoder part adopts a multi-layer self-attention mechanism to generate a weighted feature representation by calculating the attention weight of each element of the input data to other elements. The calculation formula is:

[0103]

[0104] Among them, Q is the query vector, K is the key vector, V is the value vector, and d k is the dimension of the key vector, and softmax is the normalization function;

[0105] S33. The decoder part adopts the inverse self-attention mechanism to restore the low-dimensional representation generated by the encoder to the original high-dimensional data. The reconstruction loss function is as follows:

[0106]

[0107] Among them, X i is the original data of the i-th sample, is the data reconstructed by the decoder, and N is the total number of samples. It is optimized by minimizing the reconstruction error;

[0108] S34. The variational autoencoder of the sequence generation model is used to perform anomaly detection and reconstruction on the fault data. The variational autoencoder uses a probability model to model the data, and the distribution of the latent variables adopts a Gaussian distribution. The inference formula of the latent space is as follows:

[0109] q(z|x) = N(z; μ(x), σ 2 (x));

[0110] Among them, q(z|x) is the posterior distribution of the latent variable z of the input data x, μ(x) is the mean, and σ 2 (x) is the variance. The variational autoencoder is trained by maximizing the lower bound;

[0111] S35. By optimizing the learned low-dimensional representation and the reconstruction process, the representation ability of the fault data is improved.

[0112] In this embodiment, S4 includes the following steps:

[0113] S41. Construct a fault prediction model. Process the image and time series data in the sensor data through a convolutional neural network to extract low-level local features, and gradually extract deeper abstract features through multiple convolutional layers and pooling layers;

[0114] S42. Use a long short-term memory network to model the cleaned time series data, learn the long-term dependencies between each time step in the data, and represent the hidden state sequence of each time step as:

[0115] h t = LSTM(x t , h t-1 );

[0116] Among them, h t is the hidden state at the current time step t, x t is the data input at the current time step, and h t-1 is the hidden state at the previous time step;

[0117] S43. Introduce a graph neural network to model the relationships between sensors, establish a graph structure among sensor data, and use a graph convolutional network to aggregate the features of graph nodes to generate a comprehensive feature representation for each graph node. The graph node feature update formula is as follows:

[0118]

[0119] Where, is the feature representation of node i at the l+1 layer, N(i) is the set of neighbor nodes of node i, c ij is the normalization coefficient, W (l) is the weight matrix at the l layer, and σ is the sigmoid activation function;

[0120] S44. Multilayer fuse the output features of the convolutional neural network, long short-term memory network, and graph neural network, and use a multi-layer perceptron to learn and classify each feature, and finally generate a fault prediction result;

[0121] S45. Use the backpropagation algorithm to calculate the loss function and update the network parameters. The loss function is:

[0122]

[0123] Where, N is the number of samples, y i is the true label of the i-th sample, is the model prediction value, and finally adjust the weights through the Adam optimization algorithm.

[0124] In this embodiment, S5 includes the following steps:

[0125] S51. Use an improved generative adversarial network to generate normal and abnormal cable fault data samples. The generator adopts a structure combining a convolutional neural network and a self-attention mechanism to enhance the generation ability of the generator. The generation process of the generator is as follows:

[0126]

[0127] Where, G(z) is the output of the generator, is the generator network, θ G is the network parameter of the generator, z is the input vector of the latent space, and the generator maps the latent vector z through a deep convolutional neural network to generate cable fault data samples close to the real ones;

[0128] S52. Conditionally constrain the noise input in the generator. Combine the fault type label and sensor data as conditional inputs, and use a conditional generative adversarial network to generate cable fault data of specific types. The generator generates sample data with consistency and accuracy according to different fault types. The combination process of the conditional input and the noise input of the generator is as follows:

[0129]

[0130] Among them, G(z, c) is the output of the generator, c is the conditional variable, is the generator network, z is the input vector in the latent space, θ G is the network parameter of the generator. The generator generates fault data of specific types according to the given conditional variable c and random noise z;

[0131] S53. Perform multi-dimensional perturbation processing on the generated fault data by introducing a perturbation mechanism to simulate the operating characteristics of the cable under different fault scenarios. The perturbation operations include adding random noise or scaling to the eigenvalue of temperature, humidity, vibration, current, and voltage to simulate various changes in the environment. The perturbation process is as follows:

[0132] X perturbed = X + ∈·ΔX;

[0133] Among them, X perturbed is the perturbed data sample, X is the original generated data, ∈ is the perturbation coefficient, and ΔX is the perturbation amount generated by random noise, simulating different fault modes and environmental changes;

[0134] S54. Mix the generated abnormal data samples with the original data samples and use a weighted fusion strategy to enhance the diversity of the dataset:

[0135] X mixed = α·X real + (1 - α)·X generated ;

[0136] Among them, X mixed is the mixed data sample, X real is the original data sample, X generated is the generated data sample, α is the weight coefficient, α ∈ [0, 1], controlling the proportion of the original data and the generated data in the mixed sample;

[0137] S55. Combine the generated data samples with the original data to form a new training set, and train the fault prediction model on the new training set.

[0138] In this embodiment, S6 includes the following steps:

[0139] S61. Apply the trained fault prediction model to the sensor data collected in real time, and perform fault prediction based on the output of the fault prediction model;

[0140] S62. According to the predicted value, judge the state of the wire and cable protection plastic pipe. If the predicted value exceeds the set threshold, it is determined to be a fault state;

[0141] S63. If the prediction result is a fault state, trigger the alarm mechanism and generate a fault alarm message:

[0142] Alert = {fault_type, fault_location, severity_level, predicted_time};

[0143] Among them, fault_type represents the fault type, fault_location represents the fault location, severity_level represents the severity of the fault, and predicted_time represents the predicted time of the fault occurrence;

[0144] S64. Sort the alarm information by priority. The formula for calculating the priority is:

[0145] Priority = f(severity_level, predicted_time);

[0146] Among them, severity_level represents the severity of the fault, predicted_time represents the predicted time of the fault occurrence, f(.) is the priority function calculated according to the severity and predicted time, and the alarm information with a higher priority is processed first;

[0147] S65. Transmit the alarm information to the maintenance personnel or system administrator through the communication system. The communication system includes SMS, email, and mobile application push, and different transmission methods are selected for notification according to the urgency of the alarm information;

[0148] S66. Determine the response strategy according to the alarm priority. The response strategy includes immediate response to urgent faults, regular inspection of non-urgent faults, and formulation of fault location and repair plans.

[0149] Example:

[0150] Example In a certain industrial park, there are a large number of wire and cable protection plastic pipes in the high-voltage power transmission network of the industrial park. Due to external environmental factors and the normal working load of the cables, these cable pipes often experience different types of failures. To ensure the stability and safety of power supply, the industrial park decides to implement a fault prediction system for wire and cable protection plastic pipes based on deep learning. This system uses the method of the present invention based on multi-modal data fusion and a dataset enhanced by a generative adversarial network to improve the accuracy, timeliness, and robustness of fault prediction.

[0151] The power systems in multiple areas within the industrial park have adopted this fault prediction system. Sensors are arranged at key nodes of the wires and cables and inside and outside the cable protection plastic pipes to collect data in real time. The sensors include temperature sensors, humidity sensors, vibration sensors, current sensors, and voltage sensors. Through the multi-dimensional data collected by these sensors, including parameters such as the current, voltage, temperature, humidity, and vibration of the cables, the operating status of the cables can be comprehensively monitored.

[0152] In this embodiment, the system first collects relevant data during the operation of the cables in real time through a variety of sensors installed around the cables. For example, temperature sensors are used to monitor the temperature changes inside and outside the cables, humidity sensors monitor the humidity inside and outside the plastic pipes, vibration sensors are used to detect the mechanical vibration of the cables during operation, and current and voltage sensors are used to monitor the electrical characteristics of the cables. Through these sensors, the data obtained in real time is uploaded to the central processing unit of the system for analysis.

[0153] To better predict the faults of wire and cable protection plastic pipes, the system uses the deep learning-based fault prediction method of the present invention. By inputting the collected multi-dimensional data into the deep learning model, the system can analyze potential fault patterns.

[0154] During the prediction process, the fault type given by the system is "cable short circuit caused by overheating", and it accurately indicates the specific location where the fault occurred and the severity of the fault. According to the prediction results, after receiving the alarm information, the maintenance personnel immediately go to the fault point for inspection and timely discover the damage to some areas of the cable protection plastic pipes due to excessive wear and high temperature, avoiding a larger-scale power outage and safety hazards caused by cable short circuits.

[0155] To further verify the performance of this system, a large amount of on-site data was collected in the experiment and detailed analysis was carried out. The following Table 1 shows some data for verifying the accuracy of the system's fault prediction:

[0156] Table 1 Data for Verifying the Accuracy of the System's Fault Prediction

[0157]

[0158] Throughout the embodiments, the implementer not only solves the deficiencies of traditional wire and cable fault prediction technologies in terms of accuracy, timeliness, and data diversity through the method of the present invention, but also significantly improves the reliability and robustness of fault prediction. The present invention solves the problems of insufficient dataset and low prediction accuracy in traditional cable fault prediction methods by introducing an innovative method that combines deep learning and generative adversarial networks.

[0159] Through multi-modal data fusion technology, the present invention combines sensor data such as temperature, humidity, vibration, current, and voltage to establish a more comprehensive fault prediction model. Compared with traditional single data analysis methods, the present invention can capture complex fault patterns inside and outside the cable protection plastic pipe from multiple dimensions, thereby improving the prediction accuracy. By introducing the self-attention mechanism, the present invention further optimizes the local features of the data, enabling the system to more accurately capture the detailed information of cable faults and avoiding prediction errors caused by traditional methods ignoring the correlation between data.

[0160] The present invention enhances and generates the original data through an improved generative adversarial network, overcoming the problem of insufficient cable fault data volume. The generator generates cable fault data samples with high consistency and diversity through the self-attention mechanism and conditional generative adversarial network, effectively expanding the training dataset, avoiding overfitting, and improving the generalization ability of the deep learning model. In addition, the data augmentation method can generate corresponding data samples according to different fault types and environmental changes, ensuring that the model can handle various complex fault patterns.

[0161] In response to the actual needs of wire and cable protection plastic pipe fault prediction in the power system, the present invention comprehensively considers multiple factors such as temperature and humidity fluctuations, vibration changes, and electrical loads. Through an innovative method that combines a deep learning model and a generative adversarial network, it ensures high accuracy and real-time performance of the prediction. By dynamically adjusting the strategy of generating data and combining real-time monitoring data, the model can adapt to changes in cable fault patterns under different environmental conditions and always maintain a high prediction accuracy.

[0162] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes, shall be covered by the protection scope of the present invention.

Claims

1. A method for predicting faults of wire and cable protective plastic pipes based on deep learning, characterized in that: The steps include: S1. Install a variety of sensors on the plastic pipes protecting wires and cables to collect various data during the operation of the cables in real time; S2, cleaning the collected sensor data and smoothing the data using wavelet transform method; S3, using transformation autoencoders to learn features of input data, and realizing anomaly detection and reconstruction of fault data through sequence generation models; S4. Build a fault prediction model, model the fault characteristics of sensor data by combining convolutional neural networks, long short-term memory networks, and graph neural networks, realize deep joint learning of sensor data, and conduct comprehensive analysis of fault characteristics; S5. Generate normal and abnormal fault data samples using an improved generative adversarial network, wherein the generator of the improved generative adversarial network adopts a structure based on a combination of a convolutional neural network and a self-attention mechanism to enhance the generation capability, and perform enhanced training on the fault prediction model by expanding the training set and using the newly generated training set; S6. Apply the trained multi-fault prediction model to real-time monitoring data, predict the status of the plastic pipes protecting wires and cables, and determine whether the cable is faulty based on the prediction results output by the model. If a fault occurs, the alarm mechanism is triggered and the alarm information is pushed to relevant personnel in real time.

2. According to the method for predicting faults of wire and cable protective plastic pipes based on deep learning in claim 1, it is characterized in that: The S1 comprises the following steps: S11. Install multiple sensors on the wire and cable protection plastic pipe, including: temperature sensor, humidity sensor, vibration sensor, current sensor, voltage sensor, for real-time collection of temperature data, humidity data, vibration signal data, current signal, voltage signal of the wire and cable protection pipe; S12, the sensor is connected to a data acquisition system via a wired network, and the data acquisition system collects temperature, humidity, vibration, current and voltage data in real time and uploads them to a central control unit; S13. The time interval for sensor data collection is T. The time interval T is set according to the characteristics of cable operation and fault detection requirements. T is an adjustable parameter, and the value range of T is 1 second to 3600 seconds.

3. According to a method for predicting faults of wire and cable protective plastic pipes based on deep learning in claim 1, it is characterized in that: The S2 comprises the following steps: S21, performing missing value processing on the collected temperature, humidity, vibration, current and voltage data, and using linear interpolation to fill in the missing values ​​in the data, wherein the linear interpolation method is to infer the missing values ​​based on the linear relationship between adjacent data points; S22, performing outlier detection on the collected sensor data, by setting the upper and lower thresholds of the data, using the Z-score method to identify and remove outliers, the Z value calculation formula in the Z-score method is: Among them, X i is the i-th data point, μ is the mean of the data, σ is the standard deviation of the data, and if the Z value is greater than the set threshold, it is judged as an outlier; S23, using wavelet transform method to smooth the collected sensor data; S24, normalizing data from different sensors.

4. The method for predicting faults of wire and cable protective plastic pipes based on deep learning according to claim 1 is characterized in that: The S3 comprises the following steps: S31, using a transformative autoencoder to perform feature learning on input data, where the input data is cleaned sensor data, including temperature, humidity, vibration, current and voltage data, and the autoencoder performs low-dimensional representation compression on the input data through an encoder, reconstructs the input data through a decoder, and performs self-supervised learning; S32. In the transformative autoencoder, the encoder part adopts a multi-layer self-attention mechanism to generate a weighted feature representation by calculating the attention weight of each element of the input data to other elements. The calculation formula is: Among them, Q is the query vector, K is the key vector, V is the value vector, and d k is the dimension of the key vector, and softmax is the normalization function; S33, the decoder part uses the reverse self-attention mechanism to restore the low-dimensional representation generated by the encoder to the original high-dimensional data, and the reconstruction loss function is: Among them, X i is the original data of the ith sample, is the data reconstructed by the decoder, N is the total number of samples, and the optimization is performed by minimizing the reconstruction error; S34. The fault data is anomaly detected and reconstructed through the sequence generation model variational autoencoder. The variational autoencoder uses a probability model to model the data. The distribution of the latent variable uses a Gaussian distribution. The inference formula of the latent space is: q(z|x)=N(z;μ(x),σ 2 (x)); Among them, q(z|x) is the posterior distribution of the latent variable z of the input data x, μ(x) is the mean, σ 2 (x) is the variance, and the variational autoencoder is trained by maximizing the lower bound; S35. Improve the representation capability of fault data by optimizing the low-dimensional representation and reconstruction process of learning.

5. The method for predicting faults of wire and cable protective plastic pipes based on deep learning according to claim 1 is characterized in that: The S4 comprises the following steps: S41. Build a fault prediction model, process the image and time series data in the sensor data through a convolutional neural network, extract low-level local features, and gradually extract deeper abstract features through multiple convolutional layers and pooling layers; S42. Use the long short-term memory network to model the cleaned time series data, learn the long-term dependencies between time steps in the data, and represent the hidden state sequence of each time step as: h t =LSTM(x t ,h t-1 ); Among them, h t is the hidden state at the current time step t, x t is the data input for the current time step, h t-1 is the hidden state of the previous time step; S43. Introduce graph neural network to model the relationship between sensors, establish the graph structure between sensor data, and use graph convolution network to aggregate graph node features to generate a comprehensive feature representation of each graph node. The graph node feature update formula is: in, is the feature representation of node i at layer l+1, N(i) is the set of neighbor nodes of node i, c ij is the normalization coefficient, W (l) is the weight matrix of the lth layer, σ is the sigmoid activation function; S44, performing multi-layer fusion of the output features of the convolutional neural network, the long short-term memory network and the graph neural network, using a multi-layer perceptron to learn and classify each feature, and finally generating a fault prediction result; S45, using the back propagation algorithm to calculate the loss function and update the network parameters, the loss function is: Where N is the number of samples, y i is the true label of the i-th sample, is the model prediction value, and the weight is finally adjusted through the Adam optimization algorithm.

6. The method for predicting faults of wire and cable protective plastic pipes based on deep learning according to claim 1 is characterized in that: The S5 comprises the following steps: S51. Generate normal and abnormal cable fault data samples using an improved generative adversarial network. The generator adopts a structure based on a combination of a convolutional neural network and a self-attention mechanism to enhance the generator's generation capability. The generator's generation process is as follows: Among them, G(z) is the output of the generator, is the generator network, θ G is the network parameter of the generator, z is the input vector of the latent space, and the generator maps the latent vector z through a deep convolutional neural network to generate cable fault data samples close to the real one; S52, conditionally constrain the noise input in the generator, combine the fault type label and sensor data as conditional input, use the conditional generative adversarial network to generate specific types of cable fault data, the generator generates sample data with consistency and accuracy according to different fault types, and the combination process of the conditional input of the generator and the noise input is: Among them, G(z,c) is the output of the generator, c is the conditional variable, is the generator network, z is the input vector in the latent space, θ G is the network parameter of the generator, and the generator generates a specific type of fault data according to the given conditional variable c and random noise z; S53. Perform multi-dimensional disturbance processing on the generated fault data by introducing a disturbance mechanism to simulate the operation characteristics of the cable under different fault scenarios. The disturbance operation includes adding random noise or scaling the characteristic values ​​of temperature, humidity, vibration, current, and voltage to simulate various changes in the environment. The disturbance process is: X perturbed =X+∈·ΔX; Among them, X perturbed is the disturbed data sample, X is the generated original data, ∈ is the disturbance coefficient, ΔX is the disturbance amount generated by random noise, simulating different failure modes and environmental changes; S54, mix the generated abnormal data samples with the original data samples, and use the weighted fusion strategy to enhance the diversity of the data set: X mixed =α·X real +(1-a)·X generated ; Among them, X mixed is the mixed data sample, X real is the original data sample, X generated is the generated data sample, α is the weight coefficient, α∈[0,1], which controls the ratio of original data to generated data in the mixed sample; S55. Combine the generated data samples with the original data to form a new training set, and train the fault prediction model on the new training set.

7. The method for predicting faults of wire and cable protective plastic pipes based on deep learning according to claim 1, characterized in that: The S6 comprises the following steps: S61, applying the trained fault prediction model to the sensor data collected in real time, and performing fault prediction according to the output of the fault prediction model; S62, judging the state of the plastic pipe protecting the wire and cable according to the predicted value, and judging it as a fault state if the predicted value exceeds a set threshold; S63. If the prediction result is a fault state, the alarm mechanism is triggered and fault alarm information is generated: Alert={fault_type, fault_location, severity_level, predicted_time}; Among them, fault_type indicates the fault type, fault_location indicates the fault location, severity_level indicates the severity of the fault, and predicted_time indicates the predicted time of the fault occurrence; S64. Prioritize the alarm information. The priority calculation formula is: Priority=f(severity_level,predicted_time); Among them, severity_level represents the severity of the fault, predicted_time represents the predicted time of the fault, and f(.) is the priority function calculated according to the severity and predicted time. The alarm information with higher priority will be processed first. S65, transmitting the alarm information to the maintenance personnel or system administrator through the communication system, wherein the communication system includes SMS, email, and mobile application push, and different transmission methods are selected for notification according to the urgency of the alarm information; S66. Determine a response strategy according to the alarm priority, wherein the response strategy includes immediate response to emergency faults, regular inspection of non-emergency faults, and formulation of fault location and repair plans.

Images