Power transmission component fault identification system based on deep learning

Abstract

The invention discloses a power transmission part fault identification system based on deep learning, which comprises a data acquisition module, a data preprocessing module, a feature extraction module, a model training module, a fault identification module, a fault diagnosis module and a system optimization module, and is characterized in that the data acquisition module collects data from a sensor in a power transmission system; the data preprocessing module processes the collected original data; the feature extraction module extracts features from the preprocessed data; the model training module uses the preprocessed data and the extracted features to train a deep learning model; the fault identification module carries out fault identification on the real-time data of the power transmission system; the fault diagnosis module carries out fault positioning and reason analysis; and the system optimization module continuously optimizes the whole fault identification system. According to the invention, identification and diagnosis of faults of power transmission parts are realized, and solid technical support is provided for safe and stable operation of a power system.

Description

Technical Field

[0001] The present invention relates to the field of deep learning and intelligent resources, and in particular to a power transmission component fault identification system based on deep learning. Background Art

[0002] In modern power systems, the transmission network is an important link in the transmission of electric energy, responsible for transmitting the electric energy generated by power plants to the load center. However, due to the wide distribution of transmission lines and related equipment, and the complex and changeable operating environment, transmission system failures occur from time to time. Failures in transmission components will not only cause power outages, but may also cause instability in the power grid and even cause large-scale power outages. Therefore, timely and accurate identification and diagnosis of transmission component failures are of great significance to ensure the safe and stable operation of the power system.

[0003] Traditional methods for identifying and diagnosing faults in power transmission components mainly rely on manual inspections and regular maintenance. These methods not only consume a lot of manpower and material resources, but also often fail to detect and handle faults in a timely manner due to the long detection cycle. With the development of science and technology, fault diagnosis methods based on monitoring data have gradually become a research hotspot. Various sensors are commonly installed in modern power transmission systems. These sensors can monitor the operating status of transmission lines and equipment in real time and generate a large amount of monitoring data. By analyzing these data, faults in power transmission components can be effectively identified and diagnosed. However, how to extract effective features from massive monitoring data and build efficient and accurate fault diagnosis models still faces many challenges.

[0004] In recent years, deep learning technology has made significant progress in image recognition, natural language processing, speech recognition and other fields, showing strong data processing and feature extraction capabilities. By building a multi-layer neural network, deep learning can automatically learn high-level feature representations from raw data, thereby improving the accuracy and robustness of fault diagnosis. The power transmission component fault identification system based on deep learning can realize automatic identification and diagnosis of faults through in-depth analysis of sensor data, which not only improves the accuracy and real-time performance of fault diagnosis, but also significantly improves the operating efficiency and reliability of the power transmission system, providing a strong guarantee for the safe and stable operation of the power system. Summary of the invention

[0005] In view of the above problems, the present invention aims to provide a power transmission component fault identification system based on deep learning to solve the problems raised in the above background technology.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] The present invention provides a power transmission component fault identification system based on deep learning, including a data acquisition module, a data preprocessing module, a feature extraction module, a model training module, a fault identification module, a fault diagnosis module and a system optimization module. The data acquisition module collects current, voltage, temperature and vibration data from sensors in the power transmission system; the data preprocessing module processes the collected raw data to improve the data quality and the accuracy of subsequent analysis, eliminates noise and abnormal values ​​in the data through data cleaning, data normalization and missing value filling, and converts the data into a format suitable for deep learning model input; the feature extraction module uses a deep learning algorithm to extract effective features from the preprocessed data, and automatically learns and extracts key features that can characterize the fault according to the characteristics of different power transmission components and fault types. The performance of the feature extraction module directly affects the accuracy of fault identification; the model training module uses the preprocessed data and extracted features to train the deep learning model; the fault identification module performs fault identification on the real-time data of the transmission system, monitors the operating status of the transmission components in real time, and can quickly and accurately identify faults, helping operation and maintenance personnel to find problems in a timely manner; the fault diagnosis module analyzes the specific cause and location of the fault, and locates and analyzes the cause of the fault by combining the fault identification results and the structural information of the transmission system; the system optimization module continuously optimizes the entire fault identification system, regularly evaluates the system performance, identifies the weak links in the system, and proposes improvement plans. The system optimization module continuously improves the accuracy and stability of the fault identification system by increasing data samples, updating model algorithms, and adjusting system parameters.

[0008] Furthermore, current, voltage, temperature and vibration data are collected from current sensors, voltage sensors, temperature sensors and vibration sensors in the transmission system. A distributed sensor network is used to effectively cover a wide range of transmission areas. Sensor nodes are deployed at nodes of transmission lines and components, and data is transmitted to the central processing unit through NB-IoT technology. Edge computing can reduce the computing burden of the central processing unit and improve the real-time performance of data processing. The data acquisition module adopts multi-sensor fusion technology. Multi-sensor fusion can effectively reduce the measurement error and failure rate of a single sensor by comprehensively processing and calibrating data from different sensors. The data acquisition module uses the Hadoop big data platform to realize distributed storage and parallel processing of massive data, and combines the use of NoSQL database and time series database to realize efficient storage and query of real-time data and historical data.

[0009] Furthermore, the data preprocessing module processes and optimizes the collected raw data, including data cleaning, normalization, and missing value filling:

[0010] (1) Data cleaning

[0011] Data cleaning is used to remove noise and outliers in the data. The Isolation Forest method is used to construct a random decision tree to isolate data points. Outliers are usually shallow in the tree because they are easy to be isolated. Assume that the data set is ,in, Representation dataset The data points are constructed by building multiple random trees and calculating the average path length of each data point in each tree to determine its abnormality. The expression is:

[0012]

[0013] in, Represents data points In the The path length in a tree, n is the number of trees, is the path length threshold, and data points with shorter path length are considered as outliers;

[0014] (2) Normalization

[0015] Data normalization converts data of different dimensions and ranges to a unified scale. By using an autoencoder, the low-dimensional representation of the data is learned by training a neural network, and the normalization process can be performed adaptively. The loss function of the autoencoder is for:

[0016]

[0017] in, , denote the encoder and decoder functions respectively, Representation dataset data points;

[0018] (3) Missing value filling

[0019] Generate data samples for missing value filling:

[0020]

[0021] in, denote the discriminator and the generator respectively, is the real data distribution, is the noise distribution, z is the data point in the noise distribution, Indicates that the independent variable x obeys The expected value under the distribution of Indicates that the data point z obeys The expected value under the distribution of Represents the optimization objective function, through adversarial training, the generator Ability to generate data samples to fill missing values;

[0022] After data cleaning, normalization and missing value filling, the dataset Updated to .

[0023] Furthermore, the feature extraction module performs feature learning on the input preprocessed data. The features include edges, textures, shapes, time dependencies, trends, and periodicity. It combines supervised learning and extracts low-dimensional feature representations by reducing the dimensionality of the data and compressing the features. Through the encoding and decoding process, the dimension of the feature extraction module data is effectively reduced while retaining key feature information. The low-dimensional feature representation after dimensionality reduction not only reduces the computational complexity, but also improves the training efficiency of the model and the accuracy of fault diagnosis.

[0024] Furthermore, the model training module builds the LSTM model, which includes:

[0025]

[0026]

[0027] in, is the hidden state of LSTM at time t, For LSTM The hidden state of the moment, For the updated dataset No. Features, is the output of the LSTM model, is the weight matrix, is the bias vector. During the training process, in order to avoid overfitting of the model, some neurons are randomly discarded during the training process so that the model can generalize better. Assume that the activation value of the neuron is , the generalized neuron activation value is:

[0028]

[0029] in, is the total number of neurons, Indicates the number, , To activate the probability, the learning rate is dynamically adjusted by calculating the first-order momentum and second-order momentum of the gradient:

[0030]

[0031] in, is the first-order momentum of the LSTM network gradient, is the second-order momentum of the LSTM network gradient, is the LSTM network gradient, are the LSTM network model parameters, is the evolution operator, is the first momentum coefficient, is the second momentum coefficient, is a constant approaching 0, that is .

[0032] Furthermore, the fault identification module determines whether there is a fault in the input data. In order to improve the accuracy of fault detection, a fusion method of supervised learning and unsupervised learning is adopted. Supervised learning uses a trained deep learning model to predict the input data, and its output is the fault probability. , the fault detection criteria are:

[0033]

[0034] in, is the fault detection result, To detect the threshold, the reconstruction error is calculated to determine whether the data is abnormal. The reconstruction error is:

[0035]

[0036] in, For the reconstructed data, by setting the reconstruction error threshold ,when is judged as a fault.

[0037] Furthermore, in terms of fault classification, the trained deep learning model can be used to identify the fault type. Assuming that the fault type is kind, is one of them, and the output of the model is Dimensional probability distribution , each dimension represents the probability of a fault type, and the fault classification criteria are:

[0038]

[0039] in, For the prediction results of the fault type, different weights are then assigned to different features to enhance the model's focus on key features, including:

[0040]

[0041] in, is the score function, Characterized by For query, For weighted feature representation, adversarial training is introduced. By generating adversarial samples during the training process, the model can still maintain high performance when facing perturbed data.

[0042]

[0043] in, For adversarial samples, is the disturbance intensity, is the symbolic function, is the gradient of the loss function with respect to the input data. By introducing adversarial samples during the training process, the model can cope with various interferences and noises in practical applications. In order to enhance the real-time and robustness of the fault identification module, edge computing technology is adopted to perform calculations and processing close to the data source, reducing the delay and bandwidth pressure of data transmission. The method of transfer learning is introduced in the fault identification module. By utilizing the model parameters pre-trained on a large-scale public data set, the effect of fault identification is improved. The pre-trained model parameters are transferred to the transmission component fault identification task and fine-tuned. The loss function of transfer learning for:

[0044]

[0045] in, is the weight parameter of the pre-trained model, is the weight parameter of the model, For data loss, is the regularization coefficient. By minimizing the loss function, the adaptation to the target task is achieved. Then, the prediction results are selected and the election output is :

[0046]

[0047] in, is the number of ensemble models, For the The prediction results of the model are is the indicator function.

[0048] Furthermore, the fault diagnosis module determines the specific cause and location of the fault, and proposes maintenance suggestions and preventive measures to establish a causal relationship model: ,in For a given fault characteristic Cause of failure in case of The posterior probability of For a given fault cause The observed characteristics The conditional probability of is the prior probability of the fault cause, is the marginal probability of the feature.

[0049] Furthermore, the system optimization module introduces Alluxio memory computing technology, which reduces disk I / O and improves data processing speed by caching frequently accessed data into memory. In terms of system operation, the system optimization module is reflected in the efficient management and scheduling of resources. Kubernetes containerization technology is used for resource management. By encapsulating system components into containers, dynamic allocation and scheduling of resources are achieved. The Horizontal PodAutoscaler elastic scaling technology is introduced in Kubernetes to dynamically adjust resource allocation according to the actual workload, ensuring that the system saves resources while running efficiently.

[0050] Beneficial effects of the present invention: Traditional fault diagnosis methods mostly rely on artificially set features and rules, while deep learning models can automatically learn high-level feature representations from massive data, thereby improving the accuracy and robustness of fault diagnosis. The robustness of the diagnostic model is improved by generating adversarial samples. The performance of traditional fault diagnosis models is prone to decline when facing data disturbances, while adversarial diagnosis uses adversarial training to enable the model to maintain efficient diagnostic capabilities under various disturbance conditions, optimize fault diagnosis strategies using reinforcement learning technology, and train intelligent agents to select optimal actions under given states by defining states, actions, and rewards. Model training is performed using a distributed deep learning framework, and training tasks are distributed to multiple computing nodes for parallel execution, significantly improving the efficiency of model training; dynamic resource allocation and scheduling are performed through container orchestration tools such as Kubernetes. This architecture not only improves the flexibility and scalability of the system, but also enhances the fault tolerance and reliability of the system. The present invention uses a distributed monitoring and log system to monitor the operating status of the system in real time, quickly discover and handle faults, and improve the fault tolerance of the system. At the same time, the microservice architecture enables each functional module of the system to operate independently, and when a module fails, it will not affect the operation of the overall system, thereby improving the reliability of the system. The present invention provides users with a friendly and intuitive operation interface, allowing users to quickly and easily understand the system operation status and fault diagnosis results. At the same time, the intelligent question-and-answer system can answer user questions in real time, provide personalized fault diagnosis and maintenance suggestions, and improve user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The invention is further illustrated by the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0052] Figure 1 It is a system schematic diagram of the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] See also Figure 1 The present invention will be further described in conjunction with the following examples.

[0055] See also Figure 1 The present invention aims to provide a power transmission component fault identification system based on deep learning, including a data acquisition module, a data preprocessing module, a feature extraction module, a model training module, a fault identification module, a fault diagnosis module and a system optimization module. The data acquisition module collects current, voltage, temperature and vibration data from sensors in the power transmission system; the data preprocessing module processes the collected raw data to improve the data quality and the accuracy of subsequent analysis, eliminates noise and outliers in the data through data cleaning, data normalization and missing value filling, and converts the data into a format suitable for deep learning model input; the feature extraction module uses a deep learning algorithm to extract effective features from the preprocessed data, and automatically learns and extracts key features that can characterize the fault according to the characteristics of different power transmission components and fault types. The performance of the feature extraction module directly affects the accuracy of fault identification; the model training module uses the preprocessed data and extracted features to train the deep learning model; the fault identification module identifies faults on the real-time data of the transmission system and monitors the operating status of the transmission components in real time. It can quickly and accurately identify faults and help operation and maintenance personnel find problems in time; the fault diagnosis module analyzes the specific cause and location of the fault, and locates and analyzes the cause of the fault based on the fault identification results and the structural information of the transmission system; the system optimization module continuously optimizes the entire fault identification system, regularly evaluates the system performance, identifies the weak links in the system, and proposes improvement plans. The system optimization module continuously improves the accuracy and stability of the fault identification system by increasing data samples, updating model algorithms, and adjusting system parameters.

[0056] Specifically, current, voltage, temperature, and vibration data are collected from current sensors, voltage sensors, temperature sensors, and vibration sensors in the transmission system. A distributed sensor network is used to effectively cover a wide range of transmission areas. Sensor nodes are deployed at nodes of transmission lines and components, and data is transmitted to the central processing unit through NB-IoT technology. Edge computing is used to reduce the computing burden of the central processing unit and improve the real-time performance of data processing. The data acquisition module uses multi-sensor fusion technology. Multi-sensor fusion can effectively reduce the measurement error and failure rate of a single sensor by comprehensively processing and calibrating data from different sensors. The data acquisition module uses the Hadoop big data platform to realize distributed storage and parallel processing of massive data, and combines the use of NoSQL databases and time series databases to achieve efficient storage and query of real-time and historical data.

[0057] Specifically, the data preprocessing module processes and optimizes the collected raw data, including data cleaning, normalization, and missing value filling:

[0058] (1) Data cleaning

[0059] Data cleaning is used to remove noise and outliers in the data. The Isolation Forest method is used to construct a random decision tree to isolate data points. Outliers are usually shallow in the tree because they are easy to be isolated. Assume that the data set is ,in, Representation dataset The data points are constructed by building multiple random trees and calculating the average path length of each data point in each tree to determine its abnormality. The expression is:

[0060]

[0061] in, Represents data points In the The path length in a tree, n is the number of trees, is the path length threshold, and data points with shorter path length are considered as outliers;

[0062] (2) Normalization

[0063] Data normalization converts data of different dimensions and ranges to a unified scale. By using an autoencoder, the low-dimensional representation of the data is learned by training a neural network, and the normalization process can be performed adaptively. The loss function of the autoencoder is for:

[0064]

[0065] in, , denote the encoder and decoder functions respectively, Representation dataset data points;

[0066] (3) Missing value filling

[0067] Generate data samples for missing value filling:

[0068]

[0069] in, denote the discriminator and the generator respectively, is the real data distribution, is the noise distribution, z is the data point in the noise distribution, Indicates that the independent variable x obeys The expected value under the distribution of Indicates that the data point z obeys The expected value under the distribution of Represents the optimization objective function, through adversarial training, the generator Ability to generate data samples to fill missing values;

[0070] After data cleaning, normalization and missing value filling, the dataset Updated to .

[0071] Specifically, the feature extraction module performs feature learning on the input preprocessed data. The features include edges, textures, shapes, time dependencies, trends, and periodicity. It combines supervised learning and extracts low-dimensional feature representations by reducing the dimensionality of the data and compressing the features. Through the encoding and decoding process, the dimension of the feature extraction module data is effectively reduced while retaining key feature information. The low-dimensional feature representation after dimensionality reduction not only reduces the computational complexity, but also improves the training efficiency of the model and the accuracy of fault diagnosis.

[0072] Specifically, the model training module builds the LSTM model, including:

[0073]

[0074]

[0075] in, is the hidden state of LSTM at time t, For LSTM The hidden state of the moment, For the updated dataset No. Features, is the output of the LSTM model, is the weight matrix, is the bias vector. During the training process, in order to avoid overfitting of the model, some neurons are randomly discarded during the training process so that the model can generalize better. Assume that the activation value of the neuron is , the generalized neuron activation value is:

[0076]

[0077] in, is the total number of neurons, Indicates the number, , To activate the probability, the learning rate is dynamically adjusted by calculating the first-order momentum and second-order momentum of the gradient:

[0078]

[0079] in, is the first-order momentum of the LSTM network gradient, is the second-order momentum of the LSTM network gradient, is the LSTM network gradient, are the LSTM network model parameters, is the evolution operator, is the first momentum coefficient, is the second momentum coefficient, is a constant approaching 0, that is .

[0080] Specifically, the fault identification module determines whether there is a fault in the input data. In order to improve the accuracy of fault detection, a fusion method of supervised learning and unsupervised learning is adopted. Supervised learning uses a trained deep learning model to predict the input data, and its output is the fault probability. , the fault detection criteria are:

[0081]

[0082] in, is the fault detection result, To detect the threshold, the reconstruction error is calculated to determine whether the data is abnormal. The reconstruction error is:

[0083]

[0084] in, For the reconstructed data, by setting the reconstruction error threshold ,when is judged as a fault.

[0085] Specifically, in terms of fault classification, the trained deep learning model can be used to identify the fault type. Assuming that the fault type is kind, is one of them, and the output of the model is Dimensional probability distribution , each dimension represents the probability of a fault type, and the fault classification criteria are:

[0086]

[0087] in, For the prediction results of the fault type, different weights are then assigned to different features to enhance the model's focus on key features, including:

[0088]

[0089] in, For emphasis weight, is the score function, Characterized by For query, For weighted feature representation, adversarial training is introduced. By generating adversarial samples during the training process, the model can still maintain high performance when facing perturbed data.

[0090]

[0091] in, For adversarial samples, is the disturbance intensity, is the symbolic function, is the gradient of the loss function with respect to the input data. By introducing adversarial samples during the training process, the model can cope with various interferences and noises in practical applications. In order to enhance the real-time and robustness of the fault identification module, edge computing technology is adopted to perform calculations and processing close to the data source, reducing the delay and bandwidth pressure of data transmission. The method of transfer learning is introduced in the fault identification module. By utilizing the model parameters pre-trained on a large-scale public data set, the effect of fault identification is improved. The pre-trained model parameters are transferred to the transmission component fault identification task and fine-tuned. The loss function of transfer learning for:

[0092]

[0093] in, is the weight parameter of the pre-trained model, is the weight parameter of the model, For data loss, is the regularization coefficient. By minimizing the loss function, the adaptation to the target task is achieved. Then, the prediction results are selected and the election output is :

[0094]

[0095] in, is the number of ensemble models, For the The prediction results of the model are is the indicator function.

[0096] Specifically, the fault diagnosis module determines the specific cause and location of the fault, and proposes maintenance suggestions and preventive measures to establish a causal relationship model: ,in For a given fault characteristic Cause of failure in case of The posterior probability of For a given fault cause The observed characteristics The conditional probability of is the prior probability of the fault cause, is the marginal probability of the feature.

[0097] Specifically, the system optimization module introduces Alluxio memory computing technology, which caches frequently accessed data into memory, reduces disk I / O, and improves data processing speed. In terms of system operation, the system optimization module is reflected in the efficient management and scheduling of resources. It uses Kubernetes containerization technology for resource management, and encapsulates system components into containers to achieve dynamic allocation and scheduling of resources. It introduces Horizontal Pod Autoscaler elastic scaling technology in Kubernetes, which dynamically adjusts resource allocation according to actual workloads to ensure that the system runs efficiently while saving resources.

[0098] In a specific example, the system first deploys multiple sensor nodes at the nodes of transmission lines and components through a distributed sensor network. The current, voltage, temperature and vibration data collected by the sensors are transmitted to the central processing unit, which uses edge computing technology to process part of the data in advance to reduce the load. After the data enters the data preprocessing module, the data is cleaned by the Isolation Forest method to remove noise and outliers, and then the data is normalized to convert data of different dimensions to a unified scale. The preprocessed data is transmitted to the feature extraction module, and the key features including edges, textures, shapes, etc. are extracted using a deep learning algorithm. After feature dimension reduction and compression processing, the feature data is input into the LSTM model for training. The model continuously adjusts parameters to adapt to different fault types. The trained model is deployed in the fault identification module for fault monitoring and identification of real-time data, combining supervised and unsupervised learning methods to improve the accuracy of identification. If a fault is detected, the fault diagnosis module will further analyze its cause and location and generate maintenance suggestions. The system optimization module will continuously collect more data and optimize algorithms, use Kubernetes for resource management, and ensure that the system can be updated and self-optimized in real time to improve the stability and accuracy of the overall system.

[0099] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0100] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0101] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A power transmission component fault identification system based on deep learning, characterized in that: It includes data acquisition module, data preprocessing module, feature extraction module, model training module, fault identification module, fault diagnosis module and system optimization module; the data acquisition module collects current, voltage, temperature and vibration data from sensors in the power transmission system; The data preprocessing module processes the collected raw data, eliminates noise and outliers in the data through data cleaning, data normalization, and missing value filling, and converts the data into a format suitable for deep learning model input; the feature extraction module uses deep learning algorithms to extract effective features from the preprocessed data, and automatically learns and extracts key features that can characterize faults based on the characteristics of different transmission components and fault types; The model training module uses the preprocessed data and extracted features to train the deep learning model; the fault identification module identifies faults on the real-time data of the transmission system and monitors the operating status of the transmission components in real time; the fault diagnosis module analyzes the specific cause and location of the fault, and locates and analyzes the cause of the fault based on the fault identification results and the structural information of the transmission system; the system optimization module continuously optimizes the entire fault identification system, regularly evaluates system performance, identifies weak links in the system, and proposes improvement plans.

2. A power transmission component fault identification system based on deep learning according to claim 1, characterized in that: Collect current, voltage, temperature, and vibration data from current sensors, voltage sensors, temperature sensors, and vibration sensors in the power transmission system. Use a distributed sensor network that can cover the power transmission area. Deploy sensor nodes at the nodes of the power transmission lines and components, and transmit data to the central processing unit through NB-IoT technology. Through edge computing, reduce the computing burden of the central processing unit. The data acquisition module adopts multi-sensor fusion technology. Multi-sensor fusion reduces the measurement error and failure rate of a single sensor by comprehensively processing and calibrating data from different sensors. The data acquisition module uses the Hadoop big data platform to realize distributed storage and parallel processing of massive data, and combines the use of NoSQL database and time series database to realize efficient storage and query of real-time data and historical data.

3. According to the deep learning-based power transmission component fault identification system of claim 1, it is characterized in that: The data preprocessing module processes and optimizes the collected raw data, including data cleaning, normalization, and missing value filling: Data cleaning is used to remove noise and outliers in the data. The Isolation Forest method is used to construct a random decision tree to isolate data points. Outliers are usually shallow in the tree because they are easy to be isolated. Assume that the data set is ,in, Representation dataset The data points are constructed by building multiple random trees and calculating the average path length of each data point in each tree to determine its abnormality. The expression is: in, Represents data points In the The path length in a tree, n is the number of trees, is the path length threshold, and data points with shorter path length are considered as outliers; Data normalization converts data of different dimensions and ranges to a unified scale. By using an autoencoder, the low-dimensional representation of the data is learned by training a neural network, and the normalization process can be performed adaptively. The loss function of the autoencoder is for: in, , denote the encoder and decoder functions respectively, Representation dataset data points; Generate data samples for missing value filling: in, denote the discriminator and the generator respectively, is the real data distribution, is the noise distribution, z is the data point in the noise distribution, Indicates that the independent variable x obeys The expected value under the distribution of Indicates that the data point z obeys The expected value under the distribution of Represents the optimization objective function, through adversarial training, the generator Ability to generate data samples to fill missing values; After data cleaning, normalization and missing value filling, the dataset Updated to .

4. A power transmission component fault identification system based on deep learning according to claim 1, characterized in that: The feature extraction module performs feature learning on the input preprocessed data, and the features include edges, textures, shapes, time dependencies, trends, and periodicity. It combines supervised learning and extracts low-dimensional feature representations by reducing the dimension of the data and compressing the features. Through the encoding and decoding process, the dimension of the feature extraction module data is reduced while retaining key feature information.

5. A power transmission component fault identification system based on deep learning according to claim 1, characterized in that: The model training module builds the LSTM model, which includes: in, is the hidden state of LSTM at time t, For LSTM The hidden state of the moment, For the updated dataset No. Features, is the output of the LSTM model, is the weight matrix, is the bias vector. During the training process, in order to avoid overfitting of the model, some neurons are randomly discarded during the training process so that the model can generalize better. Assume that the activation value of the neuron is , the generalized neuron activation value is: in, is the total number of neurons, Indicates the number, , To activate the probability, the learning rate is dynamically adjusted by calculating the first-order momentum and second-order momentum of the gradient: in, is the first-order momentum of the LSTM network gradient, is the second-order momentum of the LSTM network gradient, is the LSTM network gradient, are the LSTM network model parameters, is the evolution operator, is the first momentum coefficient, is the second momentum coefficient, is a constant approaching 0, that is .

6. A power transmission component fault identification system based on deep learning according to claim 1, characterized in that: The fault identification module determines whether there is a fault in the input data. In order to improve the accuracy of fault detection, a fusion method of supervised learning and unsupervised learning is adopted. Supervised learning uses a trained deep learning model to predict the input data, and its output is the fault probability. , the fault detection criteria are: in, is the fault detection result, To detect the threshold, the reconstruction error is calculated to determine whether the data is abnormal. The reconstruction error is: in, For the reconstructed data, by setting the reconstruction error threshold ,when is judged as a fault.

7. A power transmission component fault identification system based on deep learning according to claim 1, characterized in that: Use the trained deep learning model to identify the fault type. Assume that the fault type is kind, is one of them, and the output of the model is Dimensional probability distribution , each dimension represents the probability of a fault type, and the fault classification criteria are: in, For the prediction results of the fault type, different weights are then assigned to different features to enhance the model's focus on key features, including: in, is the score function, Characterized by For query, For weighted feature representation, adversarial training is introduced. By generating adversarial samples during the training process, the model can still maintain high performance when facing perturbed data. in, For adversarial samples, is the disturbance intensity, is the symbolic function, is the gradient of the loss function with respect to the input data. By introducing adversarial samples during the training process, the model can cope with various interferences and noises in practical applications. In order to enhance the real-time and robustness of the fault identification module, edge computing technology is used to perform calculations and processing close to the data source to reduce data transmission delays and bandwidth pressure. The method of transfer learning is introduced in the fault identification module. By using the model parameters pre-trained on a large-scale public data set, the effect of fault identification is improved. The pre-trained model parameters are transferred to the transmission component fault identification task and fine-tuned. The loss function of transfer learning for: in, is the weight parameter of the pre-trained model, is the weight parameter of the model, For data loss, is the regularization coefficient. By minimizing the loss function, the adaptation to the target task is achieved. Then, the prediction results are selected and the election output is : in, is the number of ensemble models, For the The prediction results of the model are is the indicator function.

8. The power transmission component fault identification system based on deep learning according to claim 1, characterized in that: The fault diagnosis module determines the specific cause and location of the fault, and proposes maintenance suggestions and preventive measures, and establishes a causal relationship model: ,in For a given fault characteristic Cause of failure in case of The posterior probability of For a given fault cause The observed characteristics The conditional probability of is the prior probability of the fault cause, is the marginal probability of the feature.

9. A power transmission component fault identification system based on deep learning according to claim 1, characterized in that: The system optimization module introduces Alluxio memory computing technology, which reduces disk I / O and improves data processing speed by caching frequently accessed data into memory. In terms of system operation, the system optimization module uses Kubernetes containerization technology for resource management, encapsulates system components into containers, realizes dynamic allocation and scheduling of resources, and introduces Horizontal Pod Autoscaler elastic scaling technology in Kubernetes to dynamically adjust resource allocation according to actual workload.

Images