An intelligent power communication fault diagnosis method and system

Abstract

The application provides a kind of intelligent power communication fault diagnosis method and system, method includes: deployment edge computing device, collection real-time data, pre-processing is carried out to real-time data, optimize the communication between nodes;Real-time feature extraction and data compression protocol are constructed;Select optimal path, compressed data is transmitted to central processor, adjust data transmission path, using cyclic redundancy check code adds check code to data to ensure that data transmission is error-free;Select key features, use interpretable machine learning model on central processor to process received key features, diagnose communication failure;Diffusion model is used to simulate fault diffusion process on central server, predict the impact of communication failure;Define network impact indicators and build decision model, design priority and automatically generate response suggestions.The application improves the efficiency and accuracy of fault handling, while enhancing the credibility of the system and the convenience of operation.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of electric power communication, and particularly relates to an intelligent electric power communication fault diagnosis method and system. BACKGROUND

[0002] In modern electric power systems, the reliability of communication technology is crucial for the stable operation of the entire power network. With the rapid development of smart grids, the complexity of electric power communication networks and the requirement for high reliability have also increased accordingly. Currently, electric power communication fault diagnosis technology mainly relies on traditional signal processing and pattern recognition methods, which can effectively diagnose common faults. However, with the increasing complexity of electric power communication network structure, traditional methods face many challenges in processing large-scale data and complex network structures, especially in terms of fault prediction accuracy and processing speed.

[0003] Current technologies usually use centralized processing systems, which have obvious limitations in data processing and fault location. First, centralized systems are prone to data transmission delays when processing large amounts of data from various nodes in the network, which is unacceptable in electric power communication fault processing that requires fast response. Second, due to the lack of effective data processing mechanisms, existing technologies often struggle to accurately identify fault types and locations when facing complex faults, especially when the fault affects multiple network nodes. In addition, existing fault diagnosis systems lack sufficient transparency, making it difficult for operators to understand the system's diagnostic logic, which reduces the system's credibility and operational convenience.

[0004] These problems mainly stem from the fact that existing technologies have failed to fully utilize the latest artificial intelligence technologies and advanced computing architectures. Although artificial intelligence has shown strong processing capabilities in other fields such as image recognition and natural language processing, its application in electric power communication fault diagnosis is not widespread enough. In addition, edge computing, as an emerging computing model, has not been fully utilized in terms of local data processing and reducing latency.

[0005] Therefore, current electric power communication fault diagnosis technology still has room for improvement, especially in terms of improving processing speed, accuracy, and system transparency. SUMMARY

[0006] The purpose of the present application is to propose an intelligent electric power communication fault diagnosis method and system that integrates self-explainable machine learning, edge computing, and Diffusion models to improve fault processing efficiency and accuracy while enhancing user credibility and operational convenience of the system.

[0007] To achieve the above purpose, in the first aspect of the present application, an intelligent electric power communication fault diagnosis method is provided, which comprises:

[0008] S1, deploying edge computing devices on nodes of the power communication network, collecting real-time data, designing preprocessing strategies to preprocess the real-time data, and defining a communication protocol between nodes to optimize communication between nodes;

[0009] S2, constructing a real-time feature extraction and data compression protocol to extract and compress the data of the preprocessed edge computing device;

[0010] S3, selecting an optimal path to transmit the compressed data to the central processor, dynamically adjusting the data transmission path according to the real-time traffic of the network, and adding a check code to the data using a cyclic redundancy check code to ensure error-free data transmission; wherein the optimal path is calculated as follows:

[0011] Let denote the distance from node to , and the path cost function is:

[0012]

[0013] wherein, denotes the path from node to node ;

[0014] S4, selecting key features, using an interpretable machine learning model on the central processor to process the received key features to diagnose communication failures;

[0015] S5, using the Diffusion model to simulate the fault diffusion process on the central server to predict the impact of communication failures;

[0016] S6, defining network impact indicators and building a decision model, designing priorities and automatically generating response suggestions based on the diagnosis results of communication failures and the impact of communication failures;

[0017] wherein, the S4 specifically includes:

[0018] Select key features, and set at least one hundred decision trees, each tree uses a randomly selected feature subset during training, and then uses information gain to evaluate the contribution of each feature to fault classification, and selects the features that most help improve classification accuracy; wherein the key features include peak value, mean value, standard deviation and energy spectrum;

[0019] For a given prediction result, select sample points around it, use a simple linear model to approximate the behavior of a complex linear model in the current local area, calculate the partial derivative of the output with respect to each input feature for the approximated linear model, and evaluate the influence of the change of each feature on the prediction result, which is calculated as follows:

[0020]

[0021] wherein, is the coefficient of feature in the linear model, representing the influence weight of the current feature in the local model; a feedback score mechanism is introduced to calculate the overall satisfaction, and an adaptive adjustment strategy is used to find the optimal model parameters, and a dynamic data reweighting method is introduced to give the highest weight to the newly collected data, and the old data is weighted according to the time effectiveness, which is represented as follows:

[0022]

[0023] wherein, represents the decay coefficient, represents the current time, represents the time of data collection.

[0024] Further, the method further comprises: S7, collecting operation data and performance data, optimizing system performance and fault response strategy through machine learning technology; S8, developing and optimizing user interface, providing monitoring and operation interface.

[0025] Further, deploying edge computing devices on the nodes of the power communication network selects the location of the edge computing nodes by adopting a graph theory-based optimization method, and models the power communication network as a graph wherein is the set of nodes in the network, is the set of edges connecting the nodes, and uses a genetic algorithm to find the optimal node location.

[0026] Further, the preprocessing includes reducing noise and extracting signals using filtering and feature extraction algorithms, wherein a convolution filter is used for data smoothing, and the filter weight is calculated by the following formula:

[0027]

[0028] wherein, represents the current time step, represents the filter center, represents the standard deviation, represents the filter length;

[0029] The optimization of communication between nodes specifically includes:

[0030] The minimum spanning tree algorithm is used to optimize the communication path to reduce the overall communication cost, which is represented as follows:

[0031]

[0032] in, This represents the total weight of the spanning tree. This represents the selected set of edges. Representing an edge The weight.

[0033] Further, step S2 specifically includes:

[0034] The dynamic range of real-time data is calculated by measuring short-time energy and variance, and an adaptive threshold is set.

[0035] In each time window Within this process, efficient feature extraction techniques are applied to calculate the energy spectral density for each decomposed subband. and information entropy ;

[0036] If the feature value exceeds the adaptive threshold, the feature is encoded and transmitted using a quantization method; the level of the quantization method is dynamically adjusted according to the statistical importance of the feature.

[0037] The system monitors the processed data stream in real time, uses anomaly detection algorithms to identify potential anomaly patterns, and immediately generates feedback signals and adjusts data processing parameters when an anomaly is detected.

[0038] Furthermore, the dynamic adjustment of the data transmission path based on the real-time network traffic is determined by load balancing calculation, which is as follows:

[0039]

[0040] in, This indicates the amount of new load. This indicates the preset load threshold.

[0041] Huffman coding is used to compress the feature data. During the encoding process, each feature data... It is mapped to a binary string, the length of which is dynamically adjusted according to the frequency of the data occurrence, with higher frequency features assigned shorter codes.

[0042] Further, step S5 specifically includes:

[0043] Power communication network modeling as a graph ,in, Represents a set of network nodes. Denotes the set of edges. Represent the set of edge weights and initialize the parameters;

[0044] Based on network operation status and external influences, an initial fault is injected into a specific node of the network, and then the fault propagation probability is calculated as follows:

[0045]

[0046] in, Indicates time Faults can occur at any time from the node. propagation to nodes The probability of; This represents the Sigmoid function, ensuring that the probability value is between 0 and 1; and These are parameters used to adjust the model's sensitivity. Represents a node to Distance or communication delay;

[0047] Calculate the cumulative impact score for each node. This reflects the impact of a fault on the node's service capabilities, and is represented as follows:

[0048]

[0049] in, Represents a node The set of upstream nodes;

[0050] Adjusted based on real-time monitoring data and Based on the parameters and the results of the fault impact analysis, formulate corresponding response measures.

[0051] Furthermore, the network influence metric is expressed as follows:

[0052]

[0053] in, and These represent the load on nodes and edges, respectively. and This represents the degree of influence of nodes and edges in fault propagation. and This represents a coefficient that adjusts the importance of the two factors.

[0054] The decision model is represented as follows:

[0055]

[0056] in, and Representing nodes respectively and edge Whether to take response measures, with a value of 0 or 1.

[0057] Furthermore, the operational data includes occurrence frequency, response time, and fault handling efficiency. The performance data uses the collected data to evaluate the performance of each part of the system, and then adjusts the system parameters and operating strategies based on the performance evaluation results.

[0058] A second aspect of the present invention provides an intelligent power communication fault diagnosis system, the system comprising:

[0059] The edge computing deployment module is used to deploy edge computing devices on nodes of the power communication network, collect real-time data, design preprocessing strategies to preprocess the real-time data, and define communication protocols between nodes to optimize communication between nodes.

[0060] The feature extraction and compression module is used to build a real-time feature extraction and data compression protocol to extract and compress features from the pre-processed data of the edge computing device.

[0061] The optimal path calculation module is used to select the optimal path to transmit the compressed data to the central processor. Simultaneously, it dynamically adjusts the data transmission path based on real-time network traffic conditions and uses a cyclic redundancy check (CRC) code to add a checksum to the data to ensure error-free data transmission. The optimal path calculation is as follows:

[0062] set up Represents a node arrive The distance and path cost function are:

[0063]

[0064] in, Represents a node To the node The path;

[0065] The key feature selection module is used to select key features, process the received key features on the central processor using an interpretable machine learning model, and diagnose communication failures.

[0066] The fault propagation calculation module is used to simulate the fault propagation process on the central server using the Diffusion model to predict the impact of communication failures.

[0067] The response module is used to define network impact indicators and build decision models. It designs priorities and automatically generates response suggestions based on the diagnosis results of communication failures and the impact caused by communication failures.

[0068] The key feature selection module specifically performs the following steps:

[0069] Key features are selected, and at least one hundred decision trees are set. Each tree is trained using a randomly selected subset of features. Information gain is then used to evaluate the contribution of each feature to fault classification, and the feature that is most helpful in improving classification accuracy is selected. The key features include peak value, mean, standard deviation, and energy spectrum.

[0070] For a given prediction result, select sample points in its vicinity, and use a simple linear model to approximate the behavior of the complex linear model in the current local context. For the approximated linear model, calculate the partial derivative of the output with respect to each input feature, and evaluate the impact of changes in each feature on the prediction result. The calculation is as follows:

[0071]

[0072] in, Features in a linear model The coefficient represents the influence weight of the current feature in the local model; a feedback scoring mechanism is introduced to calculate the overall satisfaction, and an adaptive adjustment strategy is used to find the optimal model parameters. Then, a dynamic data reweighting method is introduced to assign the highest weight to newly collected data, while the weight of old data decreases according to its timeliness, as shown below:

[0073]

[0074] in, Indicates the attenuation coefficient. Indicates the current time. w represents the time of data collection. i These are the weights in the local model.

[0075] The beneficial technical effects of the present invention are at least as follows:

[0076] (1) By deploying edge computing devices at key nodes of the power communication network, this invention enables local data processing, significantly reducing the time required for data transmission and thus improving the response speed of fault diagnosis. This directly addresses the shortcomings of traditional centralized processing systems in terms of data processing speed and real-time performance. The edge computing capability of this invention further refines and compresses data, ensuring that only the most critical information is transmitted to the central server. This not only improves the efficiency of data transmission but also reduces the complexity and time of subsequent processing, effectively enhancing the overall response speed and accuracy of the system. The refined data extracted can reach the central processing unit safely and efficiently. This optimization of the transmission process reduces the potential loss and tampering of data during transmission, ensuring the accuracy and security of data analysis.

[0077] (2) This invention employs advanced interpretable machine learning algorithms (such as LIME or SHAP) to provide a transparent decision-making process for fault diagnosis. This enables operators not only to know where the fault occurred but also to understand the cause of the fault, effectively improving the system's credibility and user acceptance. It improves diagnostic accuracy and provides decision transparency through interpretable machine learning, allowing operators to understand the cause of the fault and enhancing the system's credibility and user satisfaction.

[0078] (3) The invention employs a diffusion model to predict possible fault propagation paths, providing a scientific basis for the formulation of fault response measures. This method can predict the potential impact of a fault in its early stages, allowing for proactive measures to effectively mitigate its effects. This innovation addresses the shortcomings of existing technologies in handling complex network faults. The system algorithm and performance are optimized based on actual operational feedback to ensure the system can adapt to changes in the network environment and the evolution of fault modes. Attached Figure Description

[0079] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0080] Figure 1 This is a flowchart of an intelligent power communication fault diagnosis method according to the present invention;

[0081] Figure 2 This is a framework diagram of an intelligent power communication fault diagnosis system according to the present invention. Detailed Implementation

[0082] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0083] like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for diagnosing faults in intelligent power communication, the method comprising the following steps S1-S8:

[0084] S1. Deploy edge computing devices on nodes of the power communication network to collect real-time data, design preprocessing strategies to preprocess the real-time data, and define communication protocols between nodes to optimize communication between nodes.

[0085] In this embodiment, to maximize coverage and efficiency, a graph-based optimization method is used to select the locations of edge computing nodes. First, the power communication network is modeled as a graph. ,in It is a set of nodes in the network. This is the set of edges connecting nodes. A genetic algorithm is used to find the optimal node positions, with the goal of minimizing the overall network's data transmission latency and processing time. The fitness function is defined as:

[0086]

[0087] in, It is the selected set of edge nodes. Represents a node and The shortest path distance between them Represents a node Data processing time and It is a weighting coefficient that adjusts the importance of latency and processing time.

[0088] Furthermore, the hardware configuration is selected based on the predetermined location and expected load. The hardware configuration should support high-concurrency data processing, for example, using a multi-core CPU with fast data processing capabilities, a high-speed cache, and an SSD with fast read and write capabilities. Initialization includes installing a real-time operating system (RTOS) and real-time data processing software specifically designed for power communication data.

[0089] In this embodiment, preprocessing includes designing a data preprocessing strategy, including data buffering, filtering, and feature extraction. Considering the temporal and periodic characteristics of power data, a filtering and feature extraction algorithm is designed to reduce noise and extract useful signals. A convolutional filter is used for data smoothing, and the filter weights are calculated using the following formula:

[0090]

[0091] in, It is the current time step. It is the center of the filter. It is the standard deviation. It is the filter length.

[0092] Simultaneously, a communication protocol between edge nodes is defined to ensure efficient data transmission between nodes. The Minimum Spanning Tree (MST) algorithm is used to optimize communication paths to reduce overall communication costs. Algorithm implementation:

[0093]

[0094] in, It is the total weight of the spanning tree. It is the selected edge set. Representing an edge The weight.

[0095] Through the above steps, not only is the effective deployment and data processing capability of edge computing nodes ensured, but also the data processing efficiency and response speed of the entire power communication network are improved through mathematical model and algorithm optimization.

[0096] S2. Construct a real-time feature extraction and data compression protocol to extract and compress features from the preprocessed data of the edge computing device.

[0097] In this embodiment, for signals in a power communication network, the dynamic range of the data is first determined by calculating short-time energy and variance. An adaptive threshold is then set. in, It's recently The mean of a sample, It is the standard deviation. This is an adjustment coefficient used to adjust the threshold sensitivity based on network conditions. Also, within each time window... Within this framework, efficient feature extraction techniques, such as Discrete Wavelet Transform (DWT), are applied to decompose the signal.

[0098] For each subband of the decomposition, calculate the energy spectral density. and information entropy :

[0099]

[0100] in, It is the first Sub-band One coefficient, It is its probability density function, and N is the total number of systems.

[0101] Furthermore, a threshold-based compression strategy is adopted, where compression only occurs when the feature value exceeds a set threshold. Only then is the feature encoded and transmitted. A custom quantization method is then used to encode the feature, with the quantization level dynamically adjusted based on the feature's statistical importance, optimizing storage and transmission efficiency.

[0102] In a preferred embodiment of the invention, the processed data stream is monitored in real time, and an anomaly detection algorithm (a machine learning-based classifier) ​​is used to identify potential anomaly patterns. When an anomaly is detected, a feedback signal is immediately generated and data processing parameters, such as thresholds, are adjusted. and window size .

[0103] S3. Select the optimal path to transmit the compressed data to the central processor. At the same time, dynamically adjust the data transmission path according to the real-time network traffic. Use cyclic redundancy check codes to add check codes to the data to ensure error-free data transmission.

[0104] It should be noted that Dijkstra's algorithm is used to determine the shortest path from each edge node to the central processing unit. Nodes and edges in the network are defined as representing computational nodes and the communication links connecting them, respectively. Let... Represents a node arrive The distance and path cost function are:

[0105]

[0106] in It is a node To the node The path is determined by selecting the path with the lowest cost for data transmission.

[0107] Furthermore, the data transmission path is dynamically adjusted based on real-time network traffic. The traffic load of each communication link is monitored in real time. And switch to a less loaded alternative path if necessary. Load balancing Use the following formula to calculate:

[0108]

[0109] in, This is the newly added load. It is a preset load threshold. This indicates the load balance between node i and node j at the previous moment.

[0110] Before data transmission, use effective data compression algorithms to reduce data size and lower network load. Huffman coding is employed to compress feature data, with each feature data segment during the encoding process... It is mapped to a binary string, the length of which is dynamically adjusted according to the frequency of data occurrence; features with higher frequency are assigned shorter codes.

[0111]

[0112] in, It is a feature The frequency of occurrence, yes The allocated encoding length, Representation of features The encoding length is X, where X is the total number of features.

[0113] It should be noted that, to ensure data integrity during transmission, a Cyclic Redundancy Check (CRC) algorithm is used to add a checksum to the data packets. CRC uses a polynomial... The sender generates a checksum and treats the data as a large polynomial. And calculate the remainder. :

[0114]

[0115] in This is the length of the checksum. The receiving end verifies the data through the same process to ensure error-free data transmission. Modulo operation, in mathematics, refers to the remainder obtained after dividing two integers. In computer science, modulo operation is usually represented by "%", for example, 10 % 3 = 1, meaning that the remainder when 10 is divided by 3 is 1. Modulo operation is commonly used in operations such as calculating loops, determining parity, and generating random numbers.

[0116] Optimal path selection ensures data transmission speed and efficiency, dynamic load balancing adjusts transmission paths based on real-time network conditions to avoid congestion, data compression and encoding technologies significantly reduce the amount of data that needs to be transmitted, and data integrity and error correction ensure data accuracy and reliability.

[0117] S4. Select key features and use an interpretable machine learning model on the central processor to process the received key features and diagnose communication failures.

[0118] In this embodiment, feature selection is a crucial step in the model training process, directly impacting model performance and output interpretability. In power communication fault diagnosis, selecting the correct features is essential for improving the accuracy of fault identification. Therefore, the selected features include:

[0119] Peak Value: The highest level of the electrical signal during a fault, which can reflect the abnormal current or voltage at the moment of the fault.

[0120] Mean Value: The average level of a signal, which plays an important role in analyzing the normal performance of a power system.

[0121] Standard deviation: A statistical indicator that measures the fluctuation of signal level. It usually has large fluctuations under fault conditions.

[0122] Energy Spectrum: The frequency domain characteristics of a signal obtained through Fast Fourier Transform (FFT). Faults often exhibit abnormal energy at specific frequencies.

[0123] It should be noted that the training process of the random forest algorithm includes the following:

[0124] A. Initialize the random forest: Set up hundreds of decision trees, each of which uses a randomly selected subset of features during training to increase the diversity of the model.

[0125] B. Feature Importance Assessment (Information Gain): Information gain is used to evaluate the contribution of each feature to fault classification and select the feature that is most helpful in improving classification accuracy.

[0126] C. Model Training and Validation: Use cross-validation to evaluate the model's performance and ensure high accuracy across different subsets of data.

[0127] As a preferred embodiment of the present invention, Locally Interpretable Model-Sensitivity Analysis (LIME-SA) is employed to enhance the interpretability of the model. This method not only interprets the model's prediction results but also analyzes how small changes in input features affect the output, providing a deeper understanding of the model's behavior.

[0128] The following section further explains the Locally Interpretable Model-Sensitivity Analysis (LIME-SA):

[0129] For a given prediction result, sample points in the surrounding area are selected, and a simple linear model is used to approximate the behavior of the complex model in that local area. For the approximated linear model, the partial derivative of the output with respect to each input feature is calculated, and the impact of small changes in each feature on the prediction result is evaluated.

[0130]

[0131] in Features in a linear model The coefficient represents the influence weight of the feature in the local model.

[0132] Visualizing the contribution and sensitivity metrics of each feature helps operators understand why the model makes specific predictions and which features are key factors in decision-making.

[0133] Furthermore, a feedback rating system is introduced during model iteration, allowing operators to rate the accuracy and interpretability of the model's predictions. The rating range is set from 1 to 5, where 5 represents very satisfactory and 1 represents unsatisfactory. The overall user satisfaction with the model can be calculated using the following formula:

[0134]

[0135] in It is overall satisfaction. It is the first The feedback rating, This is the number of feedback received.

[0136] The model tuning employs adaptive tuning strategies, such as grid search or Bayesian optimization, to find the optimal model parameters. For random forests, key parameters include the number of trees. and the maximum depth of the tree The optimization objective is to maximize the accuracy of cross-validation while maintaining high user satisfaction. The optimization problem can be expressed as:

[0137]

[0138] in, It is an optimization function. It is about adjusting accuracy. Importance of user satisfaction The weight parameters.

[0139] Furthermore, to better adapt the model to new fault types and system changes, a dynamic data weighting technique is employed. Newly collected data is assigned higher weights, while older data has decreasing weights based on its timeliness. Data Weights Based on the timeliness of the data Calculated using the following formula:

[0140]

[0141] in, It is the attenuation coefficient. It is the current time. It refers to the time of data collection, w i These are the weights in the local model. This way, the latest data will have a greater impact during training.

[0142] As a preferred embodiment of the present invention, a periodic performance evaluation mechanism for the model is established, with a comprehensive performance evaluation of the model performed every evaluation period (e.g., quarterly). The evaluation includes accuracy, feedback satisfaction, and the ability to identify new fault types. Based on the evaluation results, a decision is made on whether to adjust parameters or completely retrain the model.

[0143] S5. Using a Diffusion model, simulate the fault propagation process on the central server to predict the impact of communication failures, including:

[0144] Power communication network modeling as a graph ,in, Represents a set of network nodes. Denotes the set of edges. Represent the set of edge weights and initialize the parameters;

[0145] Based on network operation status and external influences, an initial fault is injected into a specific node of the network, and then the fault propagation probability is calculated as follows:

[0146]

[0147] in, Indicates time Faults can occur at any time from the node. propagation to nodes The probability of; This represents the Sigmoid function, ensuring that the probability value is between 0 and 1; and These are parameters used to adjust the model's sensitivity. Represents a node to Distance or communication delay, Represents a node To the node The weights;

[0148] Calculate the cumulative impact score for each node. This reflects the impact of a fault on the node's service capabilities, and is represented as follows:

[0149]

[0150] in, Represents a node The set of upstream nodes, This represents the probability that a fault propagates from node u to node v at time t.

[0151] Adjusted based on real-time monitoring data and The parameters are analyzed based on the results of the fault impact analysis to more accurately reflect the current network status and environmental changes. At the same time, corresponding response measures are formulated based on the results of the fault impact analysis, such as route adjustment, load balancing, and early warning issuance.

[0152] S6. Define network impact indicators and build a decision model. Based on the diagnosis results of communication failures and the impact caused by communication failures, design priorities and automatically generate response suggestions.

[0153] In this embodiment, an impact metric is defined for each node and connection. and This represents the importance of nodes and edges in the network, and can be calculated based on factors such as traffic load, number of connections, and the impact of fault propagation. The specific formula is:

[0154]

[0155] in, and These are the loads of the nodes and edges, respectively. and It represents the degree of influence of nodes and edges in fault propagation. and It is a coefficient that adjusts the importance of the two. and These represent the impact indices of node v and edge e, respectively.

[0156] Specifically, a decision-making model is established to determine the optimal response strategy. This model is based on minimizing the risk of overall network outages and service degradation, with the following specific strategies:

[0157]

[0158] in, and Representing nodes respectively and edge Whether to take response measures (such as isolation, rerouting, etc.), with a value of 0 or 1.

[0159] In a preferred embodiment of the present invention, priority allocation and resource allocation will be performed.

[0160] Specifically, based on the output of the decision model, resources in the network are prioritized and allocated. High-priority nodes or connections will receive resource support first, such as backup power and alternative communication paths. Resource allocation also considers the actual available resources and the urgency of fault recovery, ensuring that critical services and nodes are restored first.

[0161] Furthermore, based on model recommendations, specific response strategies are executed automatically or manually, including but not limited to isolating affected nodes, switching to alternative communication paths, and increasing network bandwidth. The effectiveness of strategy execution is monitored in real time, and adjustments are made as necessary to adapt to dynamic changes in network conditions.

[0162] S7. Collect operational and performance data, and optimize system performance and fault response strategies through machine learning techniques.

[0163] Specifically, operational data is collected periodically from various nodes and links, including failure frequency, response time, and failure handling efficiency. The data collection formula can be expressed as:

[0164]

[0165] in, It is in time The collected dataset, It is the first Each monitoring point at time The data.

[0166] The performance of each component of the system is evaluated using the collected data. The main focus is on the accuracy of fault detection, the timeliness of fault response, and the stability of the system. Performance evaluation can be performed using the following formula:

[0167]

[0168] in, It is in time Overall performance score It is the first The accuracy of each node, It is response time. and It is a weighting coefficient, which is adjusted according to the system priority, and n is the total number of nodes.

[0169] In a preferred embodiment of the present invention, system parameters and operating strategies are adjusted based on performance evaluation results. This may include adjusting the parameters of the fault detection algorithm, optimizing network routing strategies, and adjusting resource allocation. Parameter adjustments can be made based on an optimization model.

[0170]

[0171] in, and These are the parameters before and after adjustment, respectively. It's the learning rate. It is a loss function based on performance scoring. It is the loss function with respect to the parameters The gradient.

[0172] Simultaneously, adjust parameters and strategies, and deploy updates within the system. Continuously monitor system performance after the update to ensure all changes deliver the expected improvements. Maintain service continuity and data integrity throughout the system update process.

[0173] S8: Develop and optimize the user interface to provide monitoring and operation interfaces.

[0174] In a preferred embodiment of the invention, a clear interface layout is designed, placing important information and frequently used functions in prominent positions. Modular design principles are used to differentiate between information display areas, operation areas, and feedback areas. An information hierarchy structure is applied, prioritizing the display of key data, such as real-time fault status and system warnings. Secondary information, such as historical data statistics and analysis reports, can be accessed through secondary menus or tabs.

[0175] Simultaneously, develop dynamic data visualization components to display key performance indicators and system status in real time. Use charts and dashboards to display data, such as network status graphs and time-series graphs of failure frequency. Select appropriate visualization methods for each data type.

[0176] Network traffic and bandwidth are represented using line graphs or area graphs.

[0177] Fault response time and processing efficiency are displayed using bar charts or pie charts.

[0178] In addition, interactive features are implemented, such as allowing users to view detailed data or historical information by clicking on an element in a chart. An interactive fault diagnosis tool is introduced, allowing users to input fault symptoms, and the system provides possible causes of the fault and suggested response measures based on the input.

[0179] Finally, a rapid feedback mechanism was designed and implemented to provide immediate visual feedback when users perform actions such as adjusting settings or taking fault response measures. Color coding, icons, and notification messages are used to indicate the results of the actions, such as success, warnings, or error messages.

[0180] By implementing the aforementioned user interface and interaction optimization schemes, the power communication system will provide a more efficient, intuitive, and user-friendly operating experience. This not only improves operator efficiency but also helps to identify and respond to system problems more quickly, ultimately enhancing the reliability and stability of the entire power communication network. Furthermore, the optimized interface and interaction design ensures that technicians and managers can quickly understand system information and make more accurate and timely decisions.

[0181] In addition, such as Figure 2 As shown in the figure, the system of this invention, an intelligent power communication fault diagnosis system, the system includes:

[0182] The edge computing deployment module 101 is used to deploy edge computing devices on nodes of the power communication network, collect real-time data, design preprocessing strategies to preprocess the real-time data, and define communication protocols between nodes to optimize communication between nodes.

[0183] The feature extraction and compression module 102 is used to construct a real-time feature extraction and data compression protocol to extract and compress features from the preprocessed data of the edge computing device;

[0184] The optimal path calculation module 103 is used to select the optimal path to transmit the compressed data to the central processor. Simultaneously, it dynamically adjusts the data transmission path based on real-time network traffic conditions and uses a cyclic redundancy check (CRC) code to add a checksum to the data to ensure error-free data transmission. The optimal path calculation is as follows:

[0185] set up Represents a node arrive The distance and path cost function are:

[0186]

[0187] in, Represents a node To the node The path;

[0188] The key feature selection module 104 is used to select key features, process the received key features on the central processor using an interpretable machine learning model, and diagnose communication failures.

[0189] The fault propagation calculation module 105 is used to simulate the fault propagation process on the central server using the Diffusion model to predict the impact of communication failures.

[0190] The response module 106 is used to define network impact indicators and build a decision model. Based on the diagnosis results of communication failures and the impact caused by communication failures, it designs priorities and automatically generates response suggestions.

[0191] The steps performed by the key feature selection module 104 specifically include:

[0192] Key features are selected, and at least one hundred decision trees are set. Each tree is trained using a randomly selected subset of features. Information gain is then used to evaluate the contribution of each feature to fault classification, and the feature that is most helpful in improving classification accuracy is selected. The key features include peak value, mean, standard deviation, and energy spectrum.

[0193] For a given prediction result, select sample points in its vicinity, and use a simple linear model to approximate the behavior of the complex linear model in the current local context. For the approximated linear model, calculate the partial derivative of the output with respect to each input feature, and evaluate the impact of changes in each feature on the prediction result. The calculation is as follows:

[0194]

[0195] in, Represents the prediction function. Features in a linear model The coefficient represents the influence weight of the current feature in the local model; a feedback scoring mechanism is introduced to calculate the overall satisfaction, and an adaptive adjustment strategy is used to find the optimal model parameters. Then, a dynamic data reweighting method is introduced to assign the highest weight to newly collected data, while the weight of old data decreases according to its timeliness, as shown below:

[0196]

[0197] in, Indicates the attenuation coefficient. Indicates the current time. Indicates the time of data collection.

[0198] It should be understood that the method steps in the embodiments of the present invention can be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium. The method can use standard programming techniques. Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if necessary, the program can be implemented in assembly or machine language. In any case, the language can be a compiled or interpreted language. Furthermore, for this purpose, the program can run on a programmable application-specific integrated circuit (ASIC).

[0199] Furthermore, the procedures described herein may be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by the context. The procedures described herein (or variations and / or combinations thereof) may be executed under the control of one or more computer systems configured with executable instructions, and may be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. The computer program comprises a plurality of instructions executable by one or more processors.

Claims

1. A method for diagnosing faults in intelligent power communication, characterized in that, The method includes: S1. Deploy edge computing devices on nodes of the power communication network to collect real-time data, design preprocessing strategies to preprocess the real-time data, and define communication protocols between nodes to optimize communication between nodes. S2. Construct a real-time feature extraction and data compression protocol to extract and compress features from the preprocessed data of the edge computing device; S3. Select the optimal path to transmit the compressed data to the central processor. Simultaneously, dynamically adjust the data transmission path based on real-time network traffic. Use a cyclic redundancy check (CRC) code to add a checksum to the data to ensure error-free data transmission. The optimal path is calculated as follows: set up Represents a node arrive The distance and path cost function are: ; in, Represents a node To the node The path; S4. Select key features and use an interpretable machine learning model on the central processor to process the received key features and diagnose communication failures. S5. Use the Diffusion model to simulate the fault propagation process on the central server and predict the impact of communication failures. S6. Define network impact indicators and build a decision model. Based on the diagnosis results of communication failures and the impact caused by communication failures, design priorities and automatically generate response suggestions. Specifically, S4 includes: Key features are selected, and at least one hundred decision trees are set. Each tree is trained using a randomly selected subset of features. Information gain is then used to evaluate the contribution of each feature to fault classification, and the feature that is most helpful in improving classification accuracy is selected. The key features include peak value, mean, standard deviation, and energy spectrum. For a given prediction result, select sample points in its vicinity, and use a simple linear model to approximate the behavior of the complex linear model in the current local context. For the approximated linear model, calculate the partial derivative of the output with respect to each input feature, and evaluate the impact of changes in each feature on the prediction result. The calculation is as follows: ; in, Features in a linear model The coefficient represents the influence weight of the current feature in the local model; This represents the prediction function; a feedback scoring mechanism is introduced to calculate overall satisfaction, and an adaptive adjustment strategy is used to find the optimal model parameters. Then, a dynamic data reweighting method is introduced, assigning the highest weight to newly collected data and decreasing the weight of older data based on their timeliness, as shown below: ; in, Indicates the attenuation coefficient. Indicates the current time. w represents the time of data collection. i These are the weights in the local model.

2. The intelligent power communication fault diagnosis method according to claim 1, characterized in that, The method further includes: S7, collecting operational and performance data, and optimizing system performance and fault response strategies through machine learning techniques; S8, developing and optimizing the user interface to provide a monitoring and operation interface.

3. The intelligent power communication fault diagnosis method according to claim 1, characterized in that, Deploying edge computing devices at nodes in a power communication network involves using graph theory-based optimization methods to select the locations of these nodes, while simultaneously modeling the power communication network as a graph. ,in It is a set of nodes in the network. It is the set of edges connecting nodes, and the optimal node position is found using a genetic algorithm.

4. The intelligent power communication fault diagnosis method according to claim 1, characterized in that, The preprocessing includes using filtering and feature extraction algorithms to reduce noise and extract the signal, wherein a convolutional filter is used for data smoothing, and the filter weights... Calculated using the following formula: ; in, Indicates the current time step. Indicates the center of the filter. Indicates standard deviation, Indicates the filter length; The optimized communication between nodes specifically includes: The minimum spanning tree algorithm is used to optimize communication paths and reduce overall communication costs, as shown below: ; in, This represents the total weight of the spanning tree. This represents the selected set of edges. Representing an edge The weight.

5. The intelligent power communication fault diagnosis method according to claim 1, characterized in that, Step S2 specifically includes: The dynamic range of real-time data is calculated by measuring short-time energy and variance, and an adaptive threshold is set. In each time window Within this process, efficient feature extraction techniques are applied to calculate the energy spectral density for each decomposed subband. and information entropy ; If the feature value exceeds the adaptive threshold, the feature is encoded and transmitted using a quantization method; the level of the quantization method is dynamically adjusted according to the statistical importance of the feature. The system monitors the processed data stream in real time, uses anomaly detection algorithms to identify potential anomaly patterns, and immediately generates feedback signals and adjusts data processing parameters when an anomaly is detected.

6. The intelligent power communication fault diagnosis method according to claim 1, characterized in that, The dynamic adjustment of data transmission paths based on real-time network traffic is determined by load balancing calculations. The calculation is as follows: ; in, This indicates the amount of new load. This indicates the preset load threshold. This indicates the load balance between node i and node j at the previous moment; Huffman coding is used to compress the feature data. During the encoding process, each feature data... It is mapped to a binary string, the length of which is dynamically adjusted according to the frequency of the data occurrence, with higher frequency features assigned shorter codes.

7. The intelligent power communication fault diagnosis method according to claim 1, characterized in that, Step S5 specifically includes: Power communication network modeling as a graph ,in, Represents a set of network nodes. Denotes the set of edges. Represent the set of edge weights and initialize the parameters; Based on network operation status and external influences, an initial fault is injected into a specific node of the network, and then the fault propagation probability is calculated as follows: ; in, Indicates time Faults can occur at any time from the node. propagation to nodes The probability of; This represents the Sigmoid function, ensuring that the probability value is between 0 and 1; and These are parameters used to adjust the model's sensitivity. Represents a node to Distance or communication delay, Represents a node To the node The weights; Calculate the cumulative impact score for each node. This reflects the impact of a fault on the node's service capabilities, and is represented as follows: ; in, Represents a node The set of upstream nodes, This represents the probability that a fault propagates from node u to node v at time t. Adjusted based on real-time monitoring data and Based on the parameters and the results of the fault impact analysis, formulate corresponding response measures.

8. The intelligent power communication fault diagnosis method according to claim 2, characterized in that, The operational data includes occurrence frequency, response time, and fault handling efficiency. The performance data is used to evaluate the performance of each part of the system using the collected data, and then the system parameters and operating strategies are adjusted based on the performance evaluation results.

9. A smart power communication fault diagnosis system, characterized in that, The system includes: The edge computing deployment module is used to deploy edge computing devices on nodes of the power communication network, collect real-time data, design preprocessing strategies to preprocess the real-time data, and define communication protocols between nodes to optimize communication between nodes. The feature extraction and compression module is used to build a real-time feature extraction and data compression protocol to extract and compress features from the preprocessed data of the edge computing device. The optimal path calculation module is used to select the optimal path to transmit the compressed data to the central processor. Simultaneously, it dynamically adjusts the data transmission path based on real-time network traffic conditions and uses a cyclic redundancy check (CRC) code to add a checksum to the data to ensure error-free data transmission. The optimal path calculation is as follows: set up Represents a node arrive The distance and path cost function are: ; in, Represents a node To the node The path; The key feature selection module is used to select key features, process the received key features on the central processor using an interpretable machine learning model, and diagnose communication failures. The fault propagation calculation module is used to simulate the fault propagation process on the central server using the Diffusion model to predict the impact of communication failures. The response module is used to define network impact indicators and build decision models. It designs priorities and automatically generates response suggestions based on the diagnosis results of communication failures and the impact caused by the communication failures. The key feature selection module specifically performs the following steps: Key features are selected, and at least one hundred decision trees are set. Each tree is trained using a randomly selected subset of features. Information gain is then used to evaluate the contribution of each feature to fault classification, and the feature that is most helpful in improving classification accuracy is selected. The key features include peak value, mean, standard deviation, and energy spectrum. For a given prediction result, select sample points in its vicinity, and use a simple linear model to approximate the behavior of the complex linear model in the current local context. For the approximated linear model, calculate the partial derivative of the output with respect to each input feature, and evaluate the impact of changes in each feature on the prediction result. The calculation is as follows: ; in, Features in a linear model The coefficient represents the influence weight of the current feature in the local model; a feedback scoring mechanism is introduced to calculate the overall satisfaction, and an adaptive adjustment strategy is used to find the optimal model parameters. Then, a dynamic data reweighting method is introduced to assign the highest weight to newly collected data, while the weight of old data decreases according to its timeliness, as shown below: ; in, Indicates the attenuation coefficient. Indicates the current time. Indicates the time of data collection.

Images