Communication System and Method for Distribution Network Condition Monitoring Based on Industrial Internet

By deploying sensors and terminal devices in the Industrial Internet, performing data preprocessing and improved traffic feature extraction, and combining distribution network topology and historical data, a health status report is generated. This solves the problem of incompatibility in the extraction of distribution network communication traffic features in existing technologies, and enables precise control of communication parameters and operation and maintenance alarms.

CN123121985APending Publication Date: 2026-11-17ANHUI HUICHEN ELECTRIC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611270281.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-20
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies for power distribution network status monitoring under the Industrial Internet cannot adapt to the periodic and sudden operating characteristics of communication traffic, resulting in low matching degree between traffic feature extraction results and actual communication conditions, and failing to achieve accurate communication parameter control and operation and maintenance alarms.

Method used

By deploying sensors and terminal devices to collect power grid parameters, performing data preprocessing and improved flow feature extraction, and combining the distribution network topology and historical operating data, a health status report is generated using a state inference engine. Based on the report, targeted network communication parameter adjustment instructions or maintenance alarms are generated.

Benefits of technology

It enables a comprehensive health status assessment of the distribution network communication network, enriches the analysis levels of anomaly types, locations and degrees, improves the linkage operation logic of communication status monitoring and operation and maintenance scheduling, and adapts to the data analysis needs in complex network transmission environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123121985A_ABST
    Figure CN123121985A_ABST
Patent Text Reader

Abstract

This invention relates to the field of power communication monitoring technology, specifically a communication system and method for power distribution network status monitoring based on the Industrial Internet. The method includes: collecting power grid operating parameters through sensors and terminal equipment deployed at key nodes of the power distribution network to form a real-time monitoring data stream; performing data alignment, outlier removal, and missing value interpolation preprocessing on the data stream to obtain regularized monitoring data; extracting communication traffic features using an improved algorithm adapted to the periodicity and burstiness of power distribution network communication traffic; comparing these features with a network status benchmark model to identify anomalies and outputting preliminary status assessment results; and generating a communication network health status report by fusing the power distribution network topology and historical operating data using a status inference engine, with the decision module issuing parameter adjustment commands or maintenance alarms. This method improves the adaptability of power distribution network communication traffic feature extraction, enhances the comprehensive judgment capability of network status, and achieves the goals of refined monitoring and intelligent operation and maintenance control of the communication network.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of power communication monitoring technology, and in particular to a communication system and method for monitoring the status of power distribution networks based on the Industrial Internet. Background Technology

[0002] In the communication scenarios of power distribution network status monitoring under the Industrial Internet architecture, existing technologies generally collect power grid operating parameters by deploying basic sensing devices and terminal equipment, and complete basic data stream shaping operations based on conventional data processing methods. The industry commonly uses general-purpose flow feature extraction algorithms to process monitoring data. The network status identification stage simply compares the extracted flow features with a basic benchmark model, determining network operation solely based on real-time monitoring data. The status analysis process does not consider the power distribution network's own topology or incorporate historical operating data for comprehensive judgment. Subsequent maintenance instructions and alarm information generation also rely solely on shallow anomaly identification results.

[0003] General-purpose traffic feature extraction algorithms fail to incorporate the inherent operational patterns of distribution network communication traffic into their design, making them unsuitable for the periodic and sudden operational characteristics of industrial internet distribution network communication traffic. Hidden network status information and traffic features within the data are difficult to fully extract, resulting in a low degree of matching between feature extraction results and actual communication conditions. After network anomaly identification, only basic anomaly information is output, lacking a systematic state deduction and analysis mechanism. This hinders a comprehensive assessment of the overall health of the distribution network communication system, makes it difficult to output precise communication parameter control strategies and maintenance alarm information, and fails to meet the application requirements of refined status monitoring and intelligent operation and maintenance of distribution network communication systems in the industrial internet environment. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a communication system and method for power distribution network status monitoring based on the Industrial Internet.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a communication method for power distribution network status monitoring based on the Industrial Internet, comprising: By deploying sensors and terminal equipment at key nodes of the power distribution network, power grid operating parameters are collected to form a real-time monitoring data stream; The real-time monitoring data stream is preprocessed, including data alignment, outlier removal, and missing value imputation, to obtain regularized real-time monitoring data. An improved traffic feature extraction algorithm is used to analyze the regular real-time monitoring data to extract communication traffic features that reflect network status and data characteristics. The improved traffic feature extraction algorithm optimizes the feature extraction process based on the periodicity and burstiness patterns of power distribution network communication traffic in the Industrial Internet. The extracted communication traffic features are compared with a preset network status benchmark model to identify network status anomalies and generate preliminary status assessment results including anomaly type, location and degree. The preliminary status assessment results are input into the status inference engine, and combined with the distribution network topology and historical operating data, a comprehensive health status report of the distribution network communication network is generated. Based on the health status report of the distribution network communication network, the decision module generates and issues targeted network communication parameter adjustment instructions or maintenance alarms, including: The decision-making module receives the health status report of the power distribution network communication network; If the report assesses the overall health of the network as good and there are no specific risks, the decision is to keep the current communication parameters unchanged. If the report indicates local performance degradation but does not reach the fault threshold, then based on the preset parameter adjustment strategy library, a communication parameter adjustment instruction for a specific link or device is generated. The communication parameter adjustment instruction includes adjusting the data reporting frequency, modifying the data transmission route, changing the channel modulation and coding method, or adjusting the quality of service strategy. If the report indicates a high-risk anomaly or infers a clear root cause of the fault, a maintenance alarm is generated that includes a detailed fault description, location information, and urgency level. The generated network communication parameter adjustment instructions or maintenance alarms are sent to the corresponding network devices or pushed to maintenance personnel through the industrial internet management platform.

[0006] As a further aspect of the present invention, the real-time monitoring data stream is preprocessed, including data alignment, outlier removal, and missing value imputation, to obtain regularized real-time monitoring data, including: Data alignment: Since the timestamps of monitoring data from different sensors and terminal devices may have slight deviations, a reference clock-based interpolation method is used to unify all data streams to the same high-precision time base to ensure data synchronization on the time axis. Outlier removal: For each data stream, calculate the statistics within the sliding window, and mark data points that deviate significantly from the data distribution within the window and exceed the physical range as outliers and remove them; Missing value imputation: For missing data points caused by transmission loss or outlier removal, a hybrid imputation method based on historical data from the same period of the data stream and related data streams from adjacent nodes is used to estimate reasonable values ​​for filling. After completing the steps, the output is the regularized real-time monitoring data that is time-synchronized, data-continuous, and conforms to physical rationality.

[0007] As a further aspect of the present invention, the step of analyzing the regularized real-time monitoring data using an improved traffic feature extraction algorithm to extract communication traffic features reflecting network status and data characteristics includes: From the regularized real-time monitoring data, data streams carrying different monitoring parameter types are separated to form multiple sub-data streams; For each of the sub-data streams, the improved traffic feature extraction algorithm is applied for processing. The improved traffic feature extraction algorithm uses a multi-scale sliding window to segment the data stream, and calculates time-domain statistical features, frequency-domain spectral features and burst features simultaneously within each window. The time-domain statistical features include the mean, variance, maximum and minimum arrival rates of data packets within the window, as well as the data packet size distribution; The frequency domain spectral features are obtained by performing a fast Fourier transform on the arrival time series of data packets within the window to extract the main frequency component and its power. The suddenness characteristic is characterized by the proportion of high flow duration within the calculation window and the kurtosis of the sudden event. The various features extracted from all sub-data streams at the same time are combined to form a multi-dimensional feature vector that reflects the overall network status and data characteristics, which serves as the communication traffic feature.

[0008] As a further aspect of the present invention, the improved flow feature extraction algorithm employs a multi-scale sliding window to segment the data stream, and simultaneously calculates time-domain statistical features, frequency-domain spectral features, and burst features within each window, including: To adapt to the diverse rates and periodicity of data streams of different monitoring parameters in distribution network communication, the improved flow feature extraction algorithm defines three sliding windows of different time lengths, which are used to capture short-term bursts, medium-period fluctuations and long-term trends, respectively. The three sliding windows of different time lengths slide synchronously on the data stream, and for the same time point, the data in the corresponding window is captured respectively; Within each window, the calculation routines for the time-domain statistical features, the frequency-domain spectral features, and the burstiness features are executed in parallel. Features calculated from three windows at the same time point are fused together. During fusion, different weights are assigned according to the feature type and window length. Trend features of long-term windows have higher weights, while burst features of short-term windows have higher weights when detecting instantaneous anomalies. Through this multi-scale parallel computing and weighted fusion, the extracted communication traffic features can reflect both the macro-trend of the network status and sensitively capture micro-anomalies.

[0009] As a further aspect of the present invention, the step of comparing the extracted communication traffic features with a preset network state benchmark model to identify network state anomalies includes: The network state benchmark model is generated based on the communication traffic characteristic data of the distribution network under historical normal operating conditions, and defines the normal value range and dynamic change mode of various communication traffic characteristics under different time periods and different network loads. Input the communication traffic feature vector extracted at the current moment into the network state baseline model; The network state baseline model compares each dimension of the communication traffic feature vector with its corresponding normal value range and calculates the deviation. If the deviation of one or more dimensions exceeds the preset threshold, it is determined that a network state abnormality has occurred. Based on the specific combination of feature dimensions that deviate and the degree of deviation, a preset abnormal pattern library is matched to determine the abnormal type; By combining the identification information of the sensors and terminal devices from which the data stream originates, the location of the anomaly can be determined; The degree of anomaly is quantified based on the magnitude of the deviation and its duration.

[0010] As a further aspect of the present invention, the network state benchmark model is trained and generated based on communication traffic characteristic data of the distribution network under historical normal operating conditions, defining the normal value range and dynamic change pattern of various communication traffic characteristics under different time periods and different network loads, including: Collect historical communication traffic characteristics data of the power distribution network during long-term stable operation and without fault alarms, as a training dataset; The training dataset is divided according to weekdays, holidays, different seasonal periods, and different scheduling operation stages; For each segmented time period, statistical analysis is performed on the corresponding communication traffic feature data in the training dataset, the probability distribution of each feature dimension is calculated, and a specific confidence interval is taken as the normal value range of the feature in that time period. Time series models are used to model the dynamic change pattern of each feature within the corresponding time period, capturing its periodicity, trend, and correlation with network load indicators. By integrating the normal value ranges of all time periods and all feature dimensions with the dynamic change pattern model, a complete network state baseline model is formed.

[0011] As a further aspect of the present invention, the preliminary state assessment results are input into the state inference engine, and combined with the distribution network topology and historical operating data, a comprehensive distribution network communication network health status report is generated, including: The state reasoning engine receives the preliminary state assessment results, which include the anomaly type, location, and severity. Obtain the current distribution network topology, including communication link connections, node levels, and key routing paths; Query the historical operation database to obtain historical fault records, performance degradation trends, and maintenance logs of devices and links associated with abnormal locations; Based on knowledge graph reasoning rules, the preliminary state assessment results, the current distribution network topology, and the relevant historical operating data are used as input facts; The knowledge graph reasoning rules define the calculation logic for anomaly propagation paths, fault root cause associations, and state influence ranges. The knowledge graph reasoning rules are applied to determine whether the current anomaly is an isolated event, whether it may trigger a cascading failure, and its potential root cause device or link. Based on the comprehensive reasoning results, a health status report of the distribution network communication network is generated, which includes the current anomaly assessment, the overall network health score, the list of risky links, and suggested points of concern.

[0012] As a further aspect of the present invention, the knowledge graph reasoning rules define the calculation logic for anomaly propagation paths, fault root cause associations, and state influence ranges, including: The calculation logic of the abnormal propagation path is based on the physical connection and logical dependency relationship of the power distribution network communication network. A state dependency graph of nodes and links is constructed. When a node or link is identified as abnormal, other nodes and links that may be affected are searched along its upstream and downstream dependencies to form a propagation path hypothesis. The calculation logic of the fault root cause association is based on the causal reasoning model, which associates and matches multiple observed abnormal phenomena with potential root causes. The causal reasoning model is trained from historical fault cases, calculates the confidence of each potential root cause in explaining all current abnormal phenomena, and selects the root cause with the highest confidence as the inference result. The calculation logic of the state's impact range, combined with the results of the anomaly propagation path and the fault root cause correlation, assesses the potential impact range and severity level of the current anomaly on the continuity of distribution network monitoring services, data integrity, and the reliability of control commands.

[0013] As a further aspect of the present invention, the present invention also includes a communication system for power distribution network status monitoring based on the Industrial Internet, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the communication method for power distribution network status monitoring based on the Industrial Internet as described above.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: By considering the periodic and sudden operational modes of communication traffic in the industrial internet distribution network, the computational logic and processing flow of traffic feature extraction are optimized and adjusted, and the computational framework and processing logic of feature extraction are reconstructed. Feature analysis is adapted to the inherent changing patterns of distribution network communication traffic, broadening the analytical dimensions of communication information in the regularized monitoring data, fully capturing the network state parameters and traffic attribute information implicit within the data, enriching the coverage of feature extraction, and ensuring that the extracted communication traffic features conform to the actual operating conditions of industrial internet communication in the distribution network. This adapts to the data analysis needs in complex network transmission environments, improving the adaptability and completeness of communication traffic feature extraction.

[0015] By incorporating distribution network topology and historical operational data into network status simulation, and relying on a status inference engine to perform in-depth analysis and computation of preliminary status assessment results, the data and judgment dimensions of network status analysis are expanded. This changes the operational mode that relies solely on real-time monitoring data for simple comparison and identification, enriching the reference criteria for network anomaly judgment, refining the analytical levels of anomaly types, locations, and degrees, and forming a comprehensive and complete health status assessment of the distribution network communication network. Based on this complete health status assessment, the decision-making module operates, generating and issuing network communication parameter adaptation and adjustment commands and maintenance alarm information, improving the linkage operation logic between distribution network communication status monitoring and maintenance scheduling, and adapting to the needs of routine status monitoring and control of large-scale distribution network communication networks. Attached Figure Description

[0016] Figure 1 This is a flowchart of the communication method for power distribution network status monitoring based on the Industrial Internet, as described in this invention. Figure 2 A flowchart for preprocessing real-time monitoring data streams; Figure 3 A flowchart illustrating how to analyze data and extract communication traffic features using an improved traffic feature extraction algorithm. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0019] See Figure 1 This invention provides a communication method for power distribution network status monitoring based on the Industrial Internet, the overall implementation process of which includes: By collecting power grid operating parameters through sensors and terminal equipment deployed at key nodes of the distribution network, a real-time monitoring data stream is formed. This real-time monitoring data stream undergoes preprocessing, including data alignment, outlier removal, and missing value imputation, resulting in well-organized real-time monitoring data. An improved traffic feature extraction algorithm is used to analyze the well-organized real-time monitoring data, extracting communication traffic features reflecting network status and data characteristics. This improved algorithm optimizes the feature extraction process based on the periodic and bursty patterns of distribution network communication traffic in the Industrial Internet. The extracted communication traffic features are compared with a preset network status benchmark model to identify network status anomalies and generate preliminary status assessment results including anomaly type, location, and severity. The preliminary status assessment results are input into a status inference engine, which, combined with the distribution network topology and historical operating data, generates a comprehensive distribution network communication network health status report. Based on the distribution network communication network health status report, a decision module generates and issues targeted network communication parameter adjustment instructions or maintenance alarms.

[0020] In one embodiment of the present invention, the process of preprocessing the real-time monitoring data stream to obtain regularized real-time monitoring data includes: (See below) Figure 2 Data alignment: Due to potential slight deviations in timestamps of monitoring data from different sensors and terminal devices, a reference clock-based interpolation method is used to unify all data streams onto the same high-precision time base, ensuring data synchronization on the time axis. Outlier removal: For each data stream, statistics within a sliding window are calculated, and data points that significantly deviate from the data distribution within the window and exceed the physical range are marked as outliers and removed. Missing value imputation: For missing data points caused by transmission loss or outlier removal, a hybrid imputation method based on historical data from the same period of the data stream and related data streams from adjacent nodes is used to estimate reasonable values ​​for filling. After completing the above steps, the output is time-synchronized, continuous, and physically consistent regular real-time monitoring data.

[0021] In practical implementation, sensors and terminal equipment deployed at key nodes of the distribution network include feeder terminal units, transformer monitoring terminals, and environmental temperature and humidity sensors. The operating parameters collected by these sensors and terminal equipment cover three-phase voltage, three-phase current, active power, equipment temperature, and environmental humidity, forming a real-time monitoring data stream. The preprocessing of the real-time monitoring data stream includes data alignment, outlier removal, and missing value interpolation to obtain well-organized real-time monitoring data. In the data alignment stage, due to millisecond-level deviations in the timestamps of monitoring data from different sensors and terminal equipment, a reference clock-based interpolation method is used to unify all data streams onto the same high-precision time base, achieving data synchronization on the time axis. Specifically, Coordinated Universal Time (UTC) is obtained from the distribution network time synchronization system as the reference clock. For each real-time monitoring data stream, based on the correspondence between the original timestamp of the real-time monitoring data stream and the reference clock, linear interpolation is used to map the original sampled values ​​to a unified integer millisecond time grid, ensuring strict alignment of real-time monitoring data streams for different parameters such as voltage, current, and temperature on the time axis.

[0022] In the outlier removal process, for each data stream, statistics within a sliding window are calculated. Data points that significantly deviate from the data distribution within the window and exceed the physical range are marked as outliers and removed. For voltage parameter data streams, the sliding window length is set to 100 consecutive data points. The statistics used are the median and interquartile range of the data points within the window. When a data point deviates from the median by more than three times the interquartile range and exceeds the voltage physical range of 0 to 120% of the rated voltage, the data point is determined as an outlier and directly deleted. For current parameter data streams, the same sliding window length and statistical judgment rules are used, with the current physical range set to 0 to 150% of the rated current. For temperature and ambient humidity parameter data streams, the physical range is determined according to the sensor specifications. Data points that exceed the physical range and are statistically significantly outliers are removed.

[0023] In the missing value interpolation stage, for missing data points caused by transmission loss or outlier removal, a hybrid interpolation method based on historical contemporaneous data from the data stream to which the missing data point belongs and related data streams from adjacent nodes is used to estimate reasonable values ​​for filling. Historical contemporaneous data is taken from the average of normal values ​​within a five-minute time window before and after the same moment on the previous day for the data stream to which the missing data point belongs. Related data streams from adjacent nodes select the same type of monitoring values ​​collected by the three normal nodes with the closest electrical distance to the sensor where the missing data point is located, and the reference values ​​for adjacent nodes are obtained by weighting based on the inverse of the electrical distance. The estimated value of the hybrid interpolation is obtained by combining the average of historical contemporaneous data and the reference values ​​for adjacent nodes according to preset weighting coefficients, where the weight of the average of historical contemporaneous data is 0.6 and the weight of the reference values ​​for adjacent nodes is 0.4. After performing the above hybrid interpolation method at all missing positions in the real-time monitoring data stream, the output is timely synchronized, continuous, and physically consistent regular real-time monitoring data.

[0024] In one embodiment of the present invention, an improved traffic feature extraction algorithm is used to analyze regular real-time monitoring data to extract communication traffic features. The process is as follows: (See...) Figure 3 From the organized real-time monitoring data, data streams carrying different monitoring parameter types are separated into multiple sub-data streams. For each sub-data stream, an improved traffic feature extraction algorithm is applied. This improved algorithm uses a multi-scale sliding window to segment the data stream, defining three sliding windows of different time lengths to capture short-term bursts, medium-period fluctuations, and long-term trends, respectively. These three sliding windows slide synchronously across the data stream, capturing data within their respective windows for the same time point. Within each window, time-domain statistical features, frequency-domain spectral features, and burst characteristics are calculated in parallel. The time-domain statistical features include the mean arrival rate of data packets within the window, the frequency domain spectral features, and the burst characteristics. The features include the difference, maximum and minimum values, and packet size distribution. Frequency domain spectral features are extracted by performing a Fast Fourier Transform on the arrival time series of packets within the window to extract the dominant frequency component and its power. Burst features are characterized by calculating the proportion of high traffic duration within the window and the peak value of the burst event. Features calculated from three windows at the same time point are fused, and different weights are assigned according to feature type and window length. Trend features of long-term windows have higher weights, while burst features of short-term windows have higher weights when detecting instantaneous anomalies. Various features extracted from all sub-data streams at the same time are combined to form a multi-dimensional feature vector reflecting the overall network status and data characteristics, which serves as the communication traffic feature.

[0025] In practical implementation, the regularized real-time monitoring data collected and preprocessed by the feeder terminal units at key nodes of the distribution network includes voltage parameter data streams, current parameter data streams, active power parameter data streams, and equipment temperature parameter data streams. When analyzing the regularized real-time monitoring data using an improved flow feature extraction algorithm, data streams carrying different monitoring parameter types are separated from the regularized real-time monitoring data to form voltage sub-data streams, current sub-data streams, active power sub-data streams, and equipment temperature sub-data streams. The improved flow feature extraction algorithm processes each sub-data stream separately. The improved flow feature extraction algorithm uses a multi-scale sliding window to segment the sub-data streams, defining three sliding windows of different time lengths: a short-time sliding window with a time length of 5 seconds, a medium-cycle sliding window with a time length of 60 seconds, and a long-time sliding window with a time length of 600 seconds. The short-time sliding window is used to capture short-term bursts, the medium-cycle sliding window is used to capture medium-cycle fluctuations, and the long-time sliding window is used to capture long-term trends. Three sliding windows of different durations slide synchronously on each sub-data stream. For the same point in time on the sub-data stream timeline, the short-term sliding window captures data packet records within a 5-second window preceding that point in time, the medium-term sliding window captures data packet records within a 60-second window preceding that point in time, and the long-term sliding window captures data packet records within a 600-second window preceding that point in time.

[0026] Within each sliding window, the improved traffic feature extraction algorithm performs parallel calculations of time-domain statistical features, frequency-domain spectral features, and burstiness features. Time-domain statistical features include the mean, variance, maximum, and minimum packet arrival rates within the window, as well as the packet size distribution. The mean packet arrival rate is the number of packets recorded within the window divided by the sliding window's duration. The variance of the packet arrival rate is the variance of the arrival rates of each sub-interval after dividing the window into equal-length sub-intervals. The maximum packet arrival rate is the largest among all equal-length sub-intervals, and the minimum packet arrival rate is the smallest among all equal-length sub-intervals. The packet size distribution is quantified as the mean and standard deviation of the payload bytes of all packets within the window. Frequency domain spectral features are derived by performing a Fast Fourier Transform (FFT) on the data packet arrival time series within the window to extract the dominant frequency component and its power. Specifically, the data packet arrival events within the sliding window are converted into a binary arrival indication sequence with a sampling period of 0.1 seconds (1 for present, 0 for absent). A Hamming window is applied to the binary arrival indication sequence, and an FFT is calculated to obtain the amplitude spectrum. The frequency point with the largest amplitude in the amplitude spectrum is searched as the dominant frequency component, and the power of the dominant frequency component is taken as the square of the amplitude value corresponding to the dominant frequency component. Burst characteristics are characterized by calculating the proportion of high-flow duration within the window and the kurtosis of the burst event. High flow is defined as a continuous period of time in which the data packet arrival rate exceeds 1.5 times the average data packet arrival rate within the window. The proportion of high-flow duration is calculated as the ratio of the sum of the lengths of all high-flow periods within the window to the length of the sliding window. The kurtosis of the burst event is taken as the kurtosis statistic of the instantaneous data packet arrival rate sequence counted at 0.1-second intervals within the window, used to measure the sharpness of the flow burst.

[0027] Features of the same type calculated from short-term, medium-term, and long-term sliding windows at the same time point are fused, with different weights assigned based on feature type and window length. For mean-type features in time-domain statistical features reflecting trends, features extracted by long-term sliding windows are assigned a weight of 0.5, features extracted by medium-term sliding windows are assigned a weight of 0.3, and features extracted by short-term sliding windows are assigned a weight of 0.2; for sudden features reflecting instantaneous anomalies, features extracted by short-term sliding windows are assigned a weight of 0.6, features extracted by medium-term sliding windows are assigned a weight of 0.3, and features extracted by long-term sliding windows are assigned a weight of 0.1; for the power of the dominant frequency component in frequency-domain spectral features, features extracted by medium-term sliding windows are assigned a weight of 0.5, features extracted by short-term sliding windows are assigned a weight of 0.25, and features extracted by long-term sliding windows are assigned a weight of 0.25. The fusion calculation is performed using the following weighted average method: , in: Represents the eigenvalues ​​after fusion. , , These represent the same type of feature values ​​extracted by short-term sliding windows, medium-term sliding windows, and long-term sliding windows, respectively. , , The weight coefficients for the corresponding sliding window satisfy the following conditions: After performing the multi-scale window feature extraction and weighted fusion for each sub-data stream, the various fusion features extracted from the voltage sub-data stream, current sub-data stream, active power sub-data stream, and equipment temperature sub-data stream at the same time are combined to form a multi-dimensional feature vector that reflects the overall network status and data characteristics. The multi-dimensional feature vector is then output as the communication traffic feature.

[0028] In one embodiment of the present invention, the method for comparing the extracted communication traffic features with a preset network state benchmark model to identify network state anomalies and generate preliminary state assessment results is as follows: The network state benchmark model is trained and generated based on the communication traffic feature data of the distribution network under historical normal operating conditions, defining the normal value range and dynamic change pattern of various communication traffic features under different time periods and different network loads; the communication traffic feature vector extracted at the current moment is input into the network state benchmark model; the network state benchmark model compares each dimension of the communication traffic feature vector with the corresponding normal value range and calculates the deviation; if the deviation of one or more dimensions exceeds a preset threshold, it is determined that a network state anomaly has occurred; according to the specific feature dimension combination and its deviation, a preset anomaly pattern library is matched to determine the anomaly type; the anomaly location is determined by combining the identification information of the sensor and terminal device from which the data flow originates; and the degree of anomaly is quantified according to the magnitude and duration of the deviation.

[0029] The training and generation method of the network state baseline model is as follows: Historical communication traffic characteristic data of the distribution network during long-term stable operation and without fault alarms are collected as training datasets; the training dataset is divided into weekdays, holidays, different seasonal periods, and different scheduling operation stages; for each divided period, the corresponding communication traffic characteristic data in the training dataset are statistically analyzed to calculate the probability distribution of each characteristic dimension, and a specific confidence interval is taken as the normal value range of the characteristic in that period; a time series model is used to model the dynamic change pattern of each characteristic in the corresponding period, capturing its periodicity, trend, and correlation with network load indicators; the normal value ranges of all periods and all characteristic dimensions are integrated with the dynamic change pattern model to form a complete network state baseline model.

[0030] In practical implementation, the network state baseline model is trained and generated based on the communication traffic characteristic data of the distribution network under historical normal operating conditions. It defines the normal value range and dynamic change pattern of various communication traffic characteristics under different time periods and network loads. When training the network state baseline model, historical communication traffic characteristic data of the distribution network during three consecutive years of stable operation without fault alarms are collected as the training dataset. The training dataset is divided according to weekdays, holidays, different seasonal periods, and different scheduling operation stages. For the peak electricity consumption period on summer weekdays after the division, statistical analysis is performed on the mean data packet arrival rate dimension of the corresponding communication traffic characteristic data in the training dataset. The probability distribution of the mean data packet arrival rate during the peak electricity consumption period on summer weekdays is calculated, and the probability density function is obtained using the kernel density estimation method. The 95% confidence interval [11.2, 28.6] data packets / second is taken as the normal value range of the mean data packet arrival rate during the peak electricity consumption period on summer weekdays. Simultaneously, for characteristic dimensions such as packet arrival rate variance, mean packet size, dominant frequency component power, high traffic duration percentage, and kurtosis, probability distributions were calculated for each time period using the same statistical method, and the 95% confidence interval was taken as the normal value range for the corresponding characteristic in that time period. An autoregressive integral moving average model was used to model the dynamic change pattern of the mean packet arrival rate during peak electricity consumption periods on summer weekdays. This model captured the periodic change pattern of the mean packet arrival rate with a 24-hour cycle, its linear upward trend with grid load growth, and its positive correlation with the number of concurrent connections in the network load index. The order of the autoregressive integral moving average model was determined by the Akaike Information Criterion. The normal value ranges for all time periods and all characteristic dimensions were integrated with the dynamic change pattern model to form a complete network state baseline model, which is stored in the model library of the distribution network monitoring system.

[0031] During the detection phase, the communication traffic feature vector at the current moment is obtained from the improved traffic feature extraction algorithm, and then input into the network state baseline model. The current moment is 14:35 on a summer weekday afternoon. The communication traffic feature vector includes fused features from the data stream of the feeder terminal unit. The values ​​of each dimension of the communication traffic feature vector are shown in Table 1.

[0032] Table 1: Dimensions and Values ​​of Communication Traffic Feature Vectors at Current Moment

[0033] The network state baseline model compares each dimension of the communication traffic feature vector with its corresponding normal range and calculates the deviation. The mean data packet arrival rate dimension corresponds to the normal range of [10.8, 29.1] data packets / second during the 2 PM to 3 PM period on summer weekdays. The current value of 42.7 data packets / second exceeds the upper limit of the normal range. The deviation of this dimension is calculated using the following formula: , in: Representing feature dimension The degree of deviation This represents the feature dimension in the current communication traffic feature vector. The measured value, Representing feature dimension The upper limit of the normal range for the corresponding time period. Representing feature dimension The standard deviation of the training dataset for the corresponding time period is 4.6 packets / second for the mean arrival rate, and the calculated deviation for the mean arrival rate is 2.96. The normal range for the variance of the arrival rate is [2.1, 9.8], and the current value of 18.3 exceeds the upper limit of the normal range with a deviation of 3.15. The normal range for the high traffic duration percentage is [0.02, 0.21], and the current value of 0.38 exceeds the upper limit of the normal range with a deviation of 2.83. The normal range for the kurtosis is [1.5, 5.8], and the current value of 9.2 exceeds the upper limit of the normal range with a deviation of 2.47. The normal range for the mean packet size is [180, 420] bytes, and the current value of 312 bytes is within the normal range with a deviation of 0. The normal range for the main frequency component power is [0.10, 0.52], and the current value of 0.47 is within the normal range with a deviation of 0.

[0034] A network status anomaly was identified when the deviations in four dimensions—mean packet arrival rate, variance packet arrival rate, percentage of high-traffic duration, and peak intensity—exceeded a preset threshold of 1.5. Based on the specific characteristic dimension combinations of the deviations—increased mean packet arrival rate, increased variance, increased percentage of high-traffic duration, and increased peak intensity—a preset anomaly pattern library was used. This library defines anomalies as sudden traffic congestion when the mean packet arrival rate exceeds 1.5 times the normal range, accompanied by increased variance and peak intensity exceeding 5.8. Combining the sensor and terminal device identification information of the data stream source, the communication traffic feature vector originated from feeder terminal unit number FTU-2207, which is installed on tower number 22 of the 10kV Industrial Park line. The anomaly location was determined to be the communication link of feeder terminal unit FTU-2207 on tower number 22 of the Industrial Park line. Based on the magnitude and duration of the deviation, the average deviation of the packet arrival rate is 2.96, the deviation of the high traffic duration percentage is 2.83, and the deviation lasts for more than 300 seconds, quantifying the anomaly as severe. Combining the above information, a preliminary status assessment result is generated, which includes the anomaly type as sudden traffic congestion, the anomaly location as the communication link of the feeder terminal unit FTU-2207 on tower 22 of the Industrial Park line, and the anomaly severity as severe.

[0035] In one embodiment of the present invention, the process of inputting the preliminary state assessment results into the state inference engine and generating a distribution network communication network health status report by combining the distribution network topology and historical operating data is as follows: The state inference engine receives the preliminary state assessment results containing the anomaly type, location, and degree; obtains the current distribution network topology, including communication link connection relationships, node levels, and key routing paths; queries the historical operating database to obtain historical fault records, performance degradation trends, and maintenance logs of devices and links associated with the anomaly location; based on knowledge graph inference rules, the preliminary state assessment results, the current distribution network topology, and related historical operating data are used as input facts; the knowledge graph inference rules define the calculation logic for anomaly propagation paths, fault root cause associations, and state impact range; the knowledge graph inference rules are applied to infer whether the current anomaly is an isolated event, whether it may trigger cascading faults, and its potential root cause devices or links; and the inference results are combined to generate a distribution network communication network health status report containing the current anomaly assessment, overall network health score, risk link list, and suggested points of concern.

[0036] In the knowledge graph reasoning rules, the calculation logic for anomaly propagation paths is based on the physical connections and logical dependencies of the distribution network communication network. A state dependency graph of nodes and links is constructed. When a node or link is identified as an anomaly, other potentially affected nodes and links are searched along its upstream and downstream dependencies to form a propagation path hypothesis. The calculation logic for fault root cause association is based on a causal reasoning model. Multiple observed anomalies are associated and matched with potential root causes. This causal reasoning model is trained from historical fault cases. The confidence level of each potential root cause in explaining all current anomalies is calculated, and the root cause with the highest confidence level is selected as the inference result. The calculation logic for the state impact range combines the results of anomaly propagation paths and fault root cause association to assess the potential impact range and severity level of the current anomaly on the continuity of distribution network monitoring services, data integrity, and the reliability of control commands.

[0037] In practice, the state inference engine receives the preliminary state assessment results, which include the anomaly type as sudden traffic congestion, the anomaly location as the communication link of the feeder terminal unit FTU-2207 on tower 22 of the Industrial Park line, and the anomaly severity as severe. The state reasoning engine obtains the current distribution network topology, which is extracted in real time from the distribution network geographic information system database. The current distribution network topology shows that feeder terminal unit FTU-2207 is located at the 7th level node of the 10 kV industrial park line backbone communication link. The uplink direct connection node of feeder terminal unit FTU-2207 is feeder terminal unit FTU-2206, and the downlink direct connection nodes of feeder terminal unit FTU-2207 are feeder terminal units FTU-2208 and FTU-2209, respectively. The communication link connection adopts a fiber optic Ethernet ring network. The eastbound and westbound paths of the ring network are redundant. The critical routing path is the hop-by-hop forwarding path from the eastbound industrial park substation to the end node of the ring network, feeder terminal unit FTU-2212.

[0038] The state inference engine queries the historical operation database to obtain historical fault records, performance degradation trends, and maintenance logs of devices and links associated with abnormal locations. The historical operation database records show that the feeder terminal unit FTU-2207 experienced two communication module reset alarms in the past six months. The communication module's operating temperature has consistently remained between 68 and 73 degrees Celsius, approaching the upper limit of 75 degrees Celsius. The optical power of the fiber optic transceiver associated with the feeder terminal unit FTU-2207 has linearly declined from -8.2 dB / mW to -11.5 dB / mW in the past three months. A year ago, during the rainy season, a signal attenuation event occurred on the fiber optic section from tower 22 to tower 21 of the industrial park line due to water ingress into the junction box. The maintenance log records that a firmware upgrade was performed on the communication module of the feeder terminal unit FTU-2207 in March 2025, but no hardware replacement was performed. Performance degradation trend data indicates that the transmit power of the FTU-2207 fiber optic transceiver in the feeder terminal unit is decreasing at an average rate of 1.1 dBmW per month, and is currently close to the receive sensitivity threshold of -13 dBmW.

[0039] The state reasoning engine, based on knowledge graph reasoning rules, takes preliminary state assessment results, the current distribution network topology, and historical operating data as input facts. The knowledge graph reasoning rules define the calculation logic for anomaly propagation paths, fault root cause correlations, and state impact ranges. Regarding anomaly propagation path calculation, based on the physical connections and logical dependencies of the distribution network communication network, a state dependency graph is constructed with feeder terminal unit FTU-2207 as the center and its upstream and downstream three-hop nodes. Feeder terminal unit FTU-2207 is marked as an anomaly in the state dependency graph. Following the downstream dependency relationship of feeder terminal unit FTU-2207, feeder terminal units FTU-2208 and FTU-2209 are searched, and the real-time communication traffic characteristics of feeder terminal units FTU-2208 and FTU-2209 are queried. After collecting the data, it was found that the average data packet arrival rate of feeder terminal unit FTU-2208 gradually increased to 35.1 data packets / second within the past 5 minutes, exceeding the upper limit of the normal range. The average data packet arrival rate of feeder terminal unit FTU-2209 remained at 21.3 data packets / second, which was within the normal range. Based on this, the propagation path was assumed to be a sudden traffic congestion spreading from feeder terminal unit FTU-2207 to feeder terminal unit FTU-2208, with a propagation hop count of one hop. Feeder terminal unit FTU-2209 was not significantly affected.

[0040] Regarding root cause correlation, an observed anomaly is matched with potential root causes based on a causal inference model, which is trained from historical fault cases. The anomalies input into the causal inference model include sudden congestion of communication traffic in feeder terminal unit FTU-2207, fiber optic transceiver transmit power degradation to -11.5 dB / mW, communication module operating temperature exceeding 73 degrees Celsius, and abnormal traffic following in feeder terminal unit FTU-2208. Potential causes include fiber optic transceiver hardware aging, poor communication module heat dissipation, fiber optic link connector contamination, and terminal equipment firmware defects. The causal inference model uses a Bayesian network structure. The conditional probability table of the Bayesian network is obtained by statistically analyzing 214 fault records with labeled root causes from the historical fault case database. The confidence level of each potential root cause in explaining all current anomalies is calculated; the results are shown in Table 2.

[0041] Table 2: Calculation Results of Confidence Levels for Each Potential Root Cause

[0042] The causal inference model selects the root cause with the highest confidence level, with the fiber optic transceiver hardware aging corresponding to a confidence level of 0.73, as the inference result. Fiber optic transceiver hardware aging can explain the continuous decline in transmit power. This decline in transmit power leads to a decrease in the channel signal-to-noise ratio (SNR), which triggers the transport layer retransmission mechanism, causing an abnormal increase in packet arrival rate and sudden congestion. Simultaneously, the higher operating temperature of the communication module further confirms the abnormally increased power consumption of the fiber optic transceiver due to aging.

[0043] Regarding the scope of impact, based on the results of the anomaly propagation path and the correlation with the root cause of the fault, the potential impact and severity level of the current anomaly on the continuity of distribution network monitoring services, data integrity, and the reliability of control commands are assessed. The anomaly propagation path indicates that the communication links of feeder terminal units FTU-2207 and FTU-2208 have been affected. The fiber optic transceiver transmission power degradation trend continues at the current rate, and it is expected that the transmission power will fall below the receiver sensitivity threshold of -13 dBmW within the next month. At that time, the fiber optic communication link between feeder terminal unit FTU-2207 and the uplink node feeder terminal unit FTU-2206 will experience intermittent interruptions, directly affecting the complete upload of operational status monitoring data of the three distribution transformers downstream of tower No. 22 on the Industrial Park line and the reliable issuance of remote opening and closing control commands. The potential impact area is defined as the power supply section from tower No. 21 to tower No. 24 on the Industrial Park line. The affected services include distribution transformer monitoring data acquisition, fault waveform file return, and load switch remote control operation. The comprehensive severity level is assessed as high risk.

[0044] The comprehensive reasoning results generate a health status report for the distribution network communication network, which includes current anomaly assessment, overall network health score, a list of risky links, and suggested areas of concern. The current anomaly assessment describes a sudden traffic congestion caused by hardware aging of the fiber optic transceiver in feeder terminal unit FTU-2207 on tower 22 of the industrial park line. The anomaly has spread to downstream feeder terminal unit FTU-2208, indicating that the anomaly is not an isolated event and there is a risk of cascading failures. The overall network health score is calculated on a 100-point scale, based on the proportion of nodes without anomalies, the average signal quality of communication links, and a weighted average of historical fault frequencies. The current score has dropped to 62 points. The list of risky links includes a high risk level of aging in the uplink fiber optic link of feeder terminal unit FTU-2207, a medium risk level of congestion in the section from feeder terminal unit FTU-2208 to feeder terminal unit FTU-2207, and a high risk level of decreased signal-to-noise ratio in the section from feeder terminal unit FTU-2206 to feeder terminal unit FTU-2207. Recommended areas of focus are: replacing the fiber optic transceiver module of feeder terminal unit FTU-2207 as soon as possible, checking and cleaning associated fiber optic connectors, and monitoring the traffic recovery status of feeder terminal unit FTU-2208.

[0045] In one embodiment of the present invention, the method for generating and issuing targeted network communication parameter adjustment instructions or maintenance alarms through the decision module based on the distribution network communication network health status report is as follows: The decision module receives the distribution network communication network health status report; if the report assesses the overall network health as good and without specific risks, the decision is to keep the current communication parameters unchanged; if the report indicates local performance degradation but does not reach the fault threshold, then based on a preset parameter adjustment strategy library, a communication parameter adjustment instruction for a specific link or device is generated, including adjusting the data reporting frequency, modifying the data transmission route, changing the channel modulation and coding method, or adjusting the quality of service strategy; if the report indicates a high-risk anomaly or infers a clear fault root cause, a maintenance alarm containing a detailed fault description, location information, and urgency level is generated; the generated network communication parameter adjustment instructions or maintenance alarms are then issued to the corresponding network devices or pushed to maintenance personnel through the industrial internet management platform.

[0046] In practical implementation, the decision-making module receives a health status report of the distribution network communication network. This report includes an assessment that the current anomaly is caused by hardware aging of the fiber optic transceiver in feeder terminal unit FTU-2207 on tower 22 of the industrial park line, leading to sudden traffic congestion that has spread to downstream feeder terminal unit FTU-2208. The overall network health score is 62 points. The risk link list includes a high risk level for the uplink fiber optic link of feeder terminal unit FTU-2207 and a medium risk level for congestion in the section from feeder terminal unit FTU-2208 to feeder terminal unit FTU-2207. The recommended focus is to replace the fiber optic transceiver module of feeder terminal unit FTU-2207 and check the associated fiber optic connectors. After parsing the distribution network communication network health status report, the decision-making module identifies a high-risk anomaly and infers a clear root cause of the fault: hardware aging of the fiber optic transceiver. A maintenance alarm is then generated, containing a detailed fault description, location information, and urgency level.

[0047] The detailed fault description of the maintenance alarm is as follows: A sudden traffic congestion occurred in the communication link of the feeder terminal unit FTU-2207 on tower 22 of the Industrial Park Line. Root cause analysis determined that the fiber optic transceiver hardware is aging, with the current transmit power degrading to -11.5 dBmW. At a degradation rate of 1.1 dBmW per month, it is projected to fall below the receive sensitivity threshold of -13 dBmW in 30 days. The congestion has spread to the downstream node feeder terminal unit FTU-2208. The location information of the maintenance alarm is as follows: the faulty device is the fiber optic transceiver module inside feeder terminal unit FTU-2207, installed on tower 22 of the Industrial Park Line, with geographical coordinates of 31°52′14″N 117°17′08″E, equipment asset number OPT-2207-03, and the associated communication link is the G.652 single-mode fiber optic cable between tower 22 and tower 21 of the Industrial Park Line. The maintenance alarm was classified as urgent based on the following criteria: severe anomaly, risk of cascading failures, and approximately 30 days remaining for resolution. The decision module distributed the generated maintenance alarm to the mobile terminals of the three on-duty maintenance personnel in the power distribution automation operation and maintenance team via the industrial internet management platform's message push service. Simultaneously, the alarm was written to the work order management system, automatically generating a fault work order with the automatically generated fault work order number WO-2026-05-09-0031.

[0048] In some embodiments, if the decision module receives a health status report indicating localized performance degradation but not reaching a fault threshold, it will generate communication parameter adjustment instructions for specific links or devices based on a preset parameter adjustment strategy library. The preset parameter adjustment strategy library is stored in a rule table structure. The rule antecedent includes the performance degradation type, degradation range, affected link characteristics, and device type; the rule consequent is the communication parameter adjustment action. When the performance degradation type is that the average arrival rate of a link's data packets exceeds the upper limit of the normal range but the deviation does not exceed 2.0 and the peak value does not exceed 5.8, a data reporting frequency adjustment strategy is matched, and the data reporting frequency adjustment step size is calculated proportionally to the deviation. The specific calculation method for adjusting the data reporting frequency is as follows: , in: This indicates the adjusted data reporting frequency. Indicates the current data reporting frequency. This indicates that the adjustment factor is 0.3. This represents the current average data packet arrival rate measurement. This indicates the upper limit of the normal range for the average arrival rate of data packets. If performance degradation persists despite multiple adjustments to the data reporting frequency, and the deviation continues to widen, the data transmission routing strategy is modified. This involves querying the available bandwidth of redundant paths in the distribution network topology and selecting the redundant path with the largest available bandwidth as the new route. When the signal-to-noise ratio (SNR) monitoring value falls below the preset SNR degradation threshold and the bit error rate increases, the channel modulation and coding scheme is changed, switching from quadrature phase-shift keying (QPS) modulation to binary phase-shift keying (BPSK), or adjusting the redundancy rate of the forward error correction coding. When link congestion causes real-time monitoring and file return services to compete for bandwidth, the quality of service (QoS) strategy is adjusted. The monitoring data stream is marked as high-priority differential service code point 46, and the file return data stream is marked as low-priority differential service code point 10, and scheduled according to priority-weighted fair queues. The decision module generates network communication parameter adjustment instructions based on the matched parameter adjustment strategy. The network communication parameter adjustment instructions include the target device identifier, the parameter to be adjusted, the target value to be adjusted, and the effective time. The instructions are sent to the communication management agent processes of feeder terminal units FTU-2207 and FTU-2208 via Simple Network Management Protocol (SMLP) messages. After the instructions are sent, the adjustment effect is continuously monitored. If the performance indicators recover to the normal range within 10 minutes, the adjustment is recorded as successful. If the performance indicators do not recover after 10 minutes, the adjustment is upgraded to a maintenance alarm.

[0049] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A communication method for power distribution network condition monitoring based on the Industrial Internet, characterized in that, The method includes: By deploying sensors and terminal equipment at key nodes of the power distribution network, power grid operating parameters are collected to form a real-time monitoring data stream; The real-time monitoring data stream is preprocessed, including data alignment, outlier removal, and missing value imputation, to obtain regularized real-time monitoring data. An improved traffic feature extraction algorithm is used to analyze the regular real-time monitoring data to extract communication traffic features that reflect network status and data characteristics. The improved traffic feature extraction algorithm optimizes the feature extraction process based on the periodicity and burstiness patterns of power distribution network communication traffic in the Industrial Internet. The extracted communication traffic features are compared with a preset network status benchmark model to identify network status anomalies and generate preliminary status assessment results including anomaly type, location and degree. The preliminary status assessment results are input into the status inference engine, and combined with the distribution network topology and historical operating data, a comprehensive health status report of the distribution network communication network is generated. Based on the health status report of the distribution network communication network, the decision module generates and issues targeted network communication parameter adjustment instructions or maintenance alarms, including: The decision-making module receives the health status report of the power distribution network communication network; If the report assesses the overall health of the network as good and there are no specific risks, the decision is to keep the current communication parameters unchanged. If the report indicates local performance degradation but does not reach the fault threshold, then based on the preset parameter adjustment strategy library, a communication parameter adjustment instruction for a specific link or device is generated. The communication parameter adjustment instruction includes adjusting the data reporting frequency, modifying the data transmission route, changing the channel modulation and coding method, or adjusting the quality of service strategy. If the report indicates a high-risk anomaly or infers a clear root cause of the fault, a maintenance alarm is generated that includes a detailed fault description, location information, and urgency level. The generated network communication parameter adjustment instructions or maintenance alarms are sent to the corresponding network devices or pushed to maintenance personnel through the industrial internet management platform.

2. The communication method for power distribution network status monitoring based on the Industrial Internet according to claim 1, characterized in that, The real-time monitoring data stream is preprocessed, including data alignment, outlier removal, and missing value imputation, to obtain well-organized real-time monitoring data, including: Data alignment: Since the timestamps of monitoring data from different sensors and terminal devices may have slight deviations, a reference clock-based interpolation method is used to unify all data streams to the same high-precision time base to ensure data synchronization on the time axis. Outlier removal: For each data stream, calculate the statistics within the sliding window, and mark data points that deviate significantly from the data distribution within the window and exceed the physical range as outliers and remove them; Missing value imputation: For missing data points caused by transmission loss or outlier removal, a hybrid imputation method based on historical data from the same period of the data stream and related data streams from adjacent nodes is used to estimate reasonable values ​​for filling. After completing the steps, the output is the regularized real-time monitoring data that is time-synchronized, data-continuous, and conforms to physical rationality.

3. The communication method for power distribution network status monitoring based on the Industrial Internet according to claim 1, characterized in that, The improved traffic feature extraction algorithm is used to analyze the regularized real-time monitoring data and extract communication traffic features that reflect network status and data characteristics, including: From the regularized real-time monitoring data, data streams carrying different monitoring parameter types are separated to form multiple sub-data streams; For each of the sub-data streams, the improved traffic feature extraction algorithm is applied for processing. The improved traffic feature extraction algorithm uses a multi-scale sliding window to segment the data stream, and calculates time-domain statistical features, frequency-domain spectral features and burst features simultaneously within each window. The time-domain statistical features include the mean, variance, maximum and minimum arrival rates of data packets within the window, as well as the data packet size distribution; The frequency domain spectral features are obtained by performing a fast Fourier transform on the arrival time series of data packets within the window to extract the main frequency component and its power. The suddenness characteristic is characterized by the proportion of high flow duration within the calculation window and the kurtosis of the sudden event. The various features extracted from all sub-data streams at the same time are combined to form a multi-dimensional feature vector that reflects the overall network status and data characteristics, which serves as the communication traffic feature.

4. The communication method for power distribution network status monitoring based on the Industrial Internet according to claim 3, characterized in that, The improved flow feature extraction algorithm uses a multi-scale sliding window to segment the data stream. Within each window, time-domain statistical features, frequency-domain spectral features, and burst features are calculated simultaneously, including: To adapt to the diverse rates and periodicity of data streams of different monitoring parameters in distribution network communication, the improved flow feature extraction algorithm defines three sliding windows of different time lengths, which are used to capture short-term bursts, medium-period fluctuations and long-term trends, respectively. The three sliding windows of different time lengths slide synchronously on the data stream, and for the same time point, the data in the corresponding window is captured respectively; Within each window, the calculation routines for the time-domain statistical features, the frequency-domain spectral features, and the burstiness features are executed in parallel. Features calculated from three windows at the same time point are fused together. During fusion, different weights are assigned according to the feature type and window length. Trend features of long-term windows have higher weights, while burst features of short-term windows have higher weights when detecting instantaneous anomalies. Through this multi-scale parallel computing and weighted fusion, the extracted communication traffic features can reflect both the macro-trend of the network status and sensitively capture micro-anomalies.

5. The communication method for power distribution network status monitoring based on the Industrial Internet according to claim 3, characterized in that, The step of comparing the extracted communication traffic features with a preset network state benchmark model to identify network state anomalies includes: The network state benchmark model is generated based on the communication traffic characteristic data of the distribution network under historical normal operating conditions, and defines the normal value range and dynamic change mode of various communication traffic characteristics under different time periods and different network loads. Input the communication traffic feature vector extracted at the current moment into the network state baseline model; The network state baseline model compares each dimension of the communication traffic feature vector with its corresponding normal value range and calculates the deviation. If the deviation of one or more dimensions exceeds the preset threshold, it is determined that a network state abnormality has occurred. Based on the specific combination of feature dimensions that deviate and the degree of deviation, a preset abnormal pattern library is matched to determine the abnormal type; By combining the identification information of the sensors and terminal devices from which the data stream originates, the location of the anomaly can be determined; The degree of anomaly is quantified based on the magnitude of the deviation and its duration.

6. The communication method for power distribution network status monitoring based on the Industrial Internet according to claim 5, characterized in that, The network state baseline model is generated based on communication traffic characteristic data of the distribution network under historical normal operating conditions. It defines the normal value range and dynamic change pattern of various communication traffic characteristics under different time periods and network loads, including: Collect historical communication traffic characteristics data of the power distribution network during long-term stable operation and without fault alarms, as a training dataset; The training dataset is divided according to weekdays, holidays, different seasonal periods, and different scheduling operation stages; For each segmented time period, statistical analysis is performed on the corresponding communication traffic feature data in the training dataset, the probability distribution of each feature dimension is calculated, and a specific confidence interval is taken as the normal value range of the feature in that time period. Time series models are used to model the dynamic change pattern of each feature within the corresponding time period, capturing its periodicity, trend, and correlation with network load indicators. By integrating the normal value ranges of all time periods and all feature dimensions with the dynamic change pattern model, a complete network state baseline model is formed.

7. The communication method for power distribution network status monitoring based on the Industrial Internet according to claim 5, characterized in that, The preliminary state assessment results are input into the state inference engine, which, combined with the distribution network topology and historical operating data, generates a comprehensive health status report for the distribution network communication network, including: The state reasoning engine receives the preliminary state assessment results, which include the anomaly type, location, and severity. Obtain the current distribution network topology, including communication link connections, node levels, and key routing paths; Query the historical operation database to obtain historical fault records, performance degradation trends, and maintenance logs of devices and links associated with abnormal locations; Based on knowledge graph reasoning rules, the preliminary state assessment results, the current distribution network topology, and the relevant historical operating data are used as input facts; The knowledge graph reasoning rules define the calculation logic for anomaly propagation paths, fault root cause associations, and state influence ranges. The knowledge graph reasoning rules are applied to determine whether the current anomaly is an isolated event, whether it may trigger a cascading failure, and its potential root cause device or link. Based on the comprehensive reasoning results, a health status report of the distribution network communication network is generated, which includes the current anomaly assessment, the overall network health score, the list of risky links, and suggested points of concern.

8. The communication method for power distribution network status monitoring based on the Industrial Internet according to claim 7, characterized in that, The knowledge graph reasoning rules define the calculation logic for anomaly propagation paths, root cause correlations, and state impact ranges, including: The calculation logic of the abnormal propagation path is based on the physical connection and logical dependency relationship of the power distribution network communication network. A state dependency graph of nodes and links is constructed. When a node or link is identified as abnormal, other nodes and links that may be affected are searched along its upstream and downstream dependencies to form a propagation path hypothesis. The calculation logic of the fault root cause association is based on the causal reasoning model, which associates and matches multiple observed abnormal phenomena with potential root causes. The causal reasoning model is trained from historical fault cases, calculates the confidence of each potential root cause in explaining all current abnormal phenomena, and selects the root cause with the highest confidence as the inference result. The calculation logic of the state's impact range, combined with the results of the anomaly propagation path and the fault root cause correlation, assesses the potential impact range and severity level of the current anomaly on the continuity of distribution network monitoring services, data integrity, and the reliability of control commands.

9. A communication system for power distribution network status monitoring based on the Industrial Internet, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the communication method for power distribution network status monitoring based on the Industrial Internet as described in any one of claims 1 to 8.