Power distribution network system fault monitoring method and system based on multi-source information

Abstract

The invention provides a power distribution network system fault monitoring method and system based on multi-source information, and the method comprises the steps: obtaining a multi-source monitoring data set of a power distribution network system, carrying out the data feature extraction of the multi-source monitoring data set, and obtaining the time sequence change feature of operation state data and the space correlation feature of environment influence data; calling a pre-constructed fault detection model to carry out cooperative fault analysis on the time sequence change characteristics and the space correlation characteristics, and generating a fault detection result of the power distribution network system; determining the fault type in the power distribution network system and the position characteristic of the fault in the network structure according to the fault detection result; and generating a fault early warning instruction containing the fault positioning coordinate based on the fault type and the position feature. The overall practicability and reliability of power distribution network system fault monitoring are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of fault detection, and more particularly, to a method and system for fault monitoring of a distribution network system based on multi-source information. Background Art

[0002] In the field of operation management of power systems, fault monitoring of distribution network systems is crucial for ensuring the stability of power supply, aiming to promptly detect abnormal states in the distribution network and determine fault-related information. In the prior art, when facing the complex and changeable operation environment of distribution networks, it is often difficult to comprehensively capture various influencing factors of fault occurrence, resulting in limitations in the accuracy and reliability of fault detection results. At the same time, it is also difficult to provide accurate location information in terms of fault location, affecting the timeliness and effectiveness of fault handling. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for fault monitoring of a distribution network system based on multi-source information. The technical solution of the present invention is realized as follows: In a first aspect, the present invention provides a method for fault monitoring of a distribution network system based on multi-source information, the method comprising: obtaining a multi-source monitoring data set of the distribution network system, the multi-source monitoring data set including operation status data and environmental impact data from different monitoring nodes; performing data feature extraction processing on the multi-source monitoring data set to obtain the time-series change features of the operation status data and the spatial correlation features of the environmental impact data; invoking a pre-constructed fault detection model to perform collaborative fault analysis on the time-series change features and the spatial correlation features to generate a fault detection result of the distribution network system; determining the fault type existing in the distribution network system and the position features of the fault in the network structure according to the fault detection result; and generating a fault warning instruction including fault location coordinates based on the fault type and the position features.

[0004] In a second aspect, the present invention provides a fault monitoring system for a distribution network system, including a computer system and a plurality of monitoring devices, the plurality of monitoring devices and the computer system being communicatively connected. The computer system includes a memory and a processor, the memory storing a computer program that runs on the processor, and the processor implementing the method as described above when executing the computer program.

[0005] Advantages of the present invention: The fault monitoring method for a distribution network system based on multi-source information provided by the present invention obtains a multi-source monitoring data set of the distribution network system. This multi-source monitoring data set includes operation status data and environmental impact data from different monitoring nodes. By integrating the internal electrical characteristic data of the system and the external interference factor data, various potential information that may induce faults can be comprehensively captured, avoiding the problem of incomplete analysis of fault causes in traditional single-data-source monitoring. After obtaining the multi-source monitoring data set, data feature extraction processing is performed on it to obtain the time-series change features of the operation status data and the spatial correlation features of the environmental impact data. This differential feature extraction method for different data characteristics can accurately capture the evolution law of electrical parameters over time during the fault occurrence process and the distribution law of environmental factors in geographical regions, making the extracted features more consistent with the actual fault occurrence mechanism. Subsequently, a pre-constructed fault detection model is called to perform collaborative fault analysis on the time-series change features and the spatial correlation features. Through the deep fusion of cross-dimensional features, the limitation that the prior art is difficult to take into account both the dynamic evolution and spatial diffusion characteristics of faults is overcome, and the reliability of the fault detection result is improved. According to the fault detection result, the fault type existing in the distribution network system and the location characteristics of the fault in the network structure are determined. Regarding the fault nature identification and location positioning as an integrated goal, the cumbersome process of manually correlating the results after traditional independent analysis is avoided, and the efficiency of obtaining fault information is improved. Finally, a fault warning instruction including the fault location coordinates is generated based on the fault type and location characteristics and sent to the target monitoring terminal to execute the fault response operation, realizing the direct mapping from fault analysis to engineering application, ensuring that the fault response operation can be carried out in a timely manner based on accurate positioning information, and effectively improving the overall practicality and reliability of the fault monitoring of the distribution network system. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 FIG. is a schematic structural diagram of a fault monitoring system for a distribution network system provided by an embodiment of the present invention.

[0007] Figure 2 FIG. is a flowchart of a fault monitoring method for a distribution network system based on multi-source information provided by an embodiment of the present invention.

[0008] Figure 3 FIG. is a schematic diagram of the composition of a computer system provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0009] Figure 1It is a schematic diagram of the architecture of the distribution network system fault monitoring system provided by the embodiment of the present invention. In the distribution network system fault monitoring system 10, there are multiple monitoring devices 100, a network 200, and a computer system 300. Communication connections are established between the multiple monitoring devices 100 and the computer system 300 through the network 200. The computer system 300 is used to execute the method provided by the embodiment of the present invention. The monitoring device 100 can be various types of sensors or a data acquisition device connected to a sensor, and is used to obtain multi-source monitoring data of the distribution network system and send it to the computer system 300 for processing.

[0010] The embodiment of the present invention provides a distribution network system fault monitoring method based on multi-source information, as Figure 2 shown. This method includes: Step S100: Obtain a multi-source monitoring data set of the distribution network system. The multi-source monitoring data set includes operation status data and environmental impact data from different monitoring nodes.

[0011] The multi-source monitoring data set is a data set related to the distribution network system collected from multiple different sources, and the data sources include different monitoring nodes. The operation status data is data that describes the real-time operation conditions of equipment and lines in the distribution network system. For example, line current data can reflect the magnitude of the current in the line, line voltage data reflects the voltage value on the line, and equipment temperature data shows the temperature of the equipment during operation. These data can directly reflect whether the operation status of the distribution network system is normal. The environmental impact data is data related to the surrounding environment of the distribution network. The meteorological condition data covers meteorological factors such as temperature, humidity, wind speed, and rainfall. The geographical area data includes information such as the geographical location and topography of the distribution network. Environmental factors can affect the operation of the distribution network. For example, severe meteorological conditions may cause line failures.

[0012] As an implementation method, step S100 may specifically include the following steps S110 to S160: Step S110: Collect real-time operation parameter data through multiple sensors deployed on the distribution network lines. The real-time operation parameter data includes line current data, line voltage data, and equipment temperature data.

[0013] The real-time operation parameter data is parameter data that reflects the operation status of lines and equipment collected in real time during the operation of the distribution network. The line current data is the magnitude of the current flowing in the line measured by a current sensor, which can reflect the load condition of the line and whether there are faults such as short circuits. The line voltage data is the voltage value on the line measured by a voltage sensor, and stable voltage is an important indicator for the normal operation of the distribution network. The equipment temperature data is the temperature of the distribution network equipment measured by a temperature sensor, and too high a temperature may mean that the equipment has a fault or is overloaded.

[0014] In the embodiments of the present invention, the multiple sensors deployed on the distribution network lines can be current transformers, voltage transformers, temperature sensors, etc. For example, a current transformer can convert a large current into a small current for measurement, a voltage transformer can convert a high voltage into a low voltage for measurement, and a temperature sensor can measure temperature based on principles such as thermocouples or thermistors. These sensors can be installed at key positions on the line, such as the outgoing end of a substation, the branch point of a line, etc., to collect line current data, line voltage data, and equipment temperature data in real time.

[0015] Step S120: Collect environmental status data around the distribution network through an environmental monitoring device. The environmental status data includes meteorological condition data and geographical area data.

[0016] The environmental status data is relevant data of the environment around the distribution network. The meteorological condition data includes meteorological factors such as temperature, humidity, wind speed, rainfall, lightning activities, etc. These factors may cause faults such as line insulation aging, pole inclination, short circuit, etc. The geographical area data includes information such as the geographical location, topography, and altitude of the distribution network. Different geographical areas may have different geological disaster risks, such as landslides, mudslides, etc., which will damage the facilities of the distribution network.

[0017] In the embodiments of the present invention, the environmental monitoring device can include a weather station, a geographic information system (GIS), etc. The weather station can collect meteorological condition data such as temperature, humidity, wind speed, rainfall, etc. in real time and transmit the data to the monitoring center. The geographic information system can store and manage the geographical area data of the distribution network and obtain information such as the geographical location and topography of the distribution network through technologies such as satellite remote sensing and geographical mapping.

[0018] Step S130: Collect device operation log data through the built-in monitoring module of the distribution network device. The device operation log data includes switch operation records and protection device action records.

[0019] The device operation log data is the record data generated during the operation of the distribution network device, which can reflect the operation status and historical operation conditions of the device. The switch operation record is the opening and closing operation record of the switch device, including information such as the operation time and operation reason, which can help the operation and maintenance personnel understand the usage frequency and operation conditions of the switch device. The protection device action record is the action record of the protection device when a fault is detected, including information such as the action time, action type, and action value, which can help the operation and maintenance personnel analyze the cause of the fault and the location where the fault occurred.

[0020] In the embodiments of the present invention, the built-in monitoring modules of the distribution network equipment can be intelligent controllers or monitoring terminals, etc. These modules can record the switch operation records and protection device action records in real time, and store the data locally or upload it to the monitoring center. For example, in the circuit breaker equipment, the built-in intelligent controller can record the time and reason for each opening and closing operation, and at the same time, when the protection device acts, record information such as the action time, action type, and action value.

[0021] Step S140: Perform timestamp alignment processing on the real-time operation parameter data, environmental status data, and equipment operation log data to ensure the consistency of the time dimension of data from different sources.

[0022] The timestamp alignment processing is to uniformly align the data from different sources according to the time dimension, so that different data can be comparable at the same time point. Since the collection frequencies and times of different data sources may vary, timestamp alignment processing is required to ensure that different sources of data can be accurately correlated when analyzing the data.

[0023] In the embodiments of the present invention, interpolation method or synchronous acquisition method can be used for timestamp alignment processing. For example, for the real-time operation parameter data, environmental status data, and equipment operation log data, if their collection times are inconsistent, the data can be filled in time by the interpolation method so that they have corresponding values at the same time point. The synchronous acquisition method can also be used to ensure that different data sources collect data at the same time point, thus avoiding the problem of inconsistent time dimensions.

[0024] Step S150: Perform data cleaning processing on various types of data after timestamp alignment, remove outliers and missing values, and generate a standardized multi-source monitoring data set.

[0025] Data cleaning processing is to pre-process the collected data and remove outliers and missing values to improve the quality and availability of the data. Outliers are values in the data that deviate significantly from the normal range, which may be caused by sensor failure, data transmission errors, etc. Missing values are situations where there are no values in certain positions in the data, which may be caused by sensor damage, data loss, etc. The standardized multi-source monitoring data set is a data set that has been cleaned and standardized so that data from different sources have the same dimension and format, which facilitates subsequent data analysis and processing. In an embodiment of the present invention, data cleaning processing can be implemented by a variety of methods. For example, for outliers, statistical analysis methods can be used, such as a method based on standard deviation, and values that deviate from the mean by a certain multiple of the standard deviation are regarded as outliers. For missing values, interpolation or filling methods based on association rules can be used to fill them. Standardization processing can use the mean-standard deviation normalization method of the data sequence to convert data of different dimensions into standardized data of unified dimension.

[0026] As an implementation manner, step S150 may specifically include the following steps S151 to S157: Step S151: Perform multi-dimensional correlation analysis on the real-time operating parameter data, environmental status data, and equipment operation log data after timestamp alignment, and build a dynamic association rule base between different types of data. The dynamic association rule base includes the correlation threshold range between line current data and equipment temperature data, and the influence coefficient range between meteorological condition data and line voltage data.

[0027] Multi-dimensional correlation analysis is a comprehensive analysis of different types of data to explore the correlations between them. The dynamic association rule base is a database that stores association rules between different types of data. These rules are dynamically updated as the data changes. The correlation threshold range between line current data and device temperature data is a correlation relationship between line current data and device temperature data. When the line current data changes within a certain range, the device temperature data will also change accordingly within a certain range. This range is the correlation threshold range. The influence coefficient range between meteorological condition data and line voltage data is the degree of influence of meteorological condition data on line voltage data. Different meteorological conditions will have different effects on line voltage, and this degree of influence is expressed by the influence coefficient range.

[0028] In the embodiments of the present invention, multi-dimensional correlation analysis can adopt data mining algorithms, such as association rule mining algorithms, machine learning algorithms, etc. For example, through the association rule mining algorithm, frequent item sets and association rules between line current data and equipment temperature data can be found, so as to determine the range of correlation thresholds. For the influence coefficient interval between meteorological condition data and line voltage data, a model between meteorological condition data and line voltage data can be established through machine learning algorithms, such as linear regression algorithms, decision tree algorithms, etc., so as to determine the influence coefficient interval.

[0029] Step S152: Based on the dynamic association rule library, perform preliminary identification of outliers for various types of data, and mark the data that exceeds the correlation threshold range or influence coefficient interval as suspected outliers.

[0030] The preliminary identification of outliers is to perform preliminary screening on various types of data according to the association rules in the dynamic association rule library to find the data that may be abnormal. Suspected outliers are those data that exceed the correlation threshold range or influence coefficient interval in the dynamic association rule library, but whether these data are truly abnormal still needs to be further verified.

[0031] In the embodiments of the present invention, the real-time operation parameter data, environmental state data, and equipment operation log data can be compared with the association rules in the dynamic association rule library first. If the relationship between the line current data and the equipment temperature data exceeds the correlation threshold range, or the relationship between the meteorological condition data and the line voltage data exceeds the influence coefficient interval, then the data is marked as a suspected outlier. For example, if the line current data suddenly increases, but the equipment temperature data does not increase correspondingly, and the correlation coefficient between the two (such as the Pearson correlation coefficient) exceeds the correlation threshold range, then the line current data is marked as a suspected outlier. It can be understood that the correlation threshold range and the influence coefficient interval can be adaptively set according to actual needs or experience, and the present invention does not limit this. During normal operation, the change of the line current usually causes a corresponding change in the equipment temperature. Because heat is generated when current passes through the equipment, according to Joule's law, when the current increases, the heat generated by the equipment increases, and the temperature will rise accordingly. Therefore, there is a positive correlation between the line current data and the equipment temperature data. In practical applications, the Pearson correlation coefficient between the line current and the equipment temperature can be calculated based on historical data as the correlation coefficient. If the calculated Pearson correlation coefficient is close to 1, it indicates that there is a strong positive linear correlation between the line current and the equipment temperature; if the Pearson correlation coefficient is close to -1, it is a strong negative linear correlation; if the Pearson correlation coefficient is close to 0, the linear correlation between the two is very weak. A correlation threshold range can be set according to the calculation results. When the real-time calculated correlation coefficient exceeds this range, it is considered that the data may be abnormal.

[0032] Step S153: Cross-validate the marked suspected outliers across data sources, call at least two types of associated data to verify the reasonableness of the suspected outliers. If all the associated data pass the verification, the outlier mark is removed; if at least one type of associated data fails the verification, it is confirmed as an outlier.

[0033] Cross-source data cross-validation is to verify suspected outliers using data from different data sources to improve the accuracy of outlier identification. Associated data are other data related to the suspected outliers, and by analyzing these associated data, it can be judged whether the suspected outliers are reasonable.

[0034] In an embodiment of the present invention, for example, assume that the marked suspected outlier is line current data, then associated data such as equipment temperature data and meteorological condition data can be called for verification. If both the equipment temperature data and the meteorological condition data can support the change of the line current data, that is, all the associated data pass the verification, then the outlier mark for the line current data is removed. If at least one type of associated data fails the verification, for example, the equipment temperature data does not change with the change of the line current data and exceeds the correlation threshold range, it is confirmed as an outlier.

[0035] Step S154: Use an adaptive interpolation algorithm to replace the data unit confirmed as an outlier. The adaptive interpolation algorithm dynamically adjusts the interpolation weight according to the data change trend within a preset time window before and after the outlier. The more drastic the data change trend, the greater the weight proportion of the recent data.

[0036] The preset time window is a time range set before and after the outlier for analyzing the data change trend. The more drastic the data change trend, the greater the weight proportion of the recent data. This is because in the case of drastic data changes, the recent data can better reflect the true change of the data. In an embodiment of the present invention, for example, the preset time window before and after the outlier can be determined first. Then, analyze the data change trend within the preset time window and calculate the data change rate. Dynamically adjust the interpolation weight according to the data change rate. The more drastic the data change trend, the greater the weight proportion of the recent data. Finally, replace the outlier according to the adjusted interpolation weight. For example, if the data change trend before and after the outlier is relatively gentle, a simple linear interpolation method can be used for replacement; if the data change trend is relatively drastic, the weight of the recent data can be increased and a weighted average method can be used for replacement.

[0037] Step S155: Use an association rule-based filling method for the data unit with missing values, generate filling values according to the mapping relationship between the missing data and other data sources in the dynamic association rule library, and the mapping relationship is pre-trained through statistical analysis of historical data in the same period.

[0038] The filling method based on association rules is a method for filling data units with missing values by using the mapping relationship between the missing data in the dynamic association rule library and other data sources. The mapping relationship is an association relationship existing between the missing data and other data sources, and this mapping relationship can be pre-trained through statistical analysis of historical data in the same period.

[0039] In an embodiment of the present invention, exemplarily, other data sources related to the missing data can be found according to the dynamic association rule library first. Then, the mapping relationship between the missing data and other data sources is determined according to the statistical analysis of historical data in the same period. Finally, filling values are generated according to the mapping relationship and the data of other data sources. For example, if there are missing values in the line current data at a certain moment, the device temperature data and meteorological condition data related to the line current data can be found according to the dynamic association rule library. Then, the mapping relationship between the line current data and the device temperature data and meteorological condition data is determined through the statistical analysis of historical data in the same period. Finally, the filling values of the line current data are generated according to the device temperature data, meteorological condition data, and the mapping relationship.

[0040] Step S156: Perform normalization processing on the data set after the outlier replacement and missing value filling are completed, and use the mean-standard deviation normalization method of the data sequence to convert data with different dimensions into standardized data with a unified dimension.

[0041] Normalization processing is to convert data with different dimensions into data with a unified dimension, so that the data is comparable. The mean-standard deviation normalization method of the data sequence is a method for performing normalization processing on the data sequence. By calculating the mean and standard deviation of the data sequence, each data point in the data sequence is subtracted by the mean and then divided by the standard deviation, so as to convert the data sequence into standardized data with a mean of 0 and a standard deviation of 1.

[0042] For example, for real-time operation parameter data, environmental state data, and device operation log data, first calculate the mean and standard deviation of each data sequence. Then, each data point is subtracted by the mean and then divided by the standard deviation to obtain the standardized data. Through normalization processing, data with different dimensions can be converted into standardized data with a unified dimension, which is convenient for subsequent data analysis and processing.

[0043] Step S157: Integrate various types of standardized data to generate a standardized multi-source monitoring data set including timestamp index, monitoring node identifier, and standardized data value.

[0044] Integration is to combine different types of standardized data to form a unified data set. The timestamp index is an index used to identify the data collection time, which can help users quickly locate and query data. The monitoring node identifier is an identifier used to identify the data collection node, which can help users determine the data source. The standardized data value is the data value after being standardized.

[0045] In an embodiment of the present invention, exemplarily, the standardized real-time operation parameter data, environmental status data, and device operation log data can be sorted according to the timestamp index first. Then, the monitoring node identifier is added to each data point. Finally, the sorted data and the monitoring node identifier are combined to generate a standardized multi-source monitoring data set including the timestamp index, the monitoring node identifier, and the standardized data value. For example, the standardized line current data, line voltage data, device temperature data, meteorological condition data, and device operation log data can be sorted according to the timestamp index, the monitoring node identifier is added to each data point, and then these data are combined into a data table to form a standardized multi-source monitoring data set.

[0046] Step S160: Store the standardized multi-source monitoring data set according to the geographical location information of the monitoring nodes, and establish an association index between the data and the physical location.

[0047] Partitioned storage is to divide the standardized multi-source monitoring data set according to the geographical location information of the monitoring nodes and store it in different regions. Establishing an association index between the data and the physical location is to associate the data with the physical location of the monitoring node by establishing an index, which is convenient for users to query and analyze data according to the physical location.

[0048] In an embodiment of the present invention, exemplarily, the geographical location information of the monitoring nodes, such as longitude and latitude coordinates, can be obtained first. Then, the monitoring nodes are divided into different regions according to the geographical location information. Finally, the standardized multi-source monitoring data set is stored according to the regions to which the monitoring nodes belong. The establishment of the association index between the data and the physical location can adopt database index technologies, such as B-tree index, hash index, etc. For example, an index table can be created in the database, and the geographical location information of the monitoring nodes and the corresponding data sets are stored in the index table, and the data can be quickly queried and located through the index table.

[0049] Step S200: Perform data feature extraction processing on the multi-source monitoring data set to obtain the temporal variation characteristics of the operation status data and the spatial association characteristics of the environmental impact data.

[0050] Data feature extraction is to extract information that can reflect the essential features of data from a multi-source monitoring data set. The temporal variation features of the operating state data are the laws and characteristics of the operating state data changing over time, and the spatial correlation features of the environmental impact data are the distribution laws and characteristics of the environmental impact data in space.

[0051] In the embodiments of the present invention, various methods can be used for data feature extraction processing of the multi-source monitoring data set, such as time series analysis methods, spatial analysis methods, etc. For example, for the temporal variation features of the operating state data, time series analysis methods such as autoregressive moving average model (ARMA), long short-term memory network (LSTM), etc. can be used to analyze the changing trend of the operating state data over time. For the spatial correlation features of the environmental impact data, spatial analysis methods such as geographic information system (GIS), spatial autocorrelation analysis, etc. can be used to analyze the distribution law of the environmental impact data in space.

[0052] As an implementation manner, step S200 may specifically include the following steps S210 to S270: Step S210: Perform time series segmentation processing on the operating state data in the multi-source monitoring data set, and divide the continuously collected data into multiple time window units.

[0053] Time series segmentation processing is to divide continuous time series data into multiple non-overlapping time periods, and each time period is called a time window unit.

[0054] In the embodiments of the present invention, the time window length can be selected appropriately according to actual needs for time series segmentation processing. For example, if you want to analyze the operating state changes of the distribution network system in a short time, a shorter time window length can be selected, such as 1 minute, 5 minutes, etc.; if you want to analyze the operating trend of the distribution network system in a longer time, a longer time window length can be selected, such as 1 hour, 1 day, etc. The continuously collected operating state data is divided according to the selected time window length to obtain multiple time window units.

[0055] Step S220: Calculate the change trend features of the operating state data within each time window unit, and the change trend features include the data fluctuation amplitude and the change direction feature.

[0056] The change trend feature is a feature used to describe the change situation of the operating state data within the time window unit. The data fluctuation amplitude reflects the size of the change range of the data within the time window, and the change direction feature indicates whether the data shows an upward trend or a downward trend.

[0057] In the embodiments of the present invention, in order to calculate the change trend characteristics of the operating state data in each time window unit, it is necessary to first perform some preprocessing on the data. For example, for operating state data such as line current data and line voltage data, there may be noise interference, and filtering needs to be performed first to make the data smoother. Then, based on the smoothed data, the data fluctuation amplitude and the change direction characteristics are calculated. The data fluctuation amplitude can be obtained by calculating the difference between the maximum value and the minimum value of the data within the time window; the change direction characteristics can be determined by comparing the magnitudes of the first data point and the last data point within the time window. If the last data point is greater than the first data point, the change direction is a positive change, and vice versa, it is a negative change.

[0058] As an implementation manner, step S220 may specifically include the following steps S221 to S226: Step S221: Perform a moving average process on the operating state data in each time window unit to generate a smoothed operating state data sequence.

[0059] The moving average process reduces the noise and fluctuations in the data by calculating the average value of the data within a certain time window, making the data smoother.

[0060] In the embodiments of the present invention, the moving average process can be implemented in the following manner. Assume that there are n operating state data points within the time window unit , and a moving window length m (m < n) is selected. Starting from the first data point, calculate the average value of the data within each window with a length of m in turn to obtain a smoothed operating state data sequence.

[0061] Step S222: Calculate the difference between adjacent data points in the smoothed operating state data sequence to obtain a data change amount sequence.

[0062] The data change amount sequence is a sequence used to describe the change situation of the smoothed operating state data between adjacent time points, reflecting the change speed and trend of the data.

[0063] In the embodiments of the present invention, for the smoothed operating state data sequence , calculate the difference between adjacent data points , to obtain a data change amount sequence . Each value in this sequence represents the change amount of the operating state data between adjacent time points.

[0064] Step S223: Statistically calculate the maximum value and the minimum value in the data change amount sequence, and calculate the difference between the two as the data fluctuation amplitude characteristic.

[0065] The data fluctuation amplitude characteristic reflects the severity of changes in the operating status data within a time window unit and is measured by the difference between the maximum and minimum values in the data variation sequence. In this embodiment of the present invention, the maximum and minimum values in the data variation sequence are first found. Then, the difference between the two is calculated, and this difference is the data fluctuation amplitude characteristic. The larger the data fluctuation amplitude characteristic, the more severe the changes in the operating status data within the time window unit.

[0066] Step S224: performing sign judgment on the data change sequence to determine whether the data change direction is positive or negative.

[0067] The data change direction is whether the running status data is in an upward or downward trend within the time window unit, which can be determined by performing a sign judgment on the data change sequence. ,if , it means that the data changes positively between the i-th adjacent time points; if , it means that the data changes negatively between the i-th adjacent time points. By counting the positive and negative changes, the direction of data change in the entire time window unit can be determined. For example, if the number of positive changes is greater than the number of negative changes, it can be considered that the data changes positively in the time window unit; otherwise, it changes negatively. If , indicating that the operating status data remains unchanged at the corresponding adjacent time points.

[0068] Step S225: Count the consecutive occurrences of positive changes and negative changes in each time window unit to generate a change direction feature.

[0069] In addition to containing information about positive or negative changes, the change direction feature can also further refine the description of the data change trend by counting the number of consecutive positive and negative changes. In this embodiment of the present invention, the data change sequence is traversed and the number of consecutive positive and negative changes is counted. For example, assuming that the data change sequence is Starting with the first data change, determine whether it is positive or negative. If there are consecutive positive changes, record the number of consecutive positive changes; if there are consecutive negative changes, record the number of consecutive negative changes. This will generate a sequence of the number of consecutive positive and negative changes within each time window unit, and use this sequence as the change direction feature.

[0070] Step S226: normalize the data fluctuation amplitude feature and the change direction feature, and combine the normalized data fluctuation amplitude feature and change direction feature to obtain a change trend feature vector of the time window unit.

[0071] Normalization is to convert data with different ranges and dimensions into a unified range (such as [0, 1]) to facilitate subsequent feature combination and comparison. The change trend feature vector of the time window unit is a vector formed by combining the normalized data fluctuation amplitude feature and the change direction feature, which can more comprehensively describe the change trend of the operation state data within the time window unit.

[0072] In the embodiment of the present invention, the data fluctuation amplitude feature and the change direction feature may have different ranges and dimensions and need to be normalized. For the data fluctuation amplitude feature A, a linear normalization method can be used to convert it into the range of [0, 1]. For the change direction feature, encoding and normalization processing can be performed according to the continuous occurrence times of positive changes and negative changes. For example, the continuous occurrence times of positive changes and the continuous occurrence times of negative changes can be normalized respectively, and then the normalized results are combined together. Finally, the normalized data fluctuation amplitude feature and the change direction feature are combined into a vector to obtain the change trend feature vector of the time window unit.

[0073] Step S230: Perform a correlation analysis on the change trend features of adjacent time window units to generate time series change features reflecting the evolution law of the data over time.

[0074] The correlation analysis is to study the relationship between the change trend features of adjacent time window units. Through this analysis, the evolution law of the operation state data over time can be revealed.

[0075] In the embodiment of the present invention, various methods can be used for the correlation analysis, such as correlation analysis, Granger causality analysis, etc. Taking correlation analysis as an example, calculate the correlation coefficient between the change trend feature vectors of adjacent time window units. The correlation coefficient can measure the degree of linear correlation between two vectors, and its value range is between [-1, 1]. If the correlation coefficient is close to 1, it means that the two vectors are positively correlated, that is, the change trends of adjacent time window units are similar; if the correlation coefficient is close to -1, it means that the two vectors are negatively correlated, that is, the change trends of adjacent time window units are opposite; if the correlation coefficient is close to 0, it means that there is almost no linear correlation between the two vectors. By performing a correlation analysis on the change trend features of adjacent time window units, time series change features reflecting the evolution law of the data over time are obtained.

[0076] Step S240: Perform spatial grid division processing on the environmental impact data in the multi-source monitoring data set, and divide the geographical area into multiple spatial grid units.

[0077] Spatial grid division processing is to divide the geographical area where the distribution network is located into multiple small spatial grid units according to certain rules, so as to facilitate spatial analysis of environmental impact data.

[0078] In the embodiment of the present invention, the spatial grid division processing can select an appropriate grid size according to the size and accuracy requirements of the geographical area. For example, if the geographical area is large, a larger grid size can be selected; if more refined spatial analysis is required, a smaller grid size can be selected. A regular grid division method, such as a square grid or a rectangular grid, can be used to divide the geographical area into multiple spatial grid units. Each spatial grid unit has a unique identifier and geographical location information, which is convenient for subsequent spatial statistics and analysis of environmental impact data.

[0079] Step S250: Calculate the distribution density characteristics of environmental impact data within each spatial grid unit, where the distribution density characteristics include data aggregation degree and regional coverage range characteristics.

[0080] The distribution density characteristics are used to describe the distribution of environmental impact data within the spatial grid unit. The data aggregation degree reflects the concentration degree of the data within the spatial grid unit, and the regional coverage range characteristics reflect the coverage range of the environmental impact data within the spatial grid unit. In the embodiment of the present invention, exemplarily, the number of data points of environmental impact data within each spatial grid unit can be first counted, and this number can reflect the data aggregation degree. Then, determine the distribution boundary of the environmental impact data within the spatial grid unit, and calculate the area of the region enclosed by the boundary, that is, the regional coverage range characteristics. By performing such calculations on each spatial grid unit, the distribution density characteristics of environmental impact data within each spatial grid unit are obtained.

[0081] As an implementation manner, step S250 can specifically include the following steps S251 to S255: Step S251: Perform data point clustering processing on the environmental impact data within each spatial grid unit to identify the dense areas and sparse areas of the data points.

[0082] Data point clustering processing is to group the environmental impact data points within the spatial grid cells according to certain rules, so that the data points within the same group have high similarity, and the data points between different groups have large differences, thereby identifying the dense areas and sparse areas of the data points. In the embodiments of the present invention, various clustering algorithms can be used for data point clustering processing, such as the K-means clustering algorithm, the DBSCAN algorithm, etc. Taking the K-means clustering algorithm as an example, first determine the number of clusters k, and then randomly initialize k cluster centers. Assign each data point to the cluster where the nearest cluster center is located, and then update the positions of the cluster centers. Repeat this process until the cluster centers no longer change or reach the maximum number of iterations. Through clustering processing, the environmental impact data points within the spatial grid cells are divided into different clusters, and each cluster can represent a dense area or a sparse area.

[0083] Step S252: Calculate the ratio of the number of data points in the dense area to the area of the spatial grid cell to obtain the data aggregation degree feature.

[0084] The data aggregation degree feature is an index used to measure the aggregation degree of environmental impact data in the dense area within the spatial grid cell, and is represented by the ratio of the number of data points in the dense area to the area of the spatial grid cell. In the embodiments of the present invention, first, according to the dense area identified in step S251, count the number of data points N in the dense area. Then, obtain the area S of the spatial grid cell. Finally, calculate the data aggregation degree feature D = N / S. The larger the value of the data aggregation degree feature, the higher the aggregation degree of the environmental impact data within the spatial grid cell.

[0085] Step S253: Determine the distribution boundary of the environmental impact data within each spatial grid cell, and calculate the area enclosed by the boundary as the area coverage range feature.

[0086] The distribution boundary is the boundary of the range covered by the environmental impact data within the spatial grid cell. By determining the distribution boundary and calculating the area it encloses, the area coverage range feature can be obtained. In the embodiments of the present invention, various methods can be used to determine the distribution boundary of the environmental impact data within each spatial grid cell, such as the convex hull algorithm, the minimum bounding rectangle algorithm, etc. Taking the convex hull algorithm as an example, use the environmental impact data points within the spatial grid cell as input, and calculate the convex hull of these data points through the convex hull algorithm. The boundary of the convex hull is the distribution boundary of the environmental impact data. Then, calculate the area enclosed by the convex hull, and use this area as the area coverage range feature.

[0087] Step S254: Calculate the overlap degree of the area coverage range features of adjacent spatial grid cells to determine the spatial diffusion degree of the environmental impact data.

[0088] The overlap degree is calculated as the degree of overlap between the area coverage ranges of adjacent spatial grid cells. Through the overlap degree, the spatial diffusion degree of environmental impact data can be determined.

[0089] In an embodiment of the present invention, exemplarily, the area coverage range information of adjacent spatial grid cells, such as the vertex coordinates of a polygon, can be obtained first. Then, the intersection area S1 and the union area S2 of the two area coverage ranges are calculated. Finally, the overlap degree J = S1 / S2 is calculated. The larger the value of the overlap degree J, the higher the overlap degree of the area coverage ranges of adjacent spatial grid cells, and the greater the spatial diffusion degree of the environmental impact data.

[0090] Step S255: Standardize the data aggregation degree feature and the area coverage range feature, and combine the standardized data aggregation degree feature and the area coverage range feature to obtain the distribution density feature vector of the spatial grid cell.

[0091] The standardization process is to convert the data aggregation degree feature and the area coverage range feature into a unified range (such as [0, 1]) to facilitate subsequent feature combination and comparison. The distribution density feature vector of the spatial grid cell is a vector formed by combining the standardized data aggregation degree feature and the area coverage range feature.

[0092] In an embodiment of the present invention, the linear normalization method can be used for the standardization process. For the area coverage range feature S, the linear normalization method is also used for normalization. Finally, the standardized data aggregation degree feature and the area coverage range feature are combined into a vector to obtain the distribution density feature vector of the spatial grid cell.

[0093] Step S260: Conduct a correlation analysis on the distribution density features of adjacent spatial grid cells to generate spatial correlation features reflecting the distribution law of data in the spatial dimension.

[0094] The correlation analysis is to study the relationship between the distribution density features of adjacent spatial grid cells. Through this analysis, the distribution law of environmental impact data in the spatial dimension can be revealed.

[0095] In the embodiments of the present invention, the correlation analysis can adopt spatial autocorrelation analysis methods, such as the global Moran's I index, the local Moran's I index, etc. Taking the global Moran's I index as an example, the spatial autocorrelation between the distribution density eigenvectors of adjacent spatial grid cells is calculated. The global Moran's I index can measure the spatial aggregation degree and distribution pattern of environmental impact data within the entire geographical area. If the global Moran's I index is positive, it indicates that the environmental impact data shows an aggregated distribution in space; if the global Moran's I index is negative, it indicates that the environmental impact data shows a dispersed distribution in space; if the global Moran's I index is close to 0, it indicates that the environmental impact data is randomly distributed in space. By performing correlation analysis on the distribution density characteristics of adjacent spatial grid cells, spatial correlation features reflecting the distribution law of data in the spatial dimension are generated.

[0096] Step S270: Input the temporal change features and the spatial correlation features into the feature fusion module for dimension unification processing to generate a set of fusion features with spatio-temporal correlation.

[0097] The feature fusion module is a module used to fuse the temporal change features and the spatial correlation features. Dimension unification processing is to transform features of different dimensions into a unified dimension space for subsequent fault analysis. In the embodiments of the present invention, the feature fusion module can adopt various architectures, such as a multi-layer perceptron (MLP), a convolutional neural network (CNN), etc. Taking the multi-layer perceptron as an example, the temporal change features and the spatial correlation features are used as inputs, and feature transformation and fusion are performed through the hidden layer of the multi-layer perceptron. In the dimension unification processing, first, the temporal change features and the spatial correlation features are concatenated, and then linear transformation and non-linear activation are performed through the neurons in the hidden layer, so that features of different dimensions are fused in a unified dimension space. Finally, a set of fusion features with spatio-temporal correlation is output.

[0098] Step S300: Invoke the pre-constructed fault detection model to perform collaborative fault analysis on the temporal change features and the spatial correlation features to generate the fault detection result of the distribution network system.

[0099] The pre-constructed fault detection model is a model obtained by training with a large amount of historical data in the training stage, which can analyze the input temporal change features and spatial correlation features and identify possible faults in the distribution network system. Collaborative fault analysis comprehensively considers the temporal change features and the spatial correlation features and uses their mutual relationship to improve the accuracy of fault detection.

[0100] In the embodiments of the present invention, the pre-constructed fault detection model can adopt a deep learning model, such as a recurrent neural network (RNN), a convolutional neural network (CNN), a long short-term memory network (LSTM), etc. In the training stage, historical time-series change features and spatial correlation feature data, as well as corresponding fault labels, are used for training to adjust the parameters of the model so that the model can accurately identify faults. In practical applications, the current time-series change features and spatial correlation features are input into the pre-constructed fault detection model, and the model performs collaborative fault analysis to output the fault detection results of the distribution network system.

[0101] As an implementation manner, step S300 may specifically include the following steps S310 to S360: Step S310: Invoke the feature collaborative encoding unit of the fault detection model. Through the time-series dynamic enhancement subunit in the feature collaborative encoding unit, perform trend enhancement processing on the time-series change features in the time dimension, and through the spatial structure enhancement subunit in the feature collaborative encoding unit, perform relationship enhancement processing on the spatial correlation features in the spatial dimension.

[0102] The feature collaborative encoding unit is used to encode and enhance the time-series change features and spatial correlation features. The time-series dynamic enhancement subunit is a sub-module in the feature collaborative encoding unit, which can perform trend enhancement processing on the time-series change features in the time dimension, making the evolution law of the data over time more obvious. The spatial structure enhancement subunit is another sub-module in the feature collaborative encoding unit, which can perform relationship enhancement processing on the spatial correlation features in the spatial dimension, making the distribution law of the data in the geographical area more prominent.

[0103] In the embodiments of the present invention, the time-series dynamic enhancement subunit can adopt models such as a recurrent neural network (RNN) or a long short-term memory network (LSTM). Taking LSTM as an example, the time-series change features are used as the input, and the time-series change features are processed through the hidden layer of the LSTM to extract the evolution law of the data over time and enhance it. The spatial structure enhancement subunit can adopt models such as a graph neural network (GNN). The spatial correlation features are represented as a graph structure, and through the information transmission of the nodes and edges of the graph neural network, the distribution law of the data in the geographical area is enhanced.

[0104] Step S320: Perform cross-dimensional correlation encoding processing on the time-series change features and spatial correlation features after enhancement processing, establish a time-space mapping relationship to achieve information interaction of features in different dimensions, and generate a collaborative feature matrix.

[0105] Cross - dimensional correlation coding processing is to correlate and code the enhanced time - series change features and spatial correlation features, establish a time - space mapping relationship, so that features in different dimensions can perform information interaction. The collaborative feature matrix is a matrix generated by cross - dimensional correlation coding processing, which contains the correlation information between time - series change features and spatial correlation features.

[0106] In the embodiments of the present invention, the cross - dimensional correlation coding processing can be implemented by the following steps.

[0107] As an implementation manner, step S320 can specifically include the following steps S321~S326: Step S321: Obtain the network topology structure data of the distribution network system, extract the node connection relationship and the line hierarchy relationship, and construct a node adjacency matrix and a line weight matrix.

[0108] The network topology structure data is information describing the connection relationship between each node and line in the distribution network system. The node connection relationship indicates which nodes are connected, and the line hierarchy relationship indicates the different levels and importance of the lines. The node adjacency matrix is a matrix used to represent the connection relationship between nodes, and the elements in the matrix indicate whether there is a connection between two nodes. The line weight matrix is a matrix used to represent the weights of the lines, and the weights can be set according to factors such as the length and capacity of the lines.

[0109] In the embodiments of the present invention, the network topology structure data of the distribution network system can be obtained through the design drawings of the distribution network, geographic information system (GIS), etc. Extract the node connection relationship and the line hierarchy relationship from the network topology structure data, and construct a node adjacency matrix and a line weight matrix according to these relationships. For example, if there is a connection between node i and node j, the element in the i - th row and j - th column of the node adjacency matrix is 1, otherwise it is 0. The elements in the line weight matrix can be assigned values according to factors such as the length and capacity of the lines.

[0110] Step S322: Perform a matrix multiplication operation on the enhanced time - series change features and the node adjacency matrix to obtain time - series correlation features with node - associated attributes.

[0111] Through the matrix multiplication operation of multiplying the enhanced time - series change features by the node adjacency matrix, the connection relationship between nodes can be incorporated into the time - series change features, and time - series correlation features with node - associated attributes can be obtained.

[0112] In an embodiment of the present invention, it is assumed that the enhanced temporal variation feature is a matrix X, and the node adjacency matrix is a matrix A. By performing matrix multiplication operation Y = XA, the obtained matrix Y is the temporal correlation feature with node association attributes. Each element in the matrix Y represents the association of the temporal variation feature on different nodes considering the node connection relationship.

[0113] Step S323: Perform a matrix multiplication operation on the enhanced spatial association feature and the line weight matrix to obtain a spatial hierarchical feature with line hierarchical attributes.

[0114] Similarly, by performing a matrix multiplication operation on the enhanced spatial association feature and the line weight matrix, the hierarchical relationship of the lines can be incorporated into the spatial association feature to obtain a spatial hierarchical feature with line hierarchical attributes. In an embodiment of the present invention, it is assumed that the enhanced spatial association feature is a matrix Z, and the line weight matrix is a matrix W. By performing matrix multiplication operation Y = ZW, the obtained matrix Y is the spatial hierarchical feature with line hierarchical attributes. Each element in the matrix Y represents the association of the spatial association feature on different lines considering the line hierarchical relationship.

[0115] Step S324: Calculate the mutual information value between the temporal correlation feature and the spatial hierarchical feature, and construct a time - space cross - attention weight matrix based on the mutual information value. The mutual information value is used to characterize the degree of association tightness between the features in the time dimension and the spatial dimension.

[0116] The mutual information value is an index to measure the degree of association between two random variables. By calculating the mutual information value between the temporal correlation feature and the spatial hierarchical feature, the degree of association tightness between the features in the time dimension and the spatial dimension can be understood. The time - space cross - attention weight matrix is a matrix constructed based on the mutual information value and is used to weight the features of different dimensions in the subsequent fusion process. In an embodiment of the present invention, exemplarily, the temporal correlation feature and the spatial hierarchical feature are first regarded as two random variables, and the mutual information value is calculated according to their joint probability distribution and marginal probability distribution. The mutual information value can be calculated by existing general mutual information value formulas, which will not be elaborated here. Then, a time - space cross - attention weight matrix is constructed based on the mutual information value. The greater the mutual information value at a position, the greater the corresponding weight.

[0117] Step S325: Perform two - way weighted fusion on the temporal correlation feature and the spatial hierarchical feature through the time - space cross - attention weight matrix to generate a fusion feature matrix.

[0118] Bidirectional weighted fusion means that during the fusion process, the influence of temporal correlation features on spatial hierarchical features and the influence of spatial hierarchical features on temporal correlation features are considered simultaneously. The two features are weighted and fused through a time-space cross-attention weight matrix to obtain a fused feature matrix.

[0119] In an embodiment of the present invention, by way of example, assume that the temporal correlation feature is X, the spatial hierarchical feature is Y, and the time-space cross-attention weight matrix is M. First, calculate X' and Y', where X' = MX and Y' = MY. X' is the temporal correlation feature weighted by the time-space cross-attention weight matrix, and Y' is the spatial hierarchical feature weighted by the time-space cross-attention weight matrix. Then, perform operations such as concatenating or adding Y' and X' to generate a fused feature matrix.

[0120] Step S326: Perform singular value decomposition on the fused feature matrix, retain the singular value components whose cumulative contribution rate exceeds a preset threshold, and generate a dimensionally reduced collaborative feature matrix. The preset threshold is determined by verifying the historical fault detection accuracy.

[0121] Singular value decomposition can decompose a matrix into the product of three matrices. By retaining the singular value components whose cumulative contribution rate exceeds a preset threshold, the noise and redundant information in the matrix can be removed to achieve dimensional reduction.

[0122] In an embodiment of the present invention, perform singular value decomposition on the fused feature matrix F , where U and V are orthogonal matrices, is a diagonal matrix, and the elements on the diagonal are singular values. Calculate the contribution rate of each singular value, where the contribution rate is the ratio of the square of the singular value to the sum of the squares of all singular values. Arrange the singular values in descending order of contribution rate and calculate the cumulative contribution rate. Retain the singular value components whose cumulative contribution rate exceeds the preset threshold to obtain a dimensionally reduced collaborative feature matrix. The preset threshold can be determined by verifying the historical fault detection accuracy. For example, by continuously adjusting the preset threshold and observing the accuracy of historical fault detection, select the preset threshold that maximizes the accuracy.

[0123] Step S330: Input the collaborative feature matrix into the hierarchical feature parsing network of the fault detection model, and sequentially perform feature extraction processing with different levels of abstraction on the input feature matrix through multiple feature parsing layers. The first feature parsing layer extracts basic fault features, the middle feature parsing layer extracts combined fault features, and the last feature parsing layer extracts advanced fault features.

[0124] The hierarchical feature analysis network performs feature extraction processing on the collaborative feature matrix at different levels of abstraction through multiple feature analysis layers, gradually mining the essential features of faults. Basic fault features are the most fundamental forms of fault manifestations, combined fault features are features composed of basic fault features, and advanced fault features are features that can reflect the essence of faults.

[0125] In an embodiment of the present invention, the hierarchical feature analysis network can adopt architectures such as a convolutional neural network (CNN) or a multi-layer perceptron (MLP). Taking CNN as an example, the first-layer feature analysis layer can perform convolution operations using a small convolutional kernel to extract basic fault features. The intermediate-layer feature analysis layer can use different convolutional kernels and pooling operations to combine and abstract the basic fault features to extract combined fault features. The last-layer feature analysis layer can use operations such as global pooling and fully connected layers to further abstract and fuse the combined fault features to extract advanced fault features.

[0126] As an implementation manner, step S330 may specifically include the following steps S331 to S335: Step S331: Input the collaborative feature matrix into the first-layer feature analysis layer of the hierarchical feature analysis network, and perform a sliding window convolution operation on the collaborative feature matrix based on a convolutional kernel of a fixed size to extract the local feature patterns within the window as basic fault features, where the basic fault features reflect the most fundamental forms of fault manifestations in the distribution network system.

[0127] The sliding window convolution operation is a commonly used operation in convolutional neural networks. By performing sliding convolution on the collaborative feature matrix using a convolutional kernel of a fixed size, the local feature patterns within the window can be extracted. Basic fault features are the most fundamental forms of fault manifestations in the distribution network system, such as current mutations and voltage drops during line short circuits. In an embodiment of the present invention, assume that the collaborative feature matrix is C and the size of the convolutional kernel is k×k. Starting from the upper left corner of the collaborative feature matrix, cover the convolutional kernel on a window of the collaborative feature matrix and perform a convolution operation to obtain a convolution result. Then, slide the convolutional kernel one step to the right or down and continue the convolution operation until the entire collaborative feature matrix is covered. Take the convolution results within each window as local feature patterns and extract these local feature patterns as basic fault features.

[0128] Step S332: Input the basic fault features output by the first-layer feature analysis layer into the intermediate-layer feature analysis layer. The intermediate-layer feature analysis layer includes multiple parallel feature combination channels, and each feature combination channel performs combination processing on the basic fault features through different feature combination algorithms to generate different types of combined fault features, where the combined fault features reflect the correlation relationships between different basic fault features.

[0129] The feature combination channels are multiple parallel processing channels in the intermediate layer feature parsing layer. Each channel uses a different feature combination algorithm to combine and process the basic fault features, obtaining different types of combined fault features. The correlation relationships between different basic fault features can help to more deeply understand the fault occurrence mechanism.

[0130] In the embodiments of the present invention, the feature combination channels in the intermediate layer feature parsing layer can adopt different feature combination algorithms, such as linear combination algorithms, non-linear combination algorithms, etc. For example, one feature combination channel can adopt a linear combination algorithm to perform weighted summation on the basic fault features; another feature combination channel can adopt a non-linear combination algorithm, such as using a neural network to perform non-linear transformation and combination on the basic fault features. By combining and processing the basic fault features through multiple parallel feature combination channels, different types of combined fault features are generated.

[0131] Step S333: Perform feature fusion processing on the multiple combined fault features output by the intermediate layer feature parsing layer, and select the feature components with higher contribution degrees in each combined fault feature through a feature voting mechanism to form a fused combined feature.

[0132] Feature fusion processing is to fuse multiple combined fault features to obtain a more representative feature. The feature voting mechanism is a method for selecting feature components. By evaluating the contribution degrees of each feature component in each combined fault feature, the feature components with higher contribution degrees are selected to form a fused combined feature.

[0133] In the embodiments of the present invention, exemplarily, the multiple combined fault features output by the intermediate layer feature parsing layer can be first standardized so that they have the same dimension and range. Then, the contribution degrees of each feature component are evaluated through a feature voting mechanism. For example, the contribution degree can be determined according to indexes such as the accuracy rate or importance of each feature component in historical fault detection. Finally, the feature components with higher contribution degrees are selected to form a fused combined feature.

[0134] Step S334: Input the fused combined feature into the last layer feature parsing layer, perform global correlation modeling on the fused combined feature through a self-attention mechanism, identify the feature patterns that play a key role in fault detection in the fused combined feature, and generate high-level fault features that characterize the essence of the fault.

[0135] The self-attention mechanism is a mechanism that can automatically focus on the correlations between different positions in a sequence. By performing global correlation modeling on the fused combined feature, the feature patterns that play a key role in fault detection can be identified. High-level fault features are features that can characterize the essence of the fault, and they can more accurately identify the fault type.

[0136] In an embodiment of the present invention, by way of example, assume that the fused combined feature is F com , first calculate the query matrix Q, the key matrix K, and the value matrix V, where Q = F com W Q , K = F com W K , V = F com W V , W Q , W K and W V are learnable weight matrices. Then, calculate the attention scores , where d k is the dimension of the key matrix. Finally, calculate the self-attention output Out = AV. By further processing and feature extraction of the self-attention output, high-level fault features characterizing the essence of the fault are generated.

[0137] Step S335: During the feature extraction process of each feature parsing layer, perform feature normalization processing on the output features of each layer to make the dimension consistency between the features of different layers. The feature normalization processing is performed by the layer normalization method, and the standardization transformation is performed based on the mean and variance of each layer of features.

[0138] The feature normalization processing is to make the features of different layers have the same dimension and range, which is convenient for subsequent feature comparison and fusion. The layer normalization method performs the standardization transformation based on the mean and variance of each layer of features. By using the layer normalization method, the output features of each layer are subjected to feature normalization processing to make the dimension of the features between different layers consistent.

[0139] Step S340: Perform a dynamic allocation operation of feature contribution degrees between the layers of the hierarchical feature parsing network. According to the matching degree between the features extracted by the current layer and the historical fault features, adjust the weight ratio of each layer of features in the final decision. The higher the matching degree, the greater the weight ratio.

[0140] The dynamic allocation operation of feature contribution degrees is to dynamically adjust the weights of each layer of features in the final decision according to the matching degree between the features extracted by the current layer and the historical fault features, so that the features with high matching degrees play a greater role in the decision. In an embodiment of the present invention, by way of example, first establish a historical fault feature database to store the feature patterns of different types of faults. Then, at each layer of the hierarchical feature parsing network, calculate the matching degree between the features extracted by the current layer and the historical fault features. The matching degree can be measured by calculating the similarity between the features, such as cosine similarity, Euclidean distance, etc. According to the matching degree, adjust the weight ratio of each layer of features in the final decision. The higher the matching degree, the greater the weight ratio. For example, a weight adjustment function can be used to dynamically adjust the weights according to the matching degree.

[0141] Step S350: Input the high-level fault features output by the hierarchical feature parsing network into the fault type judgment layer of the fault detection model, and perform fault type recognition on the high-level fault features through a multi-classifier combination strategy to generate a fault type membership vector containing the probabilities of various faults.

[0142] The multi-classifier combination strategy is to use multiple classifiers to classify the high-level fault features, and then combine the results of multiple classifiers to obtain a more accurate fault type recognition result. The fault type membership vector is a vector, where each element represents the probability that the high-level fault feature belongs to a certain type of fault. In an embodiment of the present invention, exemplarily, multiple different types of classifiers can be used, such as support vector machines (SVMs), decision trees, neural networks, etc. Input the high-level fault features into these classifiers respectively to obtain the classification results of each classifier. Then, according to indicators such as the accuracy or confidence of each classifier, the classification results are weighted and combined to generate a fault type membership vector containing the probabilities of various faults.

[0143] Step S360: Determine the fault type with the highest membership value in the fault type membership vector as the main fault type, and extract the feature contribution degree distribution information corresponding to the main fault type, and combine the main fault type and the feature contribution degree distribution information into the fault detection result of the distribution network system.

[0144] The fault type with the highest membership value represents the fault type that the high-level fault features are most likely to belong to, and it is used as the main fault type. The feature contribution degree distribution information can help understand which features play a key role in the judgment of the fault type.

[0145] In an embodiment of the present invention, first find the element with the highest membership value in the fault type membership vector, and the fault type corresponding to this element is the main fault type. Then, according to the contribution degrees of each layer of features obtained in the feature contribution degree dynamic allocation operation in different fault type judgments, extract the feature contribution degree distribution information corresponding to the main fault type. For example, in the previous steps, the matching degree of each layer of features with historical fault features and the weight ratio of each layer of features in the final decision are recorded, and the features with high matching degree and large weight ratio are selected according to the main fault type, and their contribution degree distribution information is sorted out.

[0146] Specifically, for each layer of the hierarchical feature parsing network, if the contribution degree of the features of this layer to the judgment of a certain fault type is calculated in the dynamic distribution operation of feature contribution degree, then after determining the main fault type, extract the contribution degree values of the features of each layer under this main fault type. These contribution degree values can be arranged in hierarchical order to form a contribution degree distribution sequence. At the same time, in order to more clearly display this information, the specific feature patterns corresponding to the features of each layer can also be associated, such as basic fault features, combined fault features, etc., to form a complete set of feature contribution degree distribution information.

[0147] Finally, combine the main fault type and the feature contribution degree distribution information into the fault detection result of the distribution network system. This result is presented in a structured manner. For example, it can be stored as a data object, where the main fault type is used as an attribute and the feature contribution degree distribution information is used as another attribute. Such a fault detection result not only clearly indicates the possible fault type that may occur, but also provides detailed information on the basis for this fault judgment, providing strong support for subsequent fault handling and analysis.

[0148] Step S400: Determine the fault type existing in the distribution network system and the location characteristics of the fault in the network structure according to the fault detection result.

[0149] The fault type is different fault forms that may occur in the distribution network system, such as short - circuit faults, open - circuit faults, grounding faults, etc. The location characteristics are the specific location information of the fault in the distribution network structure, including the geographical coordinates of the fault point and the area range affected by the fault, etc. By analyzing the fault detection result, the actual fault type that occurs in the distribution network system and the specific location of the fault in the network structure can be determined.

[0150] In the embodiment of the present invention, determining the fault type and location characteristics is crucial for quickly locating and repairing faults. By deeply analyzing the fault detection result, the main fault type and the feature contribution degree distribution information contained therein can be utilized, combined with the network topology structure and monitoring node information of the distribution network, to gradually narrow down the possible range of the fault, and finally determine the accurate location and type of the fault.

[0151] As an implementation manner, step S400 may specifically include the following steps S410 - S470: Step S410: Parse the fault detection result to extract the main fault type and the feature contribution degree distribution information.

[0152] Parsing means to conduct a detailed analysis and processing of the fault detection result to extract the required key information. The main fault type is the fault type determined to be most likely to occur in the fault detection, and the feature contribution degree distribution information is the information about the features that play a key role in the fault detection and their contribution degrees.

[0153] In the embodiments of the present invention, the fault detection results may be stored in a complex data structure, such as a database record containing multiple fields or a file in a specific format. The parsing process can be implemented programmatically. According to the specific format of the fault detection results, corresponding parsing algorithms are used to extract the main fault types and the distribution information of feature contribution degrees. For example, if the fault detection results are a data file in JSON format, a JSON parsing library can be used to extract the main fault types and the distribution information of feature contribution degrees by specifying the field names.

[0154] Step S420: Identify at least one key monitoring node with the highest contribution degree to fault detection based on the distribution information of feature contribution degrees.

[0155] A key monitoring node is a monitoring node whose collected data plays an important role in fault judgment during the fault detection process. The distribution information of feature contribution degrees contains the contribution degrees of the data features collected by each monitoring node to fault detection. By analyzing this information, key monitoring nodes can be identified.

[0156] In the embodiments of the present invention, exemplarily, the distribution information of feature contribution degrees can be sorted first, and each monitoring node is arranged in descending order of contribution degree. Then, select at least one monitoring node with the highest contribution degree as the key monitoring node. For example, a contribution degree threshold can be set, and the monitoring nodes with contribution degrees exceeding this threshold are selected as key monitoring nodes; or directly select the monitoring nodes ranked among the top several in terms of contribution degree.

[0157] Step S430: Determine the initial candidate location of the fault according to the geographical location information of the key monitoring node.

[0158] The geographical location information is the specific location of the key monitoring node in the geographical space, such as represented by longitude and latitude coordinates. The initial candidate location is the location where the fault may occur inferred from the location of the key monitoring node, and it is the basis for further determining the accurate location of the fault.

[0159] In the embodiments of the present invention, the following strategies can be adopted to determine the initial candidate location of the fault based on the geographical location information of the key monitoring nodes. If there are multiple key monitoring nodes, the central location of these nodes can be taken as the initial candidate location; or according to the fault type and the data characteristics collected by the monitoring nodes, analyze the possible propagation direction and scope of the fault, and combine the locations of the monitoring nodes to determine an area most likely to contain the fault point as the initial candidate location. For example, if the fault type is a short-circuit fault and some key monitoring nodes detect an abnormal increase in current, then according to the current propagation direction and the line topology structure, it can be inferred that the fault may occur on the line near these monitoring nodes, and a section of the line close to the monitoring nodes is taken as the initial candidate location.

[0160] Step S440: Obtain the network topology structure data of the distribution network system, and analyze the line connection relationship and equipment association relationship around the initial candidate location based on the network topology structure data.

[0161] The network topology structure data is information describing the connection relationship between each node (such as a substation, a pole, etc.) and the lines in the distribution network, which can be represented in the form of a graph or a matrix. The line connection relationship is the connection method and direction of each line, and the equipment association relationship is the association situation between the equipment (such as switches, transformers, etc.) on the line and the line and other equipment.

[0162] In the embodiments of the present invention, the network topology structure data of the distribution network system can be obtained through the distribution network management system, the geographic information system (GIS), etc. After obtaining the network topology structure data, methods such as graph theory algorithms or database queries can be used to analyze the line connection relationship and equipment association relationship around the initial candidate location. For example, the network topology structure data can be represented as a graph, with the initial candidate location as the central node, search for its adjacent nodes and edges to determine the surrounding line connection relationship; at the same time, query the equipment information related to these lines to determine the equipment association relationship.

[0163] Step S450: Combine the main fault type and the line connection relationship to determine the range of the area that the initial candidate location may affect.

[0164] Different fault types have different influence ranges on the distribution network, and the line connection relationship determines the possible propagation path of the fault. By combining the main fault type and the line connection relationship, the range of the area that the initial candidate location may affect can be determined more accurately.

[0165] In an embodiment of the present invention, exemplarily, for a short - circuit fault, since the short - circuit current will propagate along the line, it may affect the downstream lines and devices connected to the initial candidate location. The propagation range of the short - circuit current can be calculated based on parameters such as the impedance and capacity of the line to determine the possibly affected area. For an open - circuit fault, it may cause power outages in parts of the line and its upstream or downstream areas. The affected area range can be determined according to the connection relationship of the line and the load distribution. For example, if the main fault type is a short - circuit fault and the initial candidate location is on a main line, then along the main line and its branch lines, based on the calculation results of the short - circuit current, the area range where the possibly affected lines and devices are located can be determined.

[0166] Step S460: Calculate the matching degree between each monitoring node and the fault feature through the operation status data and the characteristic contribution degree distribution information of all monitoring nodes within the area range, and generate a fault correlation degree distribution map.

[0167] The operation status data is data about the operation of the distribution network collected by the monitoring node, such as current, voltage, power, etc. The fault feature is a feature pattern related to the fault type, such as the current mutation feature during a short - circuit fault, etc. The matching degree is the similarity between the operation status data of the monitoring node and the fault feature. By calculating the matching degree, the correlation degree between each monitoring node and the fault can be understood. The fault correlation degree distribution map is a visual chart used to show the correlation degree between each monitoring node within the area range and the fault.

[0168] In an embodiment of the present invention, the following steps can be adopted to calculate the matching degree between each monitoring node and the fault feature.

[0169] As an implementation manner, step S460 can specifically include the following steps S461 - S468: Step S461: Obtain the historical fault record data of all monitoring nodes within the area range, and count the fault occurrence frequency of each monitoring node based on the historical fault record data.

[0170] The historical fault record data is the fault information recorded by the monitoring node in the past period of time, including the time and type of the fault occurrence, etc. The fault occurrence frequency is the ratio of the number of times the monitoring node has a fault to the total time within a certain time, which reflects the likelihood of the monitoring node having a fault.

[0171] In the embodiments of the present invention, the historical fault record data of all monitoring nodes within the acquisition area can be obtained through the fault management system or the monitoring database of the distribution network. After obtaining the historical fault record data, statistical analysis methods can be used to calculate the fault occurrence frequency of each monitoring node. For example, the historical fault records can be grouped according to time intervals, the number of faults occurring at each monitoring node within each time interval can be counted, and then the fault occurrence frequency can be calculated.

[0172] Step S462: Extract the feature weight vector related to the main fault type from the feature contribution degree distribution information.

[0173] The feature weight vector is a vector composed of the weight values of each feature related to the main fault type in the feature contribution degree distribution information.

[0174] In the embodiments of the present invention, exemplarily, first, according to the main fault type, the features related to this fault type and their corresponding contribution degree values can be screened out from the feature contribution degree distribution information. Then, these contribution degree values are arranged in the order of the features to form a feature weight vector. For example, if the main fault type is a short - circuit fault, the features related to the short - circuit fault, such as current mutation feature, voltage drop feature, etc., are found in the feature contribution degree distribution information, and the contribution degree values of these features are extracted to form a feature weight vector.

[0175] Step S463: Perform feature extraction on the current operation status data of each monitoring node within the area range to obtain a real - time status feature vector.

[0176] Feature extraction is to extract the feature information that can reflect the operation status of the distribution network from the operation status data of the monitoring node. The real - time status feature vector is a vector composed of the features extracted from the current operation status data, which reflects the operation status of the monitoring node at the current moment.

[0177] In the embodiments of the present invention, the feature extraction of the current operation status data of each monitoring node within the area range can adopt a method similar to that in step S200. For example, for current data, features such as the amplitude and change rate of the current can be extracted; for voltage data, features such as the effective value and phase of the voltage can be extracted. The extracted features are arranged in a certain order to form a real - time status feature vector. Suppose the operation status data of a certain monitoring node includes current and voltage, and the extracted features include current amplitude, current change rate, voltage effective value, and voltage phase, then the real - time status feature vector can be expressed as [current amplitude, current change rate, voltage effective value, voltage phase].

[0178] Step S464: Calculate the cosine similarity between the real - time status feature vector and the feature weight vector as the feature matching degree parameter.

[0179] The cosine similarity represents the degree of similarity between two vectors by calculating the cosine value of the angle between them. The feature matching degree parameter is the cosine similarity between the real-time state feature vector and the feature weight vector, which reflects the matching degree between the current operating state of the monitoring node and the feature pattern of the main fault types. The calculated cosine similarity is used as the feature matching degree parameter.

[0180] Step S465: Determine the spatial attenuation coefficient according to the electrical distance between the initial candidate location and each monitoring node. Among them, the farther the electrical distance is, the smaller the spatial attenuation coefficient is.

[0181] The electrical distance is the degree of electrical connection between two nodes in the distribution network, usually represented by parameters such as impedance and voltage drop. The spatial attenuation coefficient is a coefficient used to measure the attenuation of the fault impact with the increase of the electrical distance, which reflects the degree of influence of the fault on the monitoring nodes at different distances.

[0182] As an implementation, step S465 may specifically include the following steps S4651 to S4656: Step S4651: Obtain the line parameter data of the distribution network system. The line parameter data includes line resistance, line reactance, and line susceptance.

[0183] The line parameter data is the parameter describing the electrical characteristics of the distribution network line. The line resistance reflects the blocking effect of the line on the current, the line reactance reflects the inductance characteristics of the line, and the line susceptance reflects the capacitance characteristics of the line. These parameters play an important role in calculating the electrical distance and fault propagation.

[0184] In the embodiment of the present invention, the line parameter data of the distribution network system can be obtained through the design documents, equipment manuals, or monitoring systems of the distribution network. The line parameter data is usually stored in the form of tables or databases, including parameter values such as the resistance, reactance, and susceptance of each line. For example, the parameter information of each line can be queried from the management system of the distribution network and stored in a data table, where each row corresponds to a line and each column corresponds to a parameter.

[0185] Step S4652: Based on the line parameter data and the network topology structure data, construct an equivalent circuit model between the initial candidate location and each monitoring node. The equivalent circuit model includes series impedance and shunt admittance.

[0186] The equivalent circuit model is a simplified circuit model used to represent the electrical connection relationship between the initial candidate location and each monitoring node. The series impedance is the sum of the line resistance and reactance, and the shunt admittance is the line susceptance. By constructing the equivalent circuit model, it is more convenient to calculate the electrical distance and fault propagation.

[0187] In an embodiment of the present invention, exemplarily, the line connection paths between the initial candidate location and each monitoring node can be determined first according to the network topology structure data. Then, according to the line parameter data, the series impedance and shunt admittance of each line are calculated. The series impedance and shunt admittance of these lines are combined according to the connection paths to construct an equivalent circuit model. For example, if there are multiple lines connecting the initial candidate location and a certain monitoring node, the series impedances of these lines can be added in sequence, and the shunt admittance is calculated accordingly, to obtain the series impedance and shunt admittance of the equivalent circuit model.

[0188] Step S4653: Calculate the impedance modulus values from the initial candidate location to each monitoring node through the equivalent circuit model, where the impedance modulus value is the square root of the sum of the square of the line resistance and the square of the line reactance.

[0189] The impedance modulus value is the magnitude of the series impedance in the equivalent circuit model, which reflects the electrical distance between the initial candidate location and each monitoring node. By calculating the impedance modulus value, the electrical connection degree between the initial candidate location and each monitoring node can be quantified.

[0190] Step S4654: Obtain the real-time power flow data of the distribution network system, and determine the active power transmission direction between the initial candidate location and each monitoring node.

[0191] The real-time power flow data is the power flow situation of the distribution network at the current moment, including the transmission direction and magnitude of the active power and reactive power. The active power transmission direction is the direction in which the active power flows from one node to another node in the distribution network, which is of great significance for judging the fault propagation and influence range.

[0192] In an embodiment of the present invention, the real-time power flow data of the distribution network system can be obtained through the monitoring system or energy management system of the distribution network. The real-time power flow data is usually provided in the form of real-time updates, including the active power, reactive power and power transmission direction of each node. According to the real-time power flow data, the active power transmission direction between the initial candidate location and each monitoring node is determined. For example, if the real-time power flow data shows that the active power of the initial candidate location is transmitted to a certain monitoring node, then it can be determined that the active power transmission direction is from the initial candidate location to that monitoring node.

[0193] Step S4655: Correct the sign of the impedance modulus value according to the active power transmission direction. If the power flows from the initial candidate location to the monitoring node, the impedance modulus value takes a positive value, otherwise it takes a negative value.

[0194] The purpose of symbol correction is to consider the influence of the active power transmission direction on the electrical distance, so that the impedance modulus can more accurately reflect the direction and degree of fault propagation. By performing symbol correction on the impedance modulus, the impact of faults on different monitoring nodes can be more accurately evaluated in subsequent calculations.

[0195] In the embodiments of the present invention, the symbol correction of the impedance modulus according to the active power transmission direction can adopt the following rules. If the active power flows from the initial candidate location to the monitoring node, it indicates that the monitoring node is downstream of the initial candidate location, and the fault is more likely to affect this monitoring node. At this time, the impedance modulus takes a positive value. If the active power flows from the monitoring node to the initial candidate location, it indicates that the monitoring node is upstream of the initial candidate location, and the impact of the fault on this monitoring node is relatively small. At this time, the impedance modulus takes a negative value.

[0196] Step S4656: Perform normalization processing on the corrected impedance modulus, and multiply the normalized impedance modulus by a preset distance influence coefficient to obtain a spatial attenuation coefficient. The distance influence coefficient is determined based on the statistical relationship between the electrical distance and the fault propagation probability in historical fault cases.

[0197] Normalization processing is to convert the corrected impedance modulus into a unified range (such as [0, 1]). The distance influence coefficient is a preset coefficient used to reflect the influence degree of the electrical distance on the fault propagation probability, and is obtained through statistical analysis of the electrical distance and the fault propagation probability in historical fault cases. Then, multiply the normalized impedance modulus by the preset distance influence coefficient to obtain the spatial attenuation coefficient.

[0198] Step S466: Perform weighted multiplication on the fault occurrence frequency, the feature matching degree parameter, and the spatial attenuation coefficient to obtain the fault correlation degree value of each monitoring node.

[0199] The fault correlation degree value is an index used to measure the degree of association between each monitoring node and the fault after comprehensively considering the fault occurrence frequency, the feature matching degree parameter, and the spatial attenuation coefficient. Through the method of weighted multiplication, the influences of these three factors can be integrated to obtain a more comprehensive correlation degree value.

[0200] Step S467: Based on the geographical location coordinates of each monitoring node and the corresponding fault correlation degree value, draw a fault correlation degree distribution map in the geographic information system.

[0201] The geographic information system (GIS) is a system used to store, analyze, and display geospatial data. The fault correlation degree distribution map is a visual chart used to display the degree of association between each monitoring node and the fault within the regional scope. By drawing the fault correlation degree distribution map in the geographic information system, the possible locations and influence ranges of faults can be more intuitively understood.

[0202] In an embodiment of the present invention, based on the geographical location coordinates of each monitoring node and the corresponding fault correlation degree value, the following steps can be adopted to draw a fault correlation degree distribution map in a geographic information system. First, import the geographical location coordinates and fault correlation degree values of each monitoring node into the geographic information system. Then, according to the magnitude of the fault correlation degree value, assign different colors or symbols to each monitoring node. For example, the higher the fault correlation degree value, the darker the color or the larger the symbol. Finally, draw the positions of each monitoring node in the geographic information system and display them according to the assigned colors or symbols to form a fault correlation degree distribution map.

[0203] Step S468: Perform Gaussian filtering on the fault correlation degree distribution map to smooth the influence of local outliers on the overall distribution trend.

[0204] Gaussian filtering is used to smooth the noise and local outliers in the image. In the fault correlation degree distribution map, there may be some local outliers, which may be caused by monitoring errors or other factors and will affect the judgment of the overall distribution trend. By performing Gaussian filtering, the influence of these local outliers can be reduced, making the fault correlation degree distribution map smoother and more accurate. In an embodiment of the present invention, exemplarily, a suitable Gaussian kernel can be selected first. The Gaussian kernel is a two-dimensional convolution kernel used to perform convolution operations on the image. The size and standard deviation of the Gaussian kernel can be adjusted according to the actual situation to achieve different smoothing effects. Then, apply the Gaussian kernel to the fault correlation degree distribution map and perform convolution operations on each pixel point to calculate the new pixel values. Through the convolution operation, the influence of local outliers is diffused to the surrounding pixel points, thereby smoothing the influence of local outliers on the overall distribution trend.

[0205] Step S470: Based on the distribution law of the correlation degree values in the fault correlation degree distribution map, determine the precise position and the boundary of the influence range of the fault, and combine the precise position and the boundary of the influence range of the fault into a position feature.

[0206] The distribution law of the correlation degree values reflects the correlation degree between each monitoring node and the fault. By analyzing these laws, the exact location of the fault and the boundary of the influence range can be determined. The exact location is the specific location where the fault occurs, and the boundary of the influence range is the boundary of the area that the fault may affect. In the embodiment of the present invention, exemplarily, the area with the highest correlation degree value can be first found in the fault correlation degree distribution map, and this area may be the exact location of the fault. The area with the highest correlation degree value can be determined by calculating the maximum value of the correlation degree value or using methods such as clustering algorithms. Then, according to the distribution of the correlation degree values, the boundary of the fault influence range is determined. A correlation degree threshold can be set, and the area with a correlation degree value greater than this threshold is regarded as the fault influence range, and its boundary is the boundary of the influence range. Its boundary is reflected in two aspects: on the one hand, it is the geographical coordinate boundary, that is, the contour formed by the geographical location coordinates of the outermost monitoring nodes within this range; on the other hand, it is the monitoring node coverage boundary, which is defined by the monitoring nodes at the edge of the fault influence range with correlation degree values close to the threshold. Finally, the exact location of the fault and the boundary of the influence range are combined into a location feature. For example, the exact location can be represented by longitude and latitude coordinates, and the boundary of the influence range can be represented by the vertex coordinates of a polygon.

[0207] Step S500: Generate a fault warning instruction including fault location coordinates based on the fault type and location feature, and send the fault warning instruction to the target monitoring terminal to perform a fault response operation.

[0208] The fault type is the specific type of fault occurring in the distribution network system, such as a short - circuit fault, an open - circuit fault, etc. The location feature is the specific location and influence range information of the fault in the distribution network. The fault location coordinates are the coordinates of the exact location where the fault occurs, usually represented by longitude and latitude. The fault warning instruction is an instruction containing information such as the fault type, location feature, and fault location coordinates, used to send fault warning information to the target monitoring terminal so as to take fault response operations in a timely manner. The target monitoring terminal is the device that receives the fault warning instruction, such as a computer in the monitoring center, a mobile terminal, etc.

[0209] As an implementation manner, step S500 may specifically include the following steps S510 - S580: Step S510: Analyze the warning level rule corresponding to the fault type, determine the priority of the warning signal, and generate a warning priority identifier.

[0210] The warning level rule is the warning level division standard formulated according to different fault types. Different fault types may correspond to different warning levels. The higher the warning level, the higher the severity of the fault. The priority of the warning signal is the signal sending order determined according to the warning level. The signal with a higher priority needs to be processed preferentially. The warning priority identifier is an identifier used to identify the priority of the warning signal, which is convenient for the target monitoring terminal to classify and process the warning signal.

[0211] In an embodiment of the present invention, exemplarily, a warning level rule database may be established first to store warning level information corresponding to different fault types. Then, the corresponding warning level is queried from the database according to the fault type. For example, if the fault type is a short - circuit fault, the warning level corresponding to the short - circuit fault queried from the database is a first - level warning. The priority of the warning signal is determined according to the warning level. The priority of the first - level warning is the highest, and so on. Finally, a warning priority identifier is generated. For example, different priorities can be represented by numbers or letters.

[0212] Step S520: Extract the precise fault location coordinates and the boundary of the influence range from the location distribution feature information.

[0213] The location distribution feature information is the location feature combined in step S470, including the precise location of the fault and the boundary of the influence range. The precise fault location coordinates are the coordinates of the specific location where the fault occurs, and the boundary of the influence range is the boundary of the area that the fault may affect. By extracting this information, accurate location information can be provided for generating a fault warning instruction in the subsequent process.

[0214] In an embodiment of the present invention, the following method can be used to extract the precise fault location coordinates and the boundary of the influence range from the location distribution feature information. If the location distribution feature information is stored in the form of a data object, which contains attributes of the precise fault location coordinates and the boundary of the influence range, these information can be directly extracted through the attribute names.

[0215] Step S530: Query the distribution network GIS system according to the precise fault location coordinates to obtain the geographical area description information corresponding to the precise fault location coordinates.

[0216] The distribution network GIS system is a system used to store and manage the geographical information of the distribution network, including the geographical location information of each device and line in the distribution network and the corresponding geographical area description information. The geographical area description information is information such as the name and address of the geographical area where the precise fault location is located. By obtaining this information, the location where the fault occurs can be more intuitively understood.

[0217] In an embodiment of the present invention, exemplarily, the precise fault location coordinates can be input into the distribution network GIS system first. Then, the distribution network GIS system queries in the geographical database according to the coordinate information to find the corresponding geographical area. Finally, the description information of this geographical area, such as the street name and the name of the community, is extracted from the geographical database.

[0218] Step S540: Combine the boundary of the influence range and the geographical area description information to generate a set of polygon coordinates of the fault - affected area.

[0219] The boundary of the affected area is the boundary of the area that a fault may affect and can be represented by a series of coordinate points. The geographical area description information is information such as the name and address of the geographical area where the exact location of the fault is located. The polygon coordinate set of the fault affected area combines the boundary of the affected area and the geographical area description information to form a coordinate set of a polygon representing the fault affected area.

[0220] In an embodiment of the present invention, exemplarily, the relative positions of these points in the geographical area can be determined first according to the coordinate points of the boundary of the affected area. Then, in combination with the geographical area description information, the coordinate points are adjusted to more accurately reflect the actual position of the fault affected area. For example, if the coordinate points of the boundary of the affected area are within a certain community, in combination with the geographical area description information of the community, the coordinate points are slightly adjusted to better conform to the actual boundary of the community. Finally, the adjusted coordinate points are connected in sequence to form a polygon, and the coordinate points of the polygon are stored in a set to obtain the polygon coordinate set of the fault affected area.

[0221] Step S550: Convert the polygon coordinate set into a positioning coordinate sequence in a standard geographical coordinate system.

[0222] The standard geographical coordinate system is a unified geographical coordinate representation method, such as the WGS84 coordinate system. The polygon coordinate set may be represented in different coordinate systems. For the convenience of use and sharing in different systems, it is necessary to convert it into a positioning coordinate sequence in a standard geographical coordinate system.

[0223] In an embodiment of the present invention, exemplarily, the coordinate system currently used by the polygon coordinate set can be determined first. Then, according to the coordinate system conversion method, each coordinate point in the polygon coordinate set is converted from the current coordinate system to a coordinate point in the standard geographical coordinate system. The coordinate system conversion method can be implemented by using professional geographical information processing software or libraries, such as the Pyproj library in Python. Finally, the converted coordinate points are arranged in sequence to form a positioning coordinate sequence. For example, if the polygon coordinate set is represented in a certain local coordinate system, the Pyproj library is used to convert each coordinate point into a coordinate point in the WGS84 coordinate system, and then these coordinate points are stored in a list to obtain the positioning coordinate sequence.

[0224] Step S560: Integrate the early warning priority identifier, the positioning coordinate sequence, and the current timestamp to generate a fault early warning data frame containing multi-dimensional information.

[0225] The warning priority identifier is an identifier used to identify the priority of warning signals. The positioning coordinate sequence is a coordinate sequence representing the fault - affected area, and the current timestamp is the time when the fault warning instruction is generated. The fault warning data frame is a data structure containing multi - dimensional information. By integrating the warning priority identifier, the positioning coordinate sequence, and the current timestamp, it can provide more comprehensive fault warning information.

[0226] In an embodiment of the present invention, exemplarily, the warning priority identifier, the positioning coordinate sequence, and the current timestamp can be stored in different variables first. Then, these variables are combined into a data object. For example, a dictionary in Python can be used to implement this. The warning priority identifier is used as a key - value pair in the dictionary, the positioning coordinate sequence is used as another key - value pair, and the current timestamp is used as the third key - value pair.

[0227] Step S570: Encapsulate the fault warning data frame into a fault warning instruction that conforms to a preset communication protocol. The fault warning instruction carries the warning priority identifier, the positioning coordinate sequence, and the geographical area description information.

[0228] The preset communication protocol is a communication protocol used for data transmission in the distribution network system, such as the Modbus protocol, the IEC60870 - 5 - 104 protocol, etc. The fault warning instruction is in a data format that conforms to the preset communication protocol and is used to transmit fault warning information in the distribution network system. By encapsulating the fault warning data frame into a fault warning instruction that conforms to the preset communication protocol, it can ensure that the fault warning information can be accurately transmitted to the target monitoring terminal.

[0229] In an embodiment of the present invention, exemplarily, the format and requirements of the preset communication protocol can be understood first, including the definitions of parts such as the data header, the data body, and the check code. Then, according to the requirements of the communication protocol, the warning priority identifier, the positioning coordinate sequence, and the geographical area description information in the fault warning data frame are encoded and added to the data body of the communication protocol. At the same time, information such as the data header and the check code is added according to the regulations of the communication protocol to form a complete fault warning instruction. For example, if the preset communication protocol is the Modbus protocol, according to the Modbus protocol format, the warning priority identifier, the positioning coordinate sequence, and the geographical area description information are encoded into the Modbus data format, and the data header and the check code are added to generate a fault warning instruction that conforms to the Modbus protocol.

[0230] Step S580: Perform data verification processing on the fault warning instruction. By verifying the integrity of the positioning coordinate sequence and the validity of the warning priority identifier, ensure that the instruction content meets the execution requirements of the fault response operation.

[0231] Data verification processing is to check and verify the data in the fault warning instruction to ensure the accuracy and integrity of the data. The integrity of the positioning coordinate sequence refers to whether the coordinate points in the positioning coordinate sequence are complete, without any missing or incorrect ones. The validity of the warning priority identifier means whether the warning priority identifier conforms to the preset rules and is within the legal range. Through data verification processing, the failure of the fault response operation caused by incorrect or incomplete data can be avoided. In an embodiment of the present invention, exemplarily, for the positioning coordinate sequence, check whether the number of coordinate points meets the requirements, whether the format of each coordinate point is correct, and whether it is within the legal geographical range. For example, check whether the longitude and latitude of the coordinate points are within a reasonable range. For the warning priority identifier, check whether it is a preset legal identifier, such as whether it is a legal priority number like "1", "2", "3", etc. If the positioning coordinate sequence is incomplete or the warning priority identifier is invalid, it is considered that the fault warning instruction does not meet the requirements and needs to be corrected or regenerated. For example, if the longitude of a certain coordinate point in the positioning coordinate sequence exceeds the reasonable range, prompt a data error and require re-checking and correcting the positioning coordinate sequence. Only when the positioning coordinate sequence is complete and the warning priority identifier is valid, is it considered that the fault warning instruction meets the execution requirements of the fault response operation and can be sent to the target monitoring terminal.

[0232] It can be understood that in the above introductions of the embodiments of the present invention, various algorithms involved, such as the Euclidean distance algorithm, linear regression algorithm, decision tree algorithm, convex hull algorithm, clustering algorithm, etc., can be obtained from the relevant content in the prior art. For the sake of saving space, they will not be elaborated in the embodiments of the present invention. In addition, those skilled in the art can make detailed supplements according to the common general knowledge in the art when implementing the solutions of the present invention. For example, according to the general knowledge in the art, normalization can be used to eliminate the dimensional conflict before feature fusion, interpolation can be used to eliminate the dimensional difference, historical data, experience or business scenario requirements can be combined to reasonably set the threshold, the model can be trained based on the general model training method, the number of layers in the model structure can be set based on actual needs, and the activation function can be selected, etc. The present invention will no longer give redundant introductions to the overly detailed implementation process.

[0233] An embodiment of the present invention provides a computer system, including a memory and a processor. The memory stores a computer program that can run on the processor, and when the processor executes the computer program, the above method is implemented.

[0234] It should be noted that Figure 3 is a schematic diagram of the hardware entity of a computer system provided by an embodiment of the present invention, as Figure 3As shown, the hardware entities of the computer system 300 include: a processor 310, a communication interface 320, and a memory 330, where: The processor 310 generally controls the overall operation of the computer system 300. The communication interface 320 enables the electronic device to communicate with other terminals or servers through a network. The memory 330 is configured to store instructions and applications executable by the processor 310, and can also cache data to be processed or already processed by the processor 310 and each module in the computer system 300. Data transmission can be carried out between the processor 310, the communication interface 320, and the memory 330 through a bus 340. It should be noted here that: The descriptions of the above storage medium and device embodiments are similar to those of the above method embodiments, and have beneficial effects similar to those of the method embodiments. For the technical details not disclosed in the storage medium and device embodiments of the present invention, please refer to the descriptions of the method embodiments of the present invention for understanding.

Claims

1. A fault monitoring method for a distribution network system based on multi-source information, characterized in that, The method includes: Obtaining a multi-source monitoring data set of the distribution network system, where the multi-source monitoring data set includes operation status data and environmental impact data from different monitoring nodes; Performing data feature extraction processing on the multi-source monitoring data set to obtain the time-series change features of the operation status data and the spatial correlation features of the environmental impact data; Invoking a pre-constructed fault detection model to perform collaborative fault analysis on the time-series change features and the spatial correlation features, and generating a fault detection result of the distribution network system; Determining the fault type existing in the distribution network system and the location features of the fault in the network structure according to the fault detection result; Generating a fault warning instruction including fault location coordinates based on the fault type and the location features.

2. The method according to claim 1, characterized in that The obtaining of the multi-source monitoring data set of the distribution network system includes: Collecting real-time operation parameter data through multiple sensors deployed on the distribution network lines, collecting environmental status data around the distribution network through environmental monitoring devices, and collecting device operation log data through the built-in monitoring module of the distribution network devices; Performing timestamp alignment processing on the real-time operation parameter data, environmental status data, and device operation log data; Performing data cleaning processing on various types of data after timestamp alignment, removing outliers and missing values, and generating a standardized multi-source monitoring data set; Storing the standardized multi-source monitoring data set in partitions according to the geographical location information of the monitoring nodes, and establishing an association index between the data and the physical location.

3. The method according to claim 1, wherein The performing of data feature extraction processing on the multi-source monitoring data set to obtain the time-series change features of the operation status data and the spatial correlation features of the environmental impact data includes: Performing time-series segmentation processing on the operation status data in the multi-source monitoring data set, and dividing the continuously collected data into multiple time window units; Calculating the change trend features of the operation status data in each time window unit, where the change trend features include the data fluctuation amplitude and the change direction features; Performing correlation analysis on the change trend features of adjacent time window units to generate time-series change features reflecting the evolution law of the data over time; Performing spatial grid division processing on the environmental impact data in the multi-source monitoring data set, and dividing the geographical area into multiple spatial grid units; Calculating the distribution density features of the environmental impact data in each spatial grid unit, where the distribution density features include the data aggregation degree and the regional coverage range features; Performing correlation analysis on the distribution density features of adjacent spatial grid units to generate spatial correlation features reflecting the distribution law of the data in the spatial dimension; Inputting the time-series change features and the spatial correlation features into a feature fusion module for dimension unification processing to generate a fusion feature set with spatio-temporal correlation.

4. The method according to claim 3, characterized in that, The calculating of the change trend features of the operation status data in each time window unit includes: Performing moving average processing on the operation status data in each time window unit to generate a smoothed operation status data sequence; Calculating the difference between adjacent data points in the smoothed operation status data sequence to obtain a data change amount sequence; Statistically analyze the maximum and minimum values in the sequence of data change amounts, and calculate the difference between the two as the data fluctuation amplitude feature; Perform a sign judgment on the sequence of data change amounts to determine whether the data change direction is a positive change or a negative change; Statistically analyze the consecutive occurrence times of positive and negative changes within each time window unit to generate a change direction feature; Perform a normalization process on the data fluctuation amplitude feature and the change direction feature, and combine the normalized data fluctuation amplitude feature and change direction feature to obtain the change trend feature vector of the time window unit; The calculation of the distribution density feature of the environmental impact data within each spatial grid unit includes: Perform data point clustering on the environmental impact data within each spatial grid unit to identify the dense and sparse areas of the data points; Calculate the ratio of the number of data points in the dense area to the area of the spatial grid unit to obtain the data aggregation degree feature; Determine the distribution boundary of the environmental impact data within each spatial grid unit, and calculate the area enclosed by the boundary as the area coverage range feature; Calculate the overlap degree of the area coverage range features of adjacent spatial grid units to determine the spatial diffusion degree of the environmental impact data; Perform a standardization process on the data aggregation degree feature and the area coverage range feature, and combine the standardized data aggregation degree feature and area coverage range feature to obtain the distribution density feature vector of the spatial grid unit.

5. The method according to claim 1, wherein The invocation of the pre-constructed fault detection model to perform collaborative fault analysis on the time series change feature and the spatial association feature to generate the fault detection result of the distribution network system includes: Invoke the feature collaborative coding unit of the fault detection model, and perform trend enhancement processing in the time dimension on the time series change feature through the time series dynamic enhancement sub-unit in the feature collaborative coding unit, and perform relationship enhancement processing in the spatial dimension on the spatial association feature through the spatial structure enhancement sub-unit in the feature collaborative coding unit; Perform cross-dimensional association coding processing on the enhanced time series change feature and spatial association feature to establish a time-space mapping relationship and generate a collaborative feature matrix; Input the collaborative feature matrix into the hierarchical feature parsing network of the fault detection model, and perform feature extraction processing with different abstraction degrees on the input feature matrix through multiple feature parsing layers in sequence. The first feature parsing layer extracts basic fault features, the middle feature parsing layers extract combined fault features, and the last feature parsing layer extracts advanced fault features; Perform a dynamic allocation operation of feature contribution degrees between the layers of the hierarchical feature parsing network. According to the matching degree between the features extracted in the current layer and the historical fault features, adjust the weight ratio of each layer of features in the final decision. The higher the matching degree, the greater the weight ratio; Input the advanced fault features output by the hierarchical feature parsing network into the fault type judgment layer of the fault detection model, and perform fault type identification on the advanced fault features through a multi-classifier combination strategy to generate a fault type membership degree vector containing the probabilities of various faults; Determine the fault type with the highest membership value according to the fault type membership degree vector as the main fault type, and extract the characteristic contribution degree distribution information corresponding to the main fault type, and combine the main fault type and the characteristic contribution degree distribution information into the fault detection result of the distribution network system.

6. The method according to claim 5, wherein The input of the collaborative feature matrix into the hierarchical feature analysis network of the fault detection model, and sequentially performing feature extraction processing with different abstraction levels on the input feature matrix through multiple feature analysis layers, including: Input the collaborative feature matrix into the first feature analysis layer of the hierarchical feature analysis network, and perform a sliding window convolution operation on the collaborative feature matrix based on a fixed-size convolution kernel to extract the local feature pattern within the window as the basic fault feature; Input the basic fault feature output by the first feature analysis layer into the intermediate layer feature analysis layer. The intermediate layer feature analysis layer includes multiple parallel feature combination channels, and each feature combination channel performs combination processing on the basic fault feature through different feature combination algorithms to generate different types of combined fault features; Perform feature fusion processing on the multiple combined fault features output by the intermediate layer feature analysis layer, and select the feature components with higher contribution degrees in each combined fault feature through a feature voting mechanism to form a fusion combined feature; Input the fusion combined feature into the last feature analysis layer, and perform global correlation modeling on the fusion combined feature through a self-attention mechanism to identify the feature patterns that play a key role in fault detection in the fusion combined feature, and generate high-level fault features representing the essence of the fault.

7. The method according to claim 1, wherein The determination of the fault type existing in the distribution network system and the location characteristics of the fault in the network structure according to the fault detection result, including: Analyze the fault detection result to extract the main fault type and the characteristic contribution degree distribution information; Identify at least one key monitoring node with the highest contribution degree to fault detection based on the characteristic contribution degree distribution information; Determine the initial candidate location of the fault according to the geographical location information of the key monitoring node; Obtain the network topology structure data of the distribution network system, and analyze the line connection relationship and equipment association relationship around the initial candidate location based on the network topology structure data; Combine the main fault type and the line connection relationship to determine the area range that the initial candidate location may affect; Calculate the matching degree between each monitoring node and the fault feature through the operation status data of all monitoring nodes within the area range and the characteristic contribution degree distribution information, and generate a fault correlation degree distribution map; Based on the distribution law of the correlation degree values in the fault correlation degree distribution map, determine the exact location and the boundary of the influence range of the fault, and combine the exact location and the boundary of the influence range of the fault into the location characteristics.

8. The method according to claim 7, wherein The calculation of the matching degree between each monitoring node and the fault feature through the operation status data of all monitoring nodes within the area range and the characteristic contribution degree distribution information, and the generation of a fault correlation degree distribution map, including: Obtain the historical fault record data of all monitoring nodes within the area range, and count the fault occurrence frequency of each monitoring node based on the historical fault record data; Extract the feature weight vector related to the main fault type from the feature contribution degree distribution information; Extract features from the current operating state data of each monitoring node within the area range to obtain a real-time state feature vector; Calculate the cosine similarity between the real-time state feature vector and the feature weight vector as the feature matching degree parameter; Determine the spatial attenuation coefficient according to the electrical distance between the initial candidate position and each monitoring node. The farther the electrical distance, the smaller the spatial attenuation coefficient; Perform weighted multiplication on the fault occurrence frequency, feature matching degree parameter, and spatial attenuation coefficient to obtain the fault correlation degree value of each monitoring node; Based on the geographical location coordinates of each monitoring node and the corresponding fault correlation degree value, draw a fault correlation degree distribution map in the geographic information system; Perform Gaussian filtering on the fault correlation degree distribution map to smooth the influence of local outliers on the overall distribution trend.

9. The method according to claim 1, characterized in that, The generation of a fault warning instruction including fault location coordinates based on the fault type and the location feature includes: Analyze the warning level rule corresponding to the fault type, determine the priority of the warning signal, and generate a warning priority identifier; Extract the exact fault location coordinates and the influence range boundary from the location distribution feature information; Query the distribution network GIS system according to the exact fault location coordinates to obtain the geographical area description information corresponding to the exact fault location coordinates; Combine the influence range boundary and the geographical area description information to generate a polygon coordinate set of the fault influence area; Convert the polygon coordinate set into a positioning coordinate sequence in the standard geographical coordinate system; Fuse the warning priority identifier, the positioning coordinate sequence, and the current timestamp to generate a fault warning data frame containing multi-dimensional information; Package the fault warning data frame into a fault warning instruction that conforms to a preset communication protocol. The fault warning instruction carries the warning priority identifier, the positioning coordinate sequence, and the geographical area description information; Perform data verification processing on the fault warning instruction to verify the integrity of the positioning coordinate sequence and the validity of the warning priority identifier.

10. A fault monitoring system for a distribution network system, characterized in that, It includes a computer system and multiple monitoring devices. The multiple monitoring devices and the computer system are communicatively connected. The computer system includes a memory and a processor. The memory stores a computer program that runs on the processor. When the processor executes the computer program, it implements the method according to any one of claims 1 to 9.

Images