Component failure prediction method and system based on extreme weather

Abstract

The present invention provides a component failure prediction method and system based on extreme weather. The method comprises the following steps: obtaining an extreme weather data set and an electrical component operating status data set, performing feature extraction on the extreme weather data set to obtain an extreme weather feature set, and performing feature extraction on the electrical component operating status data set to obtain an electrical component operating feature set; correlating and fusing the extreme weather feature set and the electrical component operating feature set to generate an associated feature sequence; calling a pre-trained fault prediction model to perform fault risk prediction on the associated feature sequence to generate a fault probability distribution sequence of the electrical component; determining the fault prediction results of the electrical component under different extreme weather conditions based on the fault probability distribution sequence, improving the adaptability of the prediction results to actual application scenarios, and thereby improving the reliability and effectiveness of electrical component fault prediction as a whole.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a component failure prediction method and system based on extreme weather. Background Art

[0002] With the continuous development of the power system, it is of great significance to analyze the operating status of electrical components and predict possible failures in advance to ensure the stable operation of the power system. At present, the existing electrical component failure prediction methods usually focus on monitoring and analyzing the operating parameters of the electrical components themselves, and achieve failure prediction by establishing a correlation model between operating parameters and failures. Some methods also consider the impact of some conventional environmental factors on electrical components. However, extreme weather is a special environmental factor that has a significant impact on the operating status of electrical components. The complex correlation between its occurrence period, impact area and other characteristics and electrical component failures has not been fully considered. Existing prediction methods are difficult to accurately capture the special failure modes and evolution laws of electrical components under extreme weather conditions, resulting in the accuracy and reliability of fault prediction results being difficult to meet actual application requirements in extreme weather scenarios. Summary of the Invention

[0003] The present invention provides a component failure prediction method and system based on extreme weather.

[0004] In a first aspect, an embodiment of the present invention provides a component failure prediction method based on extreme weather, comprising:

[0005] Acquire an extreme weather data set and an electrical component operating status data set, wherein the extreme weather data set includes records of occurrence periods and affected areas of different types of extreme weather, and the electrical component operating status data set includes records of operating parameters and historical faults of electrical components in corresponding periods;

[0006] Performing feature extraction on the extreme weather data set to obtain an extreme weather feature set, and performing feature extraction on the electrical component operating status data set to obtain an electrical component operating feature set;

[0007] Associating and fusing the extreme weather feature set and the electrical component operation feature set to generate an associated feature sequence;

[0008] Calling a pre-trained fault prediction model to perform fault risk prediction on the associated feature sequence to generate a fault probability distribution sequence of the electrical component;

[0009] Fault prediction results of electrical components under different extreme weather conditions are determined according to the fault probability distribution sequence.

[0010] In a second aspect, an embodiment of the present invention provides a computer system, including:

[0011] a memory storing a computer program;

[0012] The processor is used to load the computer program to implement the above-mentioned extreme weather-based component failure prediction method.

[0013] The component failure prediction method based on extreme weather provided by the present invention ensures that the data used for fault prediction has core dimensional information related to the impact of extreme weather by obtaining an extreme weather phenomenon data set containing records of the occurrence period and affected area of ​​different types of extreme weather, and an electrical component operation status data set containing records of the operating parameters of electrical components in the corresponding period and historical fault records, thereby laying a data foundation for subsequent accurate prediction. Feature extraction processing is performed on the two types of data sets to obtain an extreme weather feature set and an electrical component operation feature set, avoiding information redundancy or key law concealment that may be caused by the direct fusion of original heterogeneous data, and can respectively retain the spatiotemporal correlation characteristics of extreme weather and the parameter-fault mapping characteristics of electrical components, thereby improving the pertinence of feature expression. The two types of feature sets are associated and fused to generate an associated feature sequence. By establishing the internal logical association between extreme weather features and electrical component operation features, the dynamic coupling relationship between extreme weather and electrical component operation is effectively captured, breaking through the limitations of a single data source or simple data splicing in traditional fault prediction, so that the feature sequence can more comprehensively reflect the comprehensive effect of fault influencing factors. A pre-trained fault prediction model is used to assess the fault risk of associated feature sequences, generating a fault probability distribution sequence. Based on the fused joint features, the model learns the fault patterns under the combined effects of extreme weather and electrical component operation, improving the accuracy of fault risk assessment. This fault probability distribution sequence determines the fault prediction results for electrical components under different extreme weather conditions. This distinguishes the differences in fault risk under different extreme weather types, enabling targeted fault prediction and improving the adaptability of prediction results to actual application scenarios, thereby improving the overall reliability and effectiveness of electrical component fault prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a flowchart of a component failure prediction method based on extreme weather provided by an embodiment of the present invention.

[0015] Figure 2 It is a schematic diagram of the composition of a computer system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0016] See also Figure 1 , Figure 1A flowchart of a component failure prediction method based on extreme weather provided by an embodiment of the present invention, which can be executed by a computer system, includes the following steps:

[0017] Step S100: Acquire extreme weather data sets and electrical component operating status data sets. The extreme weather data sets include records of occurrence periods and affected areas of different types of extreme weather. The electrical component operating status data sets include records of operating parameters and historical faults of electrical components in corresponding periods.

[0018] Extreme weather data sets are data sets collected and integrated from information related to extreme weather. Records of the occurrence periods of different types of extreme weather events clarify the specific times at which each extreme weather event begins and ends. These records can be obtained using time-recording devices at meteorological monitoring stations, which precisely record the start and end times of extreme weather events. Impact area records define the geographic scope of extreme weather events. Geographic Information System (GIS) technology can be used to analyze meteorological satellite imagery and data from ground-based meteorological observation stations to determine the boundaries of extreme weather events.

[0019] The electrical component operating status data set is a data set built around the operating conditions of electrical components. The operating parameter records for electrical components during the corresponding period of extreme weather events provide a detailed record of the various operating parameters of the electrical components. These operating parameters include, but are not limited to, voltage, current, and temperature. For example, during a thunderstorm, voltage, current, and temperature sensors installed on a transformer at a substation monitor and record the transformer's operating parameters in real time. The voltage sensor converts the detected voltage signal into an electrical signal and transmits it to a data acquisition system for recording. The current sensor similarly records current data. The temperature sensor senses temperature changes on the transformer surface and converts this information into recordable data. Historical fault records record information about past electrical component faults, including the specific time and type of fault. These records can be obtained through the electrical component's fault monitoring system, which automatically records the relevant information when a fault occurs and stores it in a database. For example, if an electrical component experienced an overheating fault in a previous year, the fault monitoring system would record the specific date and time of the fault, as well as the fault type (overheating).

[0020] Step S200: performing feature extraction on the extreme weather data set to obtain an extreme weather feature set, and performing feature extraction on the electrical component operation status data set to obtain an electrical component operation feature set.

[0021] Feature extraction is the process of extracting essential features from raw data. Feature extraction for extreme weather data sets aims to extract key information representing the characteristics of extreme weather events from data containing records of the periods of occurrence and affected areas of different types of extreme weather events, thereby forming an extreme weather feature set. Feature extraction for electrical component operating status data sets involves extracting key information reflecting the operating status and fault characteristics of electrical components from records of their operating parameters and historical faults during the corresponding time periods, thereby forming an electrical component operating feature set.

[0022] As an implementation manner, step S200 specifically includes the following steps S210 to S250:

[0023] Step S210: Divide the occurrence period records in the extreme weather data set into time series to obtain multiple continuous weather period units, define the spatial scope of the impact area records in each weather period unit, and generate weather impact area features.

[0024] Time series segmentation of extreme weather data sets involves dividing the entire extreme weather period into multiple, continuous, smaller time periods. These smaller time periods are known as weather time units. For example, a prolonged rainstorm can be divided into multiple shorter time units at fixed intervals, with each time unit representing a time period. Time series segmentation can be implemented using a fixed time window partitioning method, which uses a fixed time window length as the starting time of the extreme weather event and then performs segmentation by sliding the window sequentially.

[0025] Defining the spatial scope of the impact area recorded within each weather time unit is crucial for determining the specific geographic extent of the extreme weather event within each weather time unit. Geographic Information System (GIS) technology can be used to precisely map the boundaries of the impact area, combining satellite imagery with data from ground-based meteorological observation stations. For example, within each weather time unit, the distribution of clouds in satellite imagery, combined with changes in meteorological elements monitored by ground-based meteorological stations, can be used to determine the boundaries of the extreme weather impact area. GIS technology can then be used to convert these boundaries into geographic coordinates, thereby defining the spatial extent of the impact area.

[0026] Generating weather impact area features involves further analyzing and processing the defined impact area to extract key information that characterizes the affected area. These features can include the impact area's spatial center location, coverage area, and electrical component density distribution. For example, the spatial center location is determined by calculating the geometric center coordinates of the impact area; the coverage area is determined by calculating the area of ​​the impact area; and electrical component density distribution features are generated by analyzing the distribution of electrical components within the impact area and counting the number of electrical components in different areas.

[0027] Step S211: parse the occurrence period records in the extreme weather data set, extract the start time mark and end time mark of each extreme weather event, calculate the time interval between the start time mark and the end time mark, and equally divide the time interval according to the preset time window length to obtain multiple continuous weather period units.

[0028] Parsing the occurrence period records in extreme weather data sets involves analyzing and processing the time information in the records to extract key time markers. Data parsing algorithms can be used to identify and extract the textual information within the occurrence period records. For example, if the occurrence period record format is "start time - end time," string segmentation can be used to extract the start and end times, which serve as the start and end time markers.

[0029] Calculating the time interval between the start and end timestamps is crucial for determining the duration of the extreme weather event. Using a time calculation algorithm, the start and end timestamps can be converted into timestamps, and the difference between the two timestamps can be calculated to determine the time interval. Equally dividing the time interval into pre-set time windows divides the duration of the extreme weather event into pre-set time windows. The pre-set time window length can be customized based on actual needs. For example, a shorter time window can be set, starting from the start time of the extreme weather event and dividing the time window into pre-set time windows until the end time.

[0030] Step S212: performing geographic information analysis on the impact area records corresponding to each weather period unit, extracting the boundary coordinate point set of the impact area, sorting the boundary coordinate point set in a clockwise direction, and generating a closed area boundary polygon.

[0031] Geographic information parsing of the impact area records corresponding to each weather time unit converts the impact area records into processable geographic coordinate data. The parsing capabilities of a geographic information system (GIS) can be used to process the impact area description. For example, if the impact area record is given as a text description, such as "around a certain area," geocoding technology can be used to convert this description into specific geographic coordinates.

[0032] Extracting the boundary coordinate point set of the impact area involves determining the boundary location of the impact area from the parsed geographic information. A boundary detection algorithm can be used to analyze the geographic coordinate data of the impact area and identify the coordinate points on the boundary. For example, an edge detection algorithm can be used to identify boundary points in the geographic coordinate data to form a boundary coordinate point set.

[0033] Sorting the set of boundary coordinate points in a clockwise direction is done to generate a closed region boundary polygon. A coordinate sorting algorithm can be used to sort the set of boundary coordinate points. For example, a reference point is selected, the angle between each boundary coordinate point and the reference point is calculated, and then the boundary coordinate points are sorted by the angle, so that they are arranged in a clockwise direction. Finally, the sorted boundary coordinate points are connected in sequence to form a closed region boundary polygon.

[0034] Step S213: Calculate the geometric center coordinates and area parameters of the region boundary polygon, determine the spatial center position of the affected region according to the geometric center coordinates, and determine the coverage size of the affected region according to the area parameters.

[0035] Calculating the geometric center coordinates of a region's boundary polygon involves processing the vertex coordinates of the region's boundary polygon using a geometric calculation algorithm. Using the polygon geometric center calculation method, the coordinates of all vertices of the region's boundary polygon are summed and then divided by the number of vertices to obtain the coordinates of the geometric center. These geometric center coordinates represent the spatial center of the affected region. Calculating the area parameter of a region's boundary polygon determines the coverage of the affected region. Using the polygon area calculation algorithm, the vertex coordinates of the region's boundary polygon are calculated. For example, by splitting the polygon into multiple triangles, calculating the area of ​​each triangle, and then summing the areas of all triangles, the area of ​​the polygon is obtained. This area parameter represents the coverage of the affected region.

[0036] Step S214: extracting geographic partition information within the regional boundary polygon, determining the electrical component distribution area within the extreme weather impact area, and generating electrical component density distribution characteristics within the area.

[0037] Extracting geographic zoning information within the region's boundary polygons analyzes and delineates the geographic structure within the impact area. The zoning functionality of a geographic information system (GIS) can be used to process the geographic information within the region's boundary polygons. For example, the impact area can be divided into different geographic zones based on factors such as topography and administrative divisions.

[0038] Determining the distribution of electrical components within the extreme weather impact zone involves analyzing the electrical component's location information and the extent of the impact zone to identify the electrical component's location within the impact zone. The electrical component's location coordinates can be compared with the region's boundary polygon to determine whether the component is within the impact zone. For example, if the electrical component's location coordinates are within the region's boundary polygon, the component is within the impact zone.

[0039] Generating a regional electrical component density distribution feature involves statistically analyzing the distribution of electrical components within the affected area. The affected area can be divided into multiple sub-areas, the number of electrical components in each sub-area is counted, and the density of electrical components is calculated based on the sub-area area. By analyzing and organizing the electrical component density of all sub-areas, a regional electrical component density distribution feature is generated.

[0040] Step S215: combining the spatial center position, coverage size, and electrical component density distribution characteristics to obtain weather impact area characteristics, wherein each weather period unit corresponds to a weather impact area characteristic.

[0041] Combining the spatial center location, coverage size, and electrical component density distribution characteristics integrates these three features to form a comprehensive weather impact area signature. Data fusion can be used to merge these three features. For example, the spatial center location, coverage size, and electrical component density distribution characteristics can be stored in a single data structure to form a record of weather impact area characteristics. Each weather time unit corresponds to a weather impact area signature, meaning that the impact area characteristics of extreme weather may vary within different weather time units. This approach allows for a more accurate description of the impact of extreme weather over different time periods.

[0042] Step S220: Analyze the intensity change trend of the weather impact area characteristics, extract the persistence characteristics and change characteristics of extreme weather in different weather time units, and combine the persistence characteristics and change characteristics to obtain an extreme weather feature set.

[0043] Analysis of the intensity trend of weather-affected regional characteristics involves analyzing how these characteristics change within different weather time periods to determine the intensity trend of extreme weather events. Time series analysis can be used to analyze various indicators of weather-affected regional characteristics. For example, analyzing the changing trends in coverage size within different weather time periods can determine whether extreme weather events are expanding or contracting.

[0044] Extracting the persistence and change characteristics of extreme weather within different weather time units is to separate the characteristics that reflect the persistence and change of extreme weather from the characteristics of the weather-affected area. Persistence characteristics can include the cumulative duration of extreme weather, while change characteristics can include the expansion or contraction trend of the affected area, changes in the density distribution of electrical components, etc. For example, by counting the duration of extreme weather within each weather time unit and calculating the sum of the durations of consecutive weather time units, the cumulative duration of extreme weather is obtained as the persistence characteristic. By analyzing the change ratio of the coverage size between adjacent weather time units, the expansion or contraction trend of the affected area is determined as the change characteristic.

[0045] Combining persistent and changing features to create an extreme weather feature set integrates these two types of feature information to form a set encompassing various extreme weather characteristics. Data combination can be used to store persistent and changing features in a single data structure to form an extreme weather feature set.

[0046] Step S221: Obtain the coverage size parameter in the weather impact area feature corresponding to each weather period unit, calculate the change ratio of the coverage size between adjacent weather period units, and generate a coverage change rate sequence.

[0047] Obtaining the coverage size parameter in the weather impact area feature corresponding to each weather period unit is to extract the coverage size indicator from the weather impact area feature. The coverage size parameter can be found in the data record of the weather impact area feature by means of data query. Calculating the change ratio of the coverage size between adjacent weather period units is to calculate the change ratio by comparing the coverage sizes of two adjacent weather period units. For example, the coverage size of the latter weather period unit is subtracted from the coverage size of the previous weather period unit, and then divided by the coverage size of the previous weather period unit to obtain the change ratio. Generating a coverage change rate sequence is to arrange the calculated change ratios in the order of the weather period units to form a sequence. The change ratios can be stored in an array or list using data storage to form a coverage change rate sequence.

[0048] Step S222: performing trend fitting on the coverage change rate sequence to determine the expansion trend direction or contraction trend direction of the extreme weather impact area, and using the expansion trend direction or contraction trend direction as the direction component of the change feature.

[0049] Trend fitting of a coverage change rate series involves using a fitting algorithm to identify the changing trend of the coverage change rate series. Linear fitting, nonlinear fitting, and other methods can be used to fit the coverage change rate series. For example, the linear fitting method finds a straight line that minimizes the error between the line and the data points in the coverage change rate series. Determining the expansion or contraction trend of the extreme weather-affected area involves using the fitted trend to determine whether the affected area is expanding or contracting. If the slope of the fitted trend line is positive, the affected area is expanding; if the slope is negative, the affected area is contracting.

[0050] Taking the expansion trend direction or the contraction trend direction as the direction component of the change feature is to take this direction information as an important component of the change feature. The direction information can be stored in the data structure of the change feature as the direction component.

[0051] Step S223: Count the duration of extreme weather in each weather period unit, calculate the sum of the durations of consecutive weather period units, obtain the cumulative duration of extreme weather, and use the cumulative duration as the time component of the duration feature.

[0052] To calculate the duration of extreme weather events within each weather time period, we calculate the start and end time of each weather time period to obtain the duration of the extreme weather events within that time period. We can use a time calculation algorithm to calculate the difference between two time points to obtain the duration.

[0053] To calculate the total duration of consecutive weather periods, add the durations of adjacent weather periods to get the cumulative duration of the extreme weather. For example, add the duration of the first weather period to the duration of the second weather period, then add the duration of the third weather period, and so on to get the cumulative duration.

[0054] Using the accumulated duration as the time component of the continuous feature is an important part of using this time information as the continuous feature. The accumulated duration can be stored in the data structure of the continuous feature as the time component.

[0055] Step S224: Analyze the changes in the electrical component density distribution characteristics in the weather-affected area characteristics in different weather time units, extract the peak position movement trajectory of the density distribution, and use the peak position movement trajectory as the spatial component of the change characteristics.

[0056] Analyzing the changes in electrical component density distribution characteristics within weather-affected regional features across different weather periods is a way to analyze the differences in electrical component density distribution characteristics across different time periods. By comparing electrical component density distribution characteristics across different weather periods, we can identify areas of variation. For example, we can analyze the changes in the maximum and minimum electrical component density values ​​across different time periods, as well as the changes in the shape of the density distribution.

[0057] Extracting the peak position trajectory of the density distribution involves finding the movement path of the density distribution peak position based on the changes in the electrical component density distribution characteristics. Using a peak detection algorithm, we can find the peak positions in the electrical component density distribution characteristics for each weather period unit and then record the coordinates of these peak positions. As the weather period unit changes, the coordinates of these peak positions are connected to form the peak position movement trajectory.

[0058] Taking the peak position movement trajectory as the spatial component of the change feature is an important component of taking this spatial information as the change feature. The peak position movement trajectory can be stored in the data structure of the change feature as the spatial component.

[0059] Step S225: combining the direction component and the spatial component to obtain a change feature, taking the time component as a continuous feature, arranging the change feature and the continuous feature in the time sequence of the weather period unit, and generating an extreme weather feature set.

[0060] Combining the directional and spatial components to obtain a change signature is to integrate the information of these two components to form a complete change signature. Data combination can be used to store the directional and spatial components in a data structure to form a change signature.

[0061] Taking the time component as a persistent feature is to take the time component information as the core content of the persistent feature. The time component can be stored in the data structure of the persistent feature to form a persistent feature.

[0062] Arranging the changing and persistent features in chronological order within the weather period units to generate an extreme weather feature set involves organizing these two types of feature information in chronological order to form a set containing various extreme weather features. Data sorting methods can be used to arrange the changing and persistent features in chronological order within the weather period units and store them in a data structure to form an extreme weather feature set.

[0063] Step S230: classify the operating parameter records in the electrical component operating status data set to obtain the state parameter sequence of the electrical component under different operating conditions, perform fault pattern matching on the historical fault records, and extract the fault type features corresponding to the state parameter sequence.

[0064] Classifying the operating parameter records in an electrical component operating status data set involves categorizing the records according to specific rules to determine the component's status under different operating conditions. Cluster analysis algorithms can be used to classify the operating parameter records. For example, operating parameter records can be divided into different categories based on the value ranges of operating parameters such as voltage, current, and temperature.

[0065] To obtain a state parameter sequence of an electrical component under different operating conditions, the classified operating parameter records are arranged in chronological order to form a state parameter sequence. Data sorting can be used to store the classified operating parameter records in chronological order in an array or list to form a state parameter sequence.

[0066] Fault pattern matching on historical fault records involves comparing the fault patterns in the records with the state parameter sequence to identify the fault type that corresponds to the state parameter sequence. Pattern matching algorithms can be used to analyze the historical fault records and the state parameter sequence. For example, by comparing the operating parameter characteristics at the time of the fault with those in the state parameter sequence, matching fault types can be identified.

[0067] Extracting fault type features corresponding to the state parameter sequence is to extract key feature information from the matched fault types. Fault types can be classified and summarized, and feature descriptions of the fault types can be extracted, such as the manifestation of the fault and possible causes.

[0068] Step S231: parsing the operating parameter records in the electrical component operating status data set, and extracting multi-dimensional operating parameters including voltage fluctuation records, current fluctuation records, and temperature change records.

[0069] Parsing operating parameter records within an electrical component operating status data set involves analyzing and processing the textual information within the operating parameter records to extract key operating parameters. Data parsing algorithms can be used to identify and process the format of the operating parameter records. For example, if the operating parameter records are stored in a table format, the parsing algorithm can extract the data from the table.

[0070] Extracting multidimensional operating parameters, including voltage fluctuation records, current fluctuation records, and temperature change records, involves identifying voltage, current, and temperature fluctuation records from parsed operating parameter records. Data filtering can be used to identify data related to voltage, current, and temperature within the operating parameter records. Voltage fluctuation records record the temporal changes in the voltage of electrical components; current fluctuation records record the temporal changes in the current; and temperature change records record the temporal changes in the temperature.

[0071] Step S232: Divide the multi-dimensional operating parameters into state intervals to obtain interval division results corresponding to each dimension.

[0072] For example, voltage fluctuation records can be divided into the first, second, and third fluctuation intervals; current fluctuation records can be divided into the first, second, and third load intervals; and temperature change records can be divided into the first, second, and third temperature intervals. State interval segmentation for multidimensional operating parameters involves dividing voltage, current, and temperature fluctuation records into different intervals according to specific rules. An interval segmentation algorithm can be used to segment records based on the value range of the operating parameters. For example, voltage fluctuation records can be segmented based on normal and abnormal voltage ranges.

[0073] Dividing voltage fluctuation records into the first, second, and third fluctuation intervals divides the voltage fluctuation records into different levels of fluctuation. The first fluctuation interval may indicate a range of relatively small voltage fluctuations; the second fluctuation interval indicates a range of moderate voltage fluctuations; and the third fluctuation interval indicates a range of relatively large voltage fluctuations.

[0074] Dividing the current fluctuation record into a first load interval, a second load interval, and a third load interval is to divide the current fluctuation record according to different load levels. The first load interval may represent a range of light current load; the second load interval represents a range of medium current load; and the third load interval represents a range of heavy current load.

[0075] Dividing the temperature change records into the first temperature interval, the second temperature interval, and the third temperature interval means dividing the temperature change records into different temperature ranges. The first temperature interval may represent a lower temperature range; the second temperature interval represents a medium temperature range; and the third temperature interval represents a higher temperature range.

[0076] Step S233: Classify the operating status of the electrical components according to the interval division result to obtain a combined status identifier including a voltage status label, a current status label, and a temperature status label, merge continuous operating parameter records with the same combined status identifier into a status segment, and generate a status parameter sequence.

[0077] Classifying the operating state of an electrical component based on the interval division results involves determining the operating state of the electrical component based on the interval division results for voltage, current, and temperature. For example, if the voltage is within a first fluctuation interval, the current is within a first load interval, and the temperature is within a first temperature interval, the operating state of the electrical component can be classified as one state.

[0078] The combined state identifier including the voltage state label, the current state label, and the temperature state label is obtained by combining the voltage, current, and temperature state labels to form a unique identifier. For example, the voltage state label, the current state label, and the temperature state label are connected by a symbol to form the combined state identifier.

[0079] Merging consecutive operating parameter records with the same combined state identifier into a state segment involves merging consecutive operating parameter records with the same combined state identifier to form a state segment. A data merging algorithm can be used to merge the operating parameter records. For example, consecutive operating parameter records with the same combined state identifier can be stored in a data structure to form a state segment.

[0080] Generating a state parameter sequence involves arranging the merged state segments in chronological order to form a state parameter sequence. Data sorting can be used to store the state segments in chronological order in an array or list to form a state parameter sequence.

[0081] Step S234: parse the historical fault records, extract the fault type description and the corresponding fault occurrence period when the fault occurs, time-match the fault occurrence period with the state segments in the state parameter sequence, and determine the state segment set corresponding to each fault type.

[0082] Parsing historical fault records involves analyzing and processing the textual information within these records to extract key fault information. Data parsing algorithms can be used to identify and process the format of these records. For example, if a historical fault record is provided as a textual description, a parsing algorithm can extract the fault type and the time period in which the fault occurred.

[0083] Extracting the fault type description and the corresponding fault occurrence time period from the parsed historical fault records involves finding the fault type description and the specific time of the fault occurrence. Data related to the fault type and fault occurrence time period can be found in the historical fault records through data filtering.

[0084] Temporally matching the fault occurrence period with the state segments in the state parameter sequence involves comparing the fault occurrence time with the time range of the state segments in the state parameter sequence to identify the state segments within the fault occurrence period. A time matching algorithm can be used to compare the fault occurrence period with the time range of the state segments. For example, this algorithm can determine whether the fault occurrence period is completely contained within the time range of a state segment or whether it overlaps with the time range of a state segment.

[0085] To determine the state segment set corresponding to each fault type, the state segments corresponding to each fault type are collected together based on the time matching results to form a state segment set. Data grouping can be used to group the state segments by fault type and store them in a data structure to form a state segment set.

[0086] Step S235: Statistically analyze the combined state identifiers in the state segment set, extract the most frequently occurring combined state identifier as the typical state feature of the fault type, and combine the typical state feature with the fault type description to obtain the fault type feature.

[0087] Statistical analysis of the combined state identifiers in the state segment set involves counting and analyzing the combined state identifiers in the state segment set to identify the most frequently occurring combined state identifier. A statistical analysis algorithm can be used to count the combined state identifiers. For example, the state segment set can be traversed, the number of occurrences of each combined state identifier can be counted, and the most frequently occurring combined state identifier can be identified.

[0088] Extracting the most frequently occurring combined state identifier as the typical state feature of the fault type means taking the most frequently occurring combined state identifier as the representative feature of the fault type. This combined state identifier can be stored in a data structure of the typical state feature.

[0089] Combining typical state features with fault type descriptions to generate fault type signatures involves integrating the typical state features and fault type descriptions to form a complete fault type signature. Data combination methods can be used to store the typical state features and fault type descriptions in a single data structure to form the fault type signature. In some implementations, the descriptions can be converted into vectors and then combined. A text vectorization algorithm can be used to convert the fault type description into a vector representation, which is then combined with the typical state features.

[0090] Step S240: performing association mapping on the state parameter sequence and the fault type feature to generate an electrical component operation feature set including a correspondence between the operation state and the fault type.

[0091] Correlation mapping between state parameter sequences and fault type characteristics involves matching the state segments in the state parameter sequence with the fault type characteristics to identify their corresponding relationships. Correlation mapping algorithms can be used to analyze the state parameter sequence and fault type characteristics. For example, matching fault types can be identified by comparing the combined state identifiers of the state segments with typical state characteristics in the fault type characteristics.

[0092] Generating an electrical component operating feature set containing the correspondence between operating states and fault types involves storing the correspondences obtained from the association mapping in a data structure to form a set containing the correspondences between operating states and fault types. Data storage can be used to store the correspondences in an array or list to form the electrical component operating feature set.

[0093] Step S241: performing a similarity comparison between each state segment in the state parameter sequence and the typical state feature in the fault type feature, and calculating a matching value between the combined state identifier of the state segment and the typical state feature.

[0094] Performing a similarity comparison between each state segment in the state parameter sequence and the typical state feature in the fault type feature is to compare the combined state identifier of the state segment and the typical state feature to find the degree of similarity between them. A similarity calculation algorithm can be used to calculate the combined state identifier and the typical state feature. For example, by comparing the individual state labels in the combined state identifier with the state labels in the typical state feature, their degree of match is calculated. Calculating the matching value between the combined state identifier of the state segment and the typical state feature is to convert the result of the similarity comparison into a numerical value used to represent the degree of match. A matching calculation algorithm can be used to calculate the matching value based on the result of the similarity comparison. For example, the matching value is obtained by dividing the number of matched state labels by the total number of state labels.

[0095] Step S242: When the matching value exceeds the preset threshold, an association relationship between the state segment and the corresponding fault type feature is established, and the state segment start time, state segment end time and fault type description in the association relationship are recorded.

[0096] When the match value exceeds the preset threshold, it is determined whether the match value meets the pre-set standard. The preset threshold is set based on specific application requirements and experience to determine whether to establish an association. If the match value exceeds the preset threshold, it indicates that there is a high degree of match between the status segment and the fault type characteristics, and an association can be established.

[0097] Establishing an association between the state segment and the corresponding fault type characteristics involves associating the state segment and the fault type characteristics to form an association record. A data association algorithm can be used to store the state segment and fault type characteristics in a data structure to form an association record. Recording the state segment start time, end time, and fault type description within the association record captures important information within the association. The state segment start time and end time can be obtained from the state segment data record; the fault type description can be obtained from the fault type characteristics. Storing this information in the association record facilitates subsequent querying and analysis.

[0098] Step S243: Analyze the parameter change trend of the associated state segment, extract the change amplitude of the voltage fluctuation record, the change frequency of the current fluctuation record and the rising rate of the temperature change record in the state segment, and combine these parameters to obtain the state change characteristics.

[0099] Parameter trend analysis for associated state segments involves analyzing the changing trends of operating parameters such as voltage, current, and temperature within the segment. Trend analysis algorithms can be used to analyze the operating parameter records within the segment. For example, analyzing the changing trends of voltage fluctuation records can determine whether the voltage is rising or falling, as well as the magnitude of the change.

[0100] Extracting the amplitude of voltage fluctuations, the frequency of current fluctuations, and the rate of rise of temperature fluctuations within a state segment is done to extract key parameters from the operating parameter records within the state segment. Parameter extraction algorithms can be used to process the operating parameter records. For example, the amplitude of voltage fluctuations can be calculated by calculating the difference between the maximum and minimum values ​​in the voltage fluctuation record; the frequency of current fluctuations can be obtained by counting the number of fluctuations in the current fluctuation record; and the rate of rise of temperature can be obtained by calculating the ratio of the temperature rise to time in the temperature change record.

[0101] Combining these parameters to generate a state change signature involves combining the extracted voltage fluctuation amplitude, current fluctuation frequency, and temperature rise rate to form a state change signature. Data combination can be used to store these parameters in a data structure to form a state change signature.

[0102] Step S244: Integrate the fault type description, status segment start time, status segment end time and status change characteristics in the association relationship to generate an association mapping table containing the corresponding relationship between the operating status and the fault type.

[0103] Integrating the fault type description, status segment start time, status segment end time, and status change characteristics in the association relationship combines this information to form a complete association record. A data integration algorithm can be used to store the fault type description, status segment start time, status segment end time, and status change characteristics in a data structure to form an association record.

[0104] To generate a correlation mapping table containing the correspondence between operating status and fault type, all correlation records are arranged according to certain rules to form a correlation mapping table. Data sorting can be used to sort the correlation records by the start time of the status segment and store them in a table or data structure to form the correlation mapping table.

[0105] Step S245: Arrange the association mapping table according to the time sequence of the state segments to form an electrical component operation feature set, wherein each association mapping table entry corresponds to a feature unit in the electrical component operation feature set.

[0106] Arranging the association mapping table in the time sequence of the state segments is to sort the entries in the association mapping table according to the start time of the state segments so that the entries in the association mapping table are arranged in the time sequence. A data sorting algorithm can be used to sort the association mapping table.

[0107] To form the electrical component operational feature set, the sorted association mapping table is stored in a data structure to form a set containing the electrical component operational features. Each association mapping table entry corresponds to a feature unit in the electrical component operational feature set, meaning that each entry in the association mapping table represents an operational feature of an electrical component.

[0108] Step S250: performing a dimensional consistency check on the extreme weather feature set and the electrical component operation feature set to ensure that the two are compatible in terms of time and space dimensions.

[0109] Performing a dimensional consistency check on the extreme weather feature set and the electrical component operation feature set examines the dimensional information of the two feature sets to ensure consistency in both temporal and spatial dimensions. A dimensionality check algorithm can be used to compare the dimensional information of the extreme weather feature set and the electrical component operation feature set.

[0110] In the time dimension, check whether the time ranges of the weather period unit in the extreme weather feature set and the state segment in the electrical component operation feature set match. For example, determine whether the start and end times of the weather period unit and the state segment overlap or are completely consistent.

[0111] In the spatial dimension, check whether the impact area in the extreme weather feature set matches the area where the electrical components are located in the electrical component operation feature set. For example, determine whether the impact area includes the area where the electrical components are located, or whether the two areas overlap.

[0112] Ensuring the matching of the time and space dimensions is crucial to ensuring that subsequent correlation and fusion operations can be performed accurately. If the time and space dimensions do not match, the correlation and fusion results may be inaccurate.

[0113] Step S300: Correlating and fusing the extreme weather feature set and the electrical component operation feature set to generate a correlation feature sequence.

[0114] Associating and fusing the extreme weather feature set and the electrical component operation feature set involves integrating the information in these two feature sets, identifying the correlation between them, and fusing the relevant information together. An association fusion algorithm can be used to process the extreme weather feature set and the electrical component operation feature set.

[0115] Generating a correlation feature sequence involves arranging the fused information in chronological order to form a sequence. Data sorting can be used to store the fused information in an array or list in chronological order to form a correlation feature sequence.

[0116] Step S310: Time-align each weather period unit in the extreme weather feature set and each state segment in the electrical component operation feature set, and determine the correspondence between weather period units and state segments with overlapping time intervals.

[0117] Time alignment of each weather period unit in the extreme weather feature set and each state segment in the electrical component operation feature set involves comparing the time ranges of the weather period unit and the state segment to identify the temporal relationship between them. A time alignment algorithm can be used to process the time ranges of the weather period unit and the state segment.

[0118] Determining the correspondence between weather time units and status segments with overlapping time intervals involves finding weather time units and status segments with overlapping time intervals from the time alignment results and establishing a correspondence between them. A correspondence determination algorithm can be used to analyze the time alignment results. For example, it can determine whether the time ranges of weather time units and status segments overlap, and if so, establish a correspondence between them.

[0119] Step S311: extract the start timestamp and end timestamp of each weather period unit in the extreme weather feature set to generate a weather period timeline, extract the start timestamp and end timestamp of each state segment in the electrical component operation feature set to generate a state segment timeline.

[0120] Extracting the start and end timestamps of each weather period unit in the extreme weather feature set involves finding the start and end times of each weather period unit from the extreme weather feature set and converting them into timestamps. A timestamp extraction algorithm can be used to process the time information in the extreme weather feature set. For example, the start and end times of the weather period units can be converted into a unified time format and then converted into timestamps.

[0121] Generating a weather period timeline involves arranging the start and end timestamps of each extracted weather period unit in chronological order to form a timeline. Data sorting can be used to store the timestamps in chronological order in an array or list to form the weather period timeline.

[0122] Extracting the start and end timestamps of each state segment in the electrical component operation feature set involves finding the start and end times of each state segment from the electrical component operation feature set and converting them into timestamps. A timestamp extraction algorithm can be used to process the time information in the electrical component operation feature set. For example, the start and end times of the state segments can be converted into a unified time format and then converted into timestamps.

[0123] Generating a state segment timeline involves arranging the start and end timestamps of each extracted state segment in chronological order to form a timeline. Data sorting can be used to store the timestamps in chronological order in an array or list to form the state segment timeline.

[0124] Step S312: unify the time bases of the weather period time axis and the status period time axis to ensure that they use the same time measurement standard.

[0125] Unifying the time bases of the weather period timeline and the status period timeline is to adjust the time measurement standards of the two timelines to make them consistent. A time base unification algorithm can be used to process the time information of the weather period timeline and the status period timeline.

[0126] Ensuring that both use the same time standard ensures that subsequent time alignment operations can be performed accurately. Inconsistent time standards may result in inaccurate time alignment results. For example, if the weather period timeline uses Beijing Time, while the status period timeline uses Greenwich Mean Time, they must be aligned to the same time standard.

[0127] Step S313: traverse each weather period unit on the weather period timeline, search for state segments on the state segment timeline that overlap with the time interval of the weather period unit, and calculate the ratio of the length of the overlapping time interval to the total time length of the weather period unit.

[0128] Traversing each weather period unit on the weather period time axis is to process each weather period unit on the weather period time axis in sequence. A loop traversal algorithm can be used to traverse the weather period time axis.

[0129] Searching for state segments on the state segment timeline that overlap with the time interval of the weather period unit involves finding state segments on the state segment timeline that overlap with the time range of the current weather period unit. A time overlap search algorithm can be used to search the state segment timeline. For example, it can be determined whether the start and end times of the state segments fall within the time range of the weather period unit, or whether there is partial overlap.

[0130] To calculate the ratio of the length of the overlapping time interval to the total length of the weather period unit, divide the length of the overlapping time interval by the total length of the weather period unit to obtain a ratio. A ratio calculation algorithm can be used to calculate the length of the overlapping time interval and the total length of the weather period unit.

[0131] Step S314: When the ratio exceeds a preset overlap threshold, it is determined that there is a corresponding relationship between the weather period unit and the state segment, and the weather period unit identifier and the state segment identifier in the corresponding relationship are recorded.

[0132] When the ratio exceeds the preset overlap threshold, the calculated ratio is determined to meet the pre-set criteria. The preset overlap threshold is set based on specific application requirements and experience to determine whether a correspondence is established. If the ratio exceeds the preset overlap threshold, it indicates a high degree of temporal overlap between the weather period unit and the status segment, and a correspondence can be determined between them.

[0133] Determining the correspondence between the weather period unit and the state segment is to associate the weather period unit and the state segment to form a correspondence record. A data association algorithm can be used to store the weather period unit and the state segment in a data structure to form a correspondence record.

[0134] Recording the weather period unit identifier and status segment identifier in the corresponding relationship is to record important information in the corresponding relationship. The weather period unit identifier can be obtained from the weather period unit data record; the status segment identifier can be obtained from the status segment data record. Storing this information in the corresponding relationship record facilitates subsequent query and analysis.

[0135] Step S315: De-duplicate all determined correspondences, delete duplicate correspondence records, and generate a weather-state correspondence table.

[0136] Deduplication of all identified correspondences involves checking all correspondence records, identifying and deleting duplicate records. A deduplication algorithm can be used to process the correspondence records. For example, the weather period unit identifier and state segment identifier in the correspondence records can be compared. If a duplicate record is found, one of the records is deleted.

[0137] Deleting duplicate correspondence records means deleting duplicate records found in the deduplication process from the correspondence record set. A data deletion algorithm can be used to process the correspondence record set.

[0138] Generating a weather-state correspondence table involves arranging the duplicated correspondence records according to a specific rule to form a correspondence table. Data sorting methods can be used to sort the correspondence records by weather time unit identifiers and store them in a table or data structure to form a weather-state correspondence table.

[0139] Step S320: According to the corresponding relationship, the extreme weather characteristics of the weather period unit are matched with the electrical component operation characteristics of the state segment, and the persistence characteristics and change characteristics in the extreme weather characteristics and the state change characteristics and fault type characteristics in the electrical component operation characteristics are extracted.

[0140] According to the correspondence, the weather time unit and the state segment are associated according to the correspondence in the weather-state correspondence table. The corresponding weather time unit and state segment can be found in the weather-state correspondence table by using a data query method.

[0141] Matching the extreme weather characteristics of a weather period unit with the electrical component operating characteristics of a status segment involves comparing identical or related fields in the extreme weather characteristics and the electrical component operating characteristics to identify matching parts. A feature field matching algorithm can be used to process the extreme weather characteristics and the electrical component operating characteristics. For example, a field in the extreme weather characteristics can be compared with a field in the electrical component operating characteristics to determine whether they are identical or related.

[0142] Extracting persistent and changing features from extreme weather characteristics, and state change and fault type features from electrical component operating characteristics, is the process of identifying key feature information from the matched features. Feature extraction algorithms can be used to process the matched features. For example, persistent and changing features can be extracted from extreme weather characteristics, and state change and fault type features can be extracted from electrical component operating characteristics.

[0143] Step S330: Expand the feature dimension of the matched feature field, calculate the ratio of the time component of the continuous feature to the duration of the state segment to obtain the time correlation coefficient, calculate the distance between the spatial component of the changing feature and the electrical component position information corresponding to the state segment to obtain the spatial correlation coefficient.

[0144] Feature dimension expansion of matched feature fields further processes the matched features to increase their dimension. Feature dimension expansion algorithms can be used to process the matched features. For example, a new feature dimension can be derived by calculating the ratio of the time component of a duration feature to the duration of a state segment.

[0145] The temporal correlation coefficient is calculated by calculating the ratio of the time component of the persistent feature to the duration of the state segment. This coefficient represents the degree of temporal correlation by dividing the time component of the persistent feature by the duration of the state segment. A ratio calculation algorithm can be used to calculate the time component of the persistent feature and the duration of the state segment.

[0146] The spatial correlation coefficient is obtained by calculating the distance between the spatial component of the change feature and the electrical component position information corresponding to the state segment. This is achieved by comparing the spatial component of the change feature with the electrical component position information, calculating the distance between them, and obtaining a coefficient representing the degree of spatial correlation. A distance calculation algorithm can be used to calculate the spatial component of the change feature and the electrical component position information.

[0147] Step S340: Using the time correlation coefficient and the space correlation coefficient as fusion weights, perform weighted fusion on the extreme weather characteristics and the electrical component operation characteristics to generate preliminary fusion characteristics, wherein the fusion weight of the electrical component operation characteristics is the difference between 1 and the comprehensive fusion weight, and the comprehensive fusion weight is determined by the weighted sum of the time correlation coefficient and the space correlation coefficient.

[0148] Using the temporal and spatial correlation coefficients as fusion weights means using these two coefficients as weighted fusion weights to adjust the proportion of extreme weather characteristics and electrical component operating characteristics in the fusion process. A weight setting algorithm can be used to use the temporal and spatial correlation coefficients as weights.

[0149] Weighted fusion of extreme weather features and electrical component operation features combines them according to the fusion weights to produce a fused feature. A weighted fusion algorithm can be used to process these features. For example, the extreme weather features can be multiplied by the time correlation coefficient, and the electrical component operation features can be multiplied by the fusion weight of the electrical component operation features. The two can then be added together to produce the fused feature. Generating a preliminary fused feature involves further processing the weighted fused features to form a complete feature. A feature generation algorithm can be used to process the weighted fused features.

[0150] The fusion weight of the electrical component operating characteristics is calculated by subtracting the comprehensive fusion weight from 1. The comprehensive fusion weight is determined by the weighted sum of the temporal correlation coefficient and the spatial correlation coefficient. A weight calculation algorithm can be used to calculate the comprehensive fusion weight and the fusion weight of the electrical component operating characteristics.

[0151] Step S341: converting the persistence features and change features in the extreme weather features and the state change features and fault type features in the electrical component operation features into numerical feature vectors so that all feature vectors have the same dimension.

[0152] Converting the persistence and variation characteristics of extreme weather events, as well as the state change and fault type characteristics of electrical component operation, into numerical feature vectors converts these feature information into vector representations for ease of subsequent calculation and processing. Feature vector conversion algorithms can be used to process these features. For example, the persistence, variation, state change, and fault type characteristics can be numerically represented and then combined into a vector.

[0153] Ensuring that all eigenvectors have the same dimensions ensures efficient computation of the eigenvectors in subsequent weighted fusion operations. A dimensionality unification algorithm can be used to adjust the dimensionality of the eigenvectors. For example, eigenvectors with smaller dimensions can be padded to make them the same dimension as the other eigenvectors.

[0154] Step S342: Calculate the weighted sum of the temporal correlation coefficient and the spatial correlation coefficient to obtain a comprehensive fusion weight, wherein the weight ratio of the temporal correlation coefficient and the spatial correlation coefficient is determined according to a preset rule.

[0155] To calculate the weighted sum of the temporal and spatial correlation coefficients, multiply the temporal correlation coefficient by its corresponding weight, multiply the spatial correlation coefficient by its corresponding weight, and then add the two together to obtain the comprehensive fusion weight. A weighted sum calculation algorithm can be used to calculate the temporal and spatial correlation coefficients and their weights.

[0156] The weighting of the temporal and spatial correlation coefficients is determined based on pre-set rules. These rules can be set based on specific application requirements and experience to determine the weighting of the temporal and spatial correlation coefficients in the overall fusion weighting. For example, the weighting can be set based on the degree to which temporal and spatial factors influence electrical component failures.

[0157] Step S343: Multiply the digitized feature vector of the extreme weather characteristics by the comprehensive fusion weight to obtain a weighted extreme weather feature vector, and multiply the digitized feature vector of the electrical component operation characteristics by the difference between 1 and the comprehensive fusion weight to obtain a weighted electrical component operation feature vector.

[0158] Multiplying the numerical feature vector of extreme weather characteristics by the comprehensive fusion weights is to multiply each element in the numerical feature vector of extreme weather characteristics by the comprehensive fusion weights to obtain a weighted extreme weather feature vector. A vector multiplication algorithm can be used to calculate the numerical feature vector of extreme weather characteristics and the comprehensive fusion weights.

[0159] Multiplying the numerical feature vector of the electrical component's operational characteristics by the difference between 1 and the comprehensive fusion weight is to multiply each element of the numerical feature vector of the electrical component's operational characteristics by the difference between 1 and the comprehensive fusion weight, thereby obtaining a weighted electrical component operational feature vector. A vector multiplication algorithm can be used to calculate the difference between the numerical feature vector of the electrical component's operational characteristics and 1 and the comprehensive fusion weight.

[0160] Step S344: Add the weighted extreme weather feature vector and the weighted electrical component operation feature vector element by element to obtain a preliminary fusion feature vector.

[0161] Element-by-element addition of the weighted extreme weather feature vector and the weighted electrical component operation feature vector adds the corresponding elements in these two vectors to produce a new vector. A vector addition algorithm can be used to calculate the weighted extreme weather feature vector and the weighted electrical component operation feature vector. For example, the first element of the weighted extreme weather feature vector is added to the first element of the weighted electrical component operation feature vector to produce the first element of the preliminary fused feature vector. This can be repeated in this manner to produce the entire preliminary fused feature vector.

[0162] Step S345: normalize the initial fusion feature vector, map each feature value into a preset numerical range, and generate a preliminary fusion feature.

[0163] Normalizing the initial fused feature vector involves processing each eigenvalue in the initial fused feature vector so that they have the same scale and range. A feature normalization algorithm can be used to process the initial fused feature vector. For example, the mean of each eigenvalue in the initial fused feature vector can be subtracted from the mean, and then divided by the standard deviation to obtain the normalized eigenvalue.

[0164] Mapping each eigenvalue to a preset numerical range involves further processing the normalized eigenvalues ​​to bring them within the preset numerical range. A numerical mapping algorithm can be used to process the normalized eigenvalues. For example, the normalized eigenvalues ​​can be multiplied by a scaling factor and then an offset added to bring them within the preset numerical range.

[0165] Generating preliminary fusion features is to combine the mapped feature values ​​into a complete feature. Feature generation algorithms can be used to process the mapped feature values.

[0166] Step S350: Redundant features are eliminated from the preliminary fusion features, repeated feature fields and feature fields with contributions lower than a preset threshold are deleted to obtain key fusion features, and the key fusion features are arranged in chronological order to generate a correlation feature sequence.

[0167] Redundant feature removal is performed on the preliminary fused features. This involves examining the feature fields in the preliminary fused features, identifying duplicate fields and fields with low contributions, and removing them. Redundant feature removal algorithms can be used to process the preliminary fused features. For example, the feature fields in the preliminary fused features can be compared. If any duplicate fields are found, one of them is removed. The contribution of each feature field to the final result is calculated, and if the contribution falls below a preset threshold, the field is removed.

[0168] Deleting duplicate feature fields and feature fields with contributions below a preset threshold removes duplicate fields and fields with low contributions found during the redundant feature elimination process from the initial fusion features. A data deletion algorithm can be used to process the initial fusion features.

[0169] The key fusion feature is obtained by using the preliminary fusion feature after removing redundant features as the key fusion feature. A feature extraction algorithm can be used to process the preliminary fusion feature after removing redundant features.

[0170] Arranging the key fusion features in chronological order to generate a related feature sequence is to sort the key fusion features in chronological order to form a sequence. The key fusion features can be stored in an array or list in chronological order using a data sorting method to form a related feature sequence.

[0171] Step S400: calling a pre-trained fault prediction model to perform fault risk prediction on the associated feature sequence, and generating a fault probability distribution sequence of the electrical component.

[0172] Calling a pre-trained fault prediction model to predict fault risk based on a correlation feature sequence involves inputting the correlation feature sequence into the pre-trained fault prediction model and leveraging the model's predictive capabilities to predict the fault risk of electrical components. Pre-trained fault prediction models are trained on large amounts of data and possess a certain level of predictive capability. A model call algorithm can be used to input the correlation feature sequence into the fault prediction model.

[0173] Generating a fault probability distribution sequence for electrical components involves arranging the fault probability results output by the fault prediction model in chronological order to form a sequence. Data sorting can be used to store the fault probability results in an array or list in chronological order to form a fault probability distribution sequence.

[0174] Step S410: inputting the associated feature sequence into the feature input layer of the fault prediction model, segmenting the feature sequence in the time dimension, and obtaining a plurality of feature subsequences of fixed length.

[0175] Inputting the associated feature sequence into the feature input layer of the fault prediction model is to pass the associated feature sequence to the feature input layer of the fault prediction model so that it can process the feature sequence. A data input algorithm can be used to input the associated feature sequence into the feature input layer.

[0176] Segmenting a feature sequence along the time dimension involves dividing the associated feature sequence into multiple feature subsequences of fixed lengths in chronological order. A time segmentation algorithm can be used to process the associated feature sequence. For example, a fixed time length can be set as the segmentation window. Starting from the start time of the associated feature sequence, the sequence is segmented sequentially according to this window length to obtain multiple feature subsequences.

[0177] To obtain multiple fixed-length feature subsequences, the segmented feature sequences are stored in a data structure to form multiple fixed-length feature subsequences. The feature subsequences can be stored in an array or list using a data storage method.

[0178] Step S420: Input each feature subsequence into the feature extraction layer of the fault prediction model, perform nonlinear transformation on the feature subsequence through a multi-layer perceptron, and extract high-level abstract features.

[0179] Inputting each feature subsequence into the feature extraction layer of the fault prediction model sequentially passes the segmented feature subsequences to the feature extraction layer of the fault prediction model so that the feature subsequences can be processed. A data input algorithm can be used to input the feature subsequences into the feature extraction layer.

[0180] Using a multilayer perceptron to perform nonlinear transformations on feature subsequences exploits the nonlinear properties of the multilayer perceptron to process the feature subsequences and extract high-level abstract features. The multilayer perceptron is an artificial neural network consisting of an input layer, hidden layers, and an output layer. The feature subsequence is first input to the input layer, which then passes it to the hidden layers. Neurons in the hidden layers perform a weighted summation of the input feature subsequences and perform a nonlinear transformation using an activation function to generate the transformed features. This process is repeated sequentially across multiple hidden layers of the multilayer perceptron, extracting increasingly higher-level abstract features. Finally, the output layer outputs the features after these multiple transformations as high-level abstract features.

[0181] Extracting high-level abstract features is to obtain representative feature information from the output of the multilayer perceptron. These high-level abstract features can better reflect the fault characteristics of electrical components and help improve the accuracy of fault prediction.

[0182] Step S421: Convert the feature subsequence into a two-dimensional feature matrix, where the row dimension corresponds to the time step of the feature subsequence and the column dimension corresponds to the feature dimension.

[0183] Converting a feature subsequence into a two-dimensional feature matrix involves arranging the feature information in the feature subsequence into a two-dimensional matrix according to certain rules. Matrix conversion algorithms can be used to process the feature subsequence. For example, the feature information at each time step in the feature subsequence can be used as a row in the matrix, and the different feature dimensions can be used as columns to form a two-dimensional feature matrix.

[0184] The row dimension corresponds to the time step of the feature subsequence, meaning that each row of the two-dimensional feature matrix represents the feature information of the feature subsequence at one time step. The column dimension corresponds to the feature dimension, meaning that each column of the two-dimensional feature matrix represents a feature dimension.

[0185] Step S422: Input the two-dimensional feature matrix into the first hidden layer of the multilayer perceptron, perform a linear transformation on the two-dimensional feature matrix using the weight matrix of the first hidden layer, and obtain the first hidden layer feature matrix.

[0186] Inputting the two-dimensional feature matrix into the first hidden layer of the multilayer perceptron involves passing the transformed two-dimensional feature matrix to the first hidden layer, enabling it to process the feature matrix. A data input algorithm can be used to input the two-dimensional feature matrix into the first hidden layer. Linearly transforming the two-dimensional feature matrix using the weight matrix of the first hidden layer involves performing a matrix multiplication operation on the two-dimensional feature matrix and the weight matrix of the first hidden layer to produce a new matrix. The weight matrix of the first hidden layer is learned during the training process of the multilayer perceptron and is used to adjust the weights of the input features. Matrix multiplication multiplies each element in the two-dimensional feature matrix by the corresponding element in the weight matrix, and then adds the results to produce a new matrix element.

[0187] The first hidden layer feature matrix is ​​obtained by using the result after the linear transformation as the first hidden layer feature matrix. The result after the linear transformation can be processed using a matrix generation algorithm.

[0188] Step S423: performing nonlinear activation on the first hidden layer feature matrix to enhance the nonlinear expression capability of the features, and obtaining the activated first hidden layer feature matrix.

[0189] Nonlinear activation of the first hidden layer feature matrix enables the multilayer perceptron to learn and express more complex patterns and feature relationships. While linear transformations can perform weighted combinations of input features, they are insufficient for complex nonlinear problems. Therefore, the first hidden layer feature matrix must be processed using a nonlinear activation function. There are many nonlinear activation functions, such as the Sigmoid function and the ReLU function. For example, the ReLU function transforms each element in the first hidden layer feature matrix. If the value of an element is greater than zero, it remains unchanged; if the value is less than or equal to zero, it is set to zero. By doing this, the ReLU function introduces nonlinearity, enhancing the expressive power of features and enabling the multilayer perceptron to learn more complex feature patterns.

[0190] The first hidden layer feature matrix is ​​input into the ReLU activation function, and the ReLU function's rules are applied to each element in the matrix. For example, if an element in the first hidden layer feature matrix has a value of -0.5, after processing it with the ReLU function, its value becomes 0; if another element has a value of 2, its value remains unchanged at 2. After processing all elements in the first hidden layer feature matrix, the activated first hidden layer feature matrix is ​​obtained. Compared to the original first hidden layer feature matrix, this activated first hidden layer feature matrix has stronger nonlinear expression capabilities and can better reflect the complex information in the input features.

[0191] Step S424: Input the activated first hidden layer feature matrix into the second hidden layer of the multilayer perceptron, perform a linear transformation through the weight matrix of the second hidden layer, and obtain the second hidden layer feature matrix.

[0192] The activated first hidden layer feature matrix is ​​input into the second hidden layer of the multilayer perceptron. This process is a continuation of the multilayer perceptron's further processing of feature information. The activated first hidden layer feature matrix carries the feature information processed by nonlinear activation and passes it to the second hidden layer for further feature extraction.

[0193] The weight matrix of the second hidden layer is learned during the training phase of the multilayer perceptron. It is used to adjust the weights of the input features to extract higher-level abstract features. A linear transformation is achieved by performing matrix multiplication on the activated first hidden layer feature matrix and the second hidden layer weight matrix. During matrix multiplication, each row element of the activated first hidden layer feature matrix is ​​multiplied by the corresponding column element of the second hidden layer weight matrix. The products are then summed to form an element in the new matrix.

[0194] For example, the first row of the activated first hidden layer feature matrix is ​​multiplied by the first column of the second hidden layer weight matrix, and the sum is calculated to obtain the first row and first column of the second hidden layer feature matrix. This operation is performed on each row of the activated first hidden layer feature matrix and each column of the second hidden layer weight matrix, ultimately resulting in the second hidden layer feature matrix. This second hidden layer feature matrix contains feature information that has undergone further linear transformation and has a higher level of abstraction than the activated first hidden layer feature matrix.

[0195] Step S425: Normalize the second hidden layer feature matrix to speed up the model convergence and prevent overfitting, and obtain a standardized second hidden layer feature matrix.

[0196] Normalizing the second hidden layer feature matrix helps improve the training efficiency and generalization ability of the multilayer perceptron. The main purpose of normalization is to adjust the elements in the second hidden layer feature matrix to an appropriate range so that different feature dimensions have similar scales, thereby accelerating the convergence of the model and preventing overfitting.

[0197] An example of a normalization method is Batch Normalization. First, the mean and variance of the second hidden layer feature matrix along each dimension are calculated. For a batch of input data, the mean and variance of all samples along each dimension are calculated. Then, the mean of each dimension is subtracted from each element in the second hidden layer feature matrix, and the result is divided by the standard deviation of that dimension to obtain the normalized element.

[0198] To ensure that the normalization operation does not destroy the feature representation of the original data, two learnable parameters are introduced: a scaling factor and an offset. The normalized elements are multiplied by the scaling factor and then added to the offset to obtain the final normalized result. This process yields the normalized second hidden layer feature matrix. The elements in this matrix have similar scales, which enables faster convergence of the multilayer perceptron during training, reduces the risk of overfitting, and improves the model's generalization ability.

[0199] Step S426: Perform global average pooling on the normalized second hidden layer feature matrix to convert the two-dimensional feature matrix into a one-dimensional feature vector as a high-level abstract feature.

[0200] Global average pooling is performed on the normalized second hidden layer feature matrix to convert the two-dimensional feature matrix into a one-dimensional feature vector, thereby obtaining high-level abstract features suitable for subsequent processing. The core idea of ​​global average pooling is to average each feature channel (i.e., each column) of the normalized second hidden layer feature matrix.

[0201] Specifically, for each column of the normalized second hidden layer feature matrix, all elements in that column are summed and then divided by the number of elements to obtain the column's average value. Arranging the average values ​​of all columns in sequence yields a one-dimensional feature vector. For example, if the normalized second hidden layer feature matrix has 10 columns, then after global average pooling, a one-dimensional feature vector of length 10 is obtained.

[0202] Global average pooling not only reduces the feature dimension and data complexity, but also preserves the key information in the feature matrix, generating representative, high-level abstract features. This one-dimensional feature vector more concisely represents the key information of the input feature subsequence, providing a better foundation for subsequent time series modeling and fault probability prediction.

[0203] Step S430: Input the high-level abstract features into the time series modeling layer of the fault prediction model, perform time series dependency modeling on the high-level abstract features through a recurrent neural network, extract the time correlation information between different feature subsequences, and obtain features processed by time series modeling.

[0204] High-level abstract features are input into the time series modeling layer of the fault prediction model. While these high-level abstract features extract key information about feature subsequences, they do not yet account for the temporal correlations between different feature subsequences. Time is crucial in electrical component fault prediction, as causal relationships or changing trends may exist between feature subsequences at different times. Therefore, a recurrent neural network is needed to model the temporal dependencies of these high-level abstract features.

[0205] Recurrent neural networks (RNNs) are neural networks specifically designed for processing sequential data. They use the hidden state of the previous moment and the input of the current moment to update the hidden state of the current moment, thereby capturing the temporal dependencies in sequential data. In this embodiment, multiple high-level abstract features are arranged in chronological order to form a temporal feature sequence, which is then input into the recurrent neural network.

[0206] Step S431: Arrange multiple high-level abstract features in chronological order to form a temporal feature sequence, wherein each element corresponds to a high-level abstract feature of a feature subsequence.

[0207] Arranging multiple high-level features in chronological order allows the recurrent neural network to process these features in chronological order. The corresponding high-level features can be arranged in order based on the time steps of the feature subsequences. For example, the high-level feature corresponding to the first feature subsequence is ranked first in the sequence, the high-level feature corresponding to the second feature subsequence is ranked second, and so on.

[0208] Each element in the resulting temporal feature sequence is a high-level abstract feature, representing the key information of a feature subsequence. This arrangement allows the recurrent neural network to process these features in chronological order, thereby better capturing the temporal correlations between different feature subsequences.

[0209] Step S432: Input the temporal feature sequence into the input gate of the recurrent neural network, calculate the update weight of the feature at each moment through the activation function of the input gate, and determine the feature information to be updated.

[0210] The time series feature sequence is fed into the input gate of the recurrent neural network. The input gate is a key component of the recurrent neural network. Its function is to control whether the current input information is updated to the cell state. The activation function of the input gate is usually the Sigmoid function, whose output value is between 0 and 1.

[0211] When a time series feature sequence is fed into the input gate, it calculates the update weight for each feature at each moment using an activation function based on the current input feature and the hidden state at the previous moment. This update weight represents the importance of the current input feature to the cell state update. For example, if the update weight is close to 1, the current input feature is very important and needs to be included in the cell state update; if the update weight is close to 0, the current input feature has little impact on the cell state update and can be ignored.

[0212] By calculating the update weight, the feature information to be updated is determined. The current input feature is multiplied by the update weight to obtain the feature information to be updated. This information will be used for subsequent cell state updates.

[0213] Step S433: Input the temporal feature sequence into the forget gate of the recurrent neural network, calculate the forget weight of the feature at each moment through the activation function of the forget gate, and determine the historical feature information to be forgotten.

[0214] The time series feature sequence is input into the forget gate of the recurrent neural network. The forget gate is also a key component of the recurrent neural network. Its main function is to control which historical information in the cell state needs to be forgotten. The forget gate also uses the Sigmoid function as the activation function.

[0215] When a time series feature sequence is input, the forget gate calculates the forget weight for each feature at each moment using an activation function based on the current input feature and the hidden state at the previous moment. The forget weight ranges from 0 to 1 and indicates the degree to which the historical information in the cell state has been forgotten. For example, if the forget weight is close to 0, a significant amount of the historical information in the cell state needs to be forgotten; if the forget weight is close to 1, the historical information in the cell state can be largely retained.

[0216] The historical feature information to be forgotten is determined based on the forgetting weight. The historical information in the cell state is multiplied by the forgetting weight to obtain the historical information after forgetting. This information will be used in subsequent cell state updates.

[0217] Step S434: Input the temporal feature sequence into the candidate memory gate of the recurrent neural network, generate candidate memory features through the activation function, combine the update weight of the input gate and the forget weight of the forget gate, and update the cell state of the recurrent neural network.

[0218] The temporal feature sequence is input into the candidate memory gate of the recurrent neural network, which generates candidate memory features. The candidate memory gate uses the tanh function as an activation function, with output values ​​between -1 and 1. When the temporal feature sequence is input, the candidate memory gate generates a candidate memory feature using the activation function based on the current input features and the previous hidden state. This candidate memory feature represents new information that may be added to the cell state at the current moment.

[0219] The cell state of the recurrent neural network is updated by combining the update weights of the input gate and the forget weights of the forget gate. First, the historical information in the cell state is forgotten based on the forget weights of the forget gate. Then, a portion of the candidate memory features is added to the cell state based on the update weights of the input gate. Specifically, the forgotten historical information is added to the updated candidate memory features to obtain the updated cell state. Through this operation, the recurrent neural network can dynamically update the cell state based on the input features and historical information at different times, capturing the temporal correlations between different feature subsequences.

[0220] Step S435: The updated cell state is input into the output gate of the recurrent neural network, and the hidden state feature at the current moment is generated through the activation function of the output gate. The hidden state feature contains the time correlation information between the current feature subsequence and the historical feature subsequence.

[0221] The updated cell state is input into the output gate of the recurrent neural network. The output gate generates the current hidden state features based on the updated cell state. The output gate also uses the Sigmoid function as the activation function.

[0222] When the updated cell state is input, the output gate calculates the output weight through the activation function based on the current input features and the updated cell state. This output weight indicates which information in the updated cell state is output as the hidden state feature at the current moment.

[0223] Multiplying the updated cell state by the output weight yields the current hidden state feature. This hidden state feature not only contains information about the current feature subsequence but also includes temporal correlations with historical feature subsequences. Because the forget gate and input gate are factored into the cell state update process, historical and current information are rationally integrated. Therefore, the current hidden state feature reflects the temporal dependencies between different feature subsequences, providing an important basis for subsequent fault probability prediction.

[0224] Step S436: Concatenate the hidden state features at all moments to generate a temporal modeling feature containing complete temporal dependency information.

[0225] The purpose of concatenating hidden state features across all time steps is to integrate them into a single feature representation containing complete temporal dependency information. Because the hidden state features at each time step only contain partial temporal correlation information for that time step and previous steps, concatenation can combine this information across all time steps to create a more comprehensive feature representation.

[0226] Specifically, the hidden state features of each moment are arranged in chronological order and then concatenated. For example, if the hidden state feature of each moment is a one-dimensional vector of length n and there are m moments, then after concatenation, a one-dimensional vector of length m×n will be obtained. This vector is the temporal modeling feature that contains complete temporal dependency information.

[0227] This time series modeling feature can fully reflect the temporal correlation between different feature subsequences, providing rich information for subsequent fault probability prediction and helping to improve the accuracy of fault prediction.

[0228] Step S440: Input the features processed by time series modeling into the probability prediction layer of the fault prediction model, perform probability distribution conversion on them through the probability distribution conversion function, and generate the fault probability distribution corresponding to each feature subsequence.

[0229] The features processed by time series modeling are input into the probability prediction layer of the fault prediction model. The main task of the probability prediction layer is to convert the features processed by time series modeling into the fault probability distribution corresponding to each feature subsequence. The probability distribution conversion function is the core of the probability prediction layer. Common probability distribution conversion functions include the Softmax function.

[0230] When input features processed by time series modeling are used, the probability prediction layer calculates the probability value corresponding to each fault type based on these features using the probability distribution conversion function. For example, if an electrical component has three possible fault types, then after processing the probability distribution conversion function, a probability vector of length 3 is obtained, where each element in the vector represents the probability of each fault type occurring. The sum of these probability values ​​is 1, which conforms to the properties of a probability distribution.

[0231] Through this process, a fault probability distribution corresponding to each characteristic subsequence is generated. This fault probability distribution can intuitively reflect the probability of different fault types occurring in the electrical component at the moment represented by each characteristic subsequence.

[0232] Step S450: Arrange the fault probability distributions in the time order of the characteristic subsequences to generate a fault probability distribution sequence of the electrical components, wherein each fault probability distribution includes probability values ​​corresponding to different fault types.

[0233] Arranging the fault probability distributions in chronological order of the characteristic subsequences is intended to integrate the fault probability distributions at different moments into a complete sequence. Because each characteristic subsequence corresponds to a moment in time, arranging their corresponding fault probability distributions in chronological order clearly demonstrates the changing probabilities of different fault types occurring in electrical components at different moments in time.

[0234] Specifically, the fault probability distributions corresponding to each feature subsequence are arranged in chronological order to form a sequence of fault probability distributions. For example, if there are five feature subsequences, and the fault probability distribution corresponding to each feature subsequence is a probability vector of length 3, then after arrangement, a sequence of five probability vectors of length 3 will be obtained.

[0235] Each fault probability distribution contains probability values ​​corresponding to different fault types, which provide a concrete basis for predicting electrical component failures. By analyzing this series of fault probability distributions, we can understand the likelihood trends of various faults occurring in electrical components under different extreme weather conditions, providing important information for subsequent fault prediction.

[0236] Step S500: determining the fault prediction results of the electrical components under different extreme weather conditions according to the fault probability distribution sequence.

[0237] The ultimate goal of the entire fault prediction method is to determine the failure prediction results of electrical components under different extreme weather conditions based on the fault probability distribution sequence. The fault probability distribution sequence already reflects the probability of different fault types occurring in electrical components at different times. By further analyzing and processing this sequence, specific fault prediction results can be obtained.

[0238] Step S510: parse each fault probability distribution in the fault probability distribution sequence, and extract the probability value corresponding to each fault type.

[0239] Parsing each fault probability distribution in a fault probability distribution sequence involves analyzing and processing each element in the sequence to extract key probability information. A fault probability distribution is a vector containing probability values ​​corresponding to different fault types. Each element in this vector needs to be extracted to obtain the probability value corresponding to each fault type.

[0240] For example, a fault probability distribution is a vector of length 4, representing the probability of four different fault types. By traversing this vector and extracting the value of each element, the probability of each fault type can be obtained. Data parsing algorithms can be used to process the fault probability distribution to ensure accurate extraction of the probability value corresponding to each fault type.

[0241] Step S520: Perform time series analysis on the probability value of each fault type, calculate the average and maximum values ​​of the probability values ​​in different weather time units, and determine the risk level of the fault type under extreme weather conditions.

[0242] We perform time series analysis on the probability of each fault type because the probability of a particular fault type occurring in an electrical component can vary within different weather periods. This analysis can help us understand the temporal trends of these probabilities and provide a basis for determining the risk level of each fault type.

[0243] To calculate the average probability value within different weather time units, we add the probability values ​​of the fault type within each weather time unit and divide it by the number of weather time units to obtain the average value. This average value can reflect the average probability of occurrence of the fault type throughout the extreme weather process.

[0244] Calculating the maximum probability value within different weather time periods is to find the maximum value of the probability value of the fault type within each weather time period. This maximum value can reflect the highest probability of occurrence of the fault type during extreme weather.

[0245] The risk level of this fault type under extreme weather conditions is determined based on the average and maximum values. A weighted summation approach can be used: multiplying the average and maximum values ​​by different weights and then adding them together to obtain the overall risk value for this fault type. Weighting can be determined based on specific application requirements and experience. For example, if you are more concerned about the risk of failure in extreme conditions, you can assign a higher weight to the maximum value.

[0246] Step S521: extract the probability values ​​of the same fault type in different weather time units in the fault probability distribution sequence to form a probability time series of the fault type.

[0247] Extracting the probability values ​​of the same fault type in different weather time units from the fault probability distribution sequence is to filter and organize the fault probability distribution sequence to obtain a sequence of the probability of the fault type changing over time. Because the fault probability distribution sequence contains the probability values ​​of multiple fault types at different times, it is necessary to find the probability values ​​of the same fault type in different weather time units. For example, there are 10 elements in the fault probability distribution sequence, each element is a vector of length 5, representing the probability values ​​of 5 different fault types. To extract the probability time series of a certain fault type, it is necessary to extract the probability values ​​corresponding to the fault type from these 10 vectors, and then arrange these probability values ​​in chronological order to form a probability time series of length 10. A data screening algorithm can be used to process the fault probability distribution sequence to ensure the accurate extraction of the probability values ​​of the same fault type in different weather time units.

[0248] Step S522: Perform sliding window smoothing on the probability time series to eliminate short-term fluctuation interference and obtain a smoothed probability time series.

[0249] Sliding window smoothing is performed on probability time series because there may be some short-term fluctuations in probability time series. These fluctuations may be caused by noise or accidental factors, which will affect the accurate judgment of the risk level of the fault type. Therefore, a sliding window smoothing method is needed to eliminate these short-term fluctuations.

[0250] The basic idea behind sliding window smoothing is to select a window of fixed length, slide this window across the probability time series, and average the probability values ​​within the window to obtain a smoothed value at the center of the window. For example, if a window of length 3 is selected, for the second element in the probability time series, the first, second, and third elements are added together and then divided by 3 to obtain the smoothed value of the second element. The window is then continuously slid across the entire probability time series to obtain a smoothed probability time series. Sliding window smoothing can smooth the probability time series, highlighting long-term trends in probability and providing more reliable data for subsequent average and maximum value calculations.

[0251] Step S523: Calculate the arithmetic mean of the smoothed probability time series in each weather period unit as the average risk value of the period.

[0252] The arithmetic mean of the smoothed probability time series within each weather period is calculated to obtain the average probability of occurrence of the fault type within each weather period, which serves as the average risk value for that period. The average risk value for that period is obtained by adding the probability values ​​for each weather period in the smoothed probability time series and dividing it by the number of probability values ​​within that weather period.

[0253] Step S524: Find the maximum value of the smoothed probability time series in each weather period unit as the peak risk value of the period.

[0254] Finding the maximum value of the smoothed probability time series within each weather period is to identify the highest possible probability of the fault type within each weather period, which serves as the peak risk value for that period. Traverse the probability values ​​belonging to a weather period in the smoothed probability time series and find the maximum value.

[0255] Step S525: perform weighted summation on the average risk value and the peak risk value to obtain the comprehensive risk value of the fault type in the corresponding weather period unit, and use the comprehensive risk value as the risk level of the fault type under extreme weather conditions.

[0256] The weighted sum of the average and peak risk values ​​is used to comprehensively consider the average and maximum probability of occurrence of a fault type in each weather period, resulting in a more comprehensive risk assessment. Different weights are assigned to the average and peak risk values ​​based on specific application requirements and experience.

[0257] Step S530: Compare the risk level with a preset risk threshold. When the risk level exceeds the risk threshold, mark the fault type as a high-risk fault type.

[0258] The risk level is compared to a preset risk threshold to determine whether the risk of a particular fault type under extreme weather conditions is high enough to warrant attention. The preset risk threshold is a pre-set standard based on factors such as the criticality of the electrical component, maintenance costs, and the consequences of a failure.

[0259] When the risk level exceeds the risk threshold, it indicates that the fault type is more likely to occur under extreme weather conditions and may seriously affect the normal operation of the electrical components. Therefore, the fault type needs to be marked as a high-risk fault type. A comparison algorithm can be used to compare the risk level with a preset risk threshold. When the risk level is greater than the preset risk threshold, the fault type is marked as high risk.

[0260] Step S540: Count the high-risk fault type sets corresponding to different extreme weather types, and establish a correspondence table between extreme weather types and high-risk fault types.

[0261] To compile statistics on high-risk fault types corresponding to different extreme weather types, we categorize and organize all fault types marked as high-risk, grouping them according to the extreme weather type. For example, extreme weather types include heavy rain, strong winds, and lightning, and we calculate the high-risk fault types that occur under each extreme weather type.

[0262] Establishing a table that maps extreme weather types to high-risk fault types presents statistical results in a tabular format, facilitating subsequent query and analysis. Each row in the table represents an extreme weather type, while the columns list the corresponding high-risk fault types. Data statistics and table generation algorithms can be used to categorize and count high-risk fault types and generate a corresponding table.

[0263] Step S550: Generate a fault prediction result including extreme weather types, high-risk fault types and corresponding probability values ​​according to the correspondence table and the probability values ​​in the fault probability distribution sequence.

[0264] According to the correspondence table and the probability values ​​in the fault probability distribution sequence, a fault prediction result including extreme weather types, high-risk fault types and corresponding probability values ​​is generated. This is to integrate the information obtained in the previous steps to form a complete fault prediction report.

[0265] Step S551: Parse the correspondence table and extract a list of high-risk fault types corresponding to each extreme weather type.

[0266] Parsing the correspondence table involves analyzing and processing the information in the correspondence table to extract key fault type information. The correspondence table records the high-risk fault types corresponding to different extreme weather types. A list of high-risk fault types corresponding to each extreme weather type needs to be extracted.

[0267] For example, a correspondence table contains three extreme weather types, each of which corresponds to several high-risk fault types. By traversing the correspondence table, the high-risk fault types corresponding to each extreme weather type are extracted and formed into a list. A data parsing algorithm can be used to process the correspondence table to ensure that the list of high-risk fault types corresponding to each extreme weather type is accurately extracted.

[0268] Step S552: traverse the high-risk fault type list and search the fault probability distribution sequence for the probability value of each high-risk fault type in the corresponding extreme weather period unit.

[0269] Traversing the list of high-risk fault types involves processing each high-risk fault type in turn to find its probability value for the corresponding extreme weather period. Because the fault probability distribution sequence records the probability values ​​of different fault types at different times, we need to find the probability value of each high-risk fault type for the corresponding extreme weather period.

[0270] Step S553: ​​Combine extreme weather types, high-risk fault types and their corresponding probability values ​​to generate fault prediction entries.

[0271] Combining extreme weather types, high-risk fault types, and their corresponding probability values ​​integrates the information obtained in the previous steps to form a complete fault prediction entry. Each fault prediction entry contains the extreme weather type, high-risk fault type, and the probability value of the fault type in the corresponding extreme weather period.

[0272] Step S554: Classify and sort all fault prediction items according to extreme weather types to form a classified fault prediction item set.

[0273] All fault prediction items are classified and sorted according to extreme weather types, which means that the generated fault prediction items are grouped according to extreme weather types, so as to better present and analyze the fault prediction situations under different extreme weather types.

[0274] For example, all fault prediction entries can be categorized by extreme weather types such as heavy rain, high winds, and lightning. Fault prediction entries belonging to the same extreme weather type can be grouped together to form a classified set of fault prediction entries. Data classification algorithms can be used to process the fault prediction entries to ensure they are accurately categorized and organized by extreme weather type.

[0275] Step S555: converting the classified fault prediction item set into a structured data format to generate a fault prediction result including extreme weather types, high-risk fault types and corresponding probability values.

[0276] Converting the classified fault prediction entry set into a structured data format is to present the fault prediction results in a standardized and easy-to-process manner. Structured data formats can be JSON, XML, and other formats. These formats have clear structures and specifications, making them easy to store, transmit, and analyze.

[0277] By converting the classified set of fault prediction items into a structured data format, a fault prediction result is generated, including extreme weather types, high-risk fault types, and corresponding probability values. A data format conversion algorithm can be used to process the classified set of fault prediction items, converting them into a specified structured data format, ultimately yielding a complete fault prediction result. This result provides an important reference for the maintenance and management of electrical components, helping relevant personnel take proactive measures to reduce the risk of electrical component failures in extreme weather conditions.

[0278] See also Figure 2 , Figure 2This is a schematic diagram of the structure of a computer system provided in an embodiment of the present invention. The computer system includes at least a processor 101, a communication interface 102, and a memory 103. The processor 101, communication interface 102, and memory 103 may be connected via a bus or other means. The processor 101 (also known as the Central Processing Unit (CPU)) is the computing and control core of the computer system, capable of parsing various instructions within the computer system and processing various data within the computer system. The communication interface 102 may optionally include a standard wired interface or a wireless interface (such as Wi-Fi, a mobile communication interface, etc.), which can be used to send and receive data under the control of the processor 101. The communication interface 102 may also be used for data transmission and interaction within the computer system. The memory 103 is a storage device in the computer system for storing programs and data. It is understood that the memory 103 herein may include both the built-in memory of the computer system and, of course, the extended memory supported by the computer system. The memory 103 provides storage space, which stores the computer system's operating system, but this is not limited to this in the present invention. In one embodiment, the processor 101 executes the component failure prediction method based on extreme weather provided in the above embodiment of the present invention by running the computer program in the memory 103 .

Claims

1. A component failure prediction method based on extreme weather, characterized in that: include: Acquire an extreme weather data set and an electrical component operating status data set, wherein the extreme weather data set includes records of occurrence periods and affected areas of different types of extreme weather, and the electrical component operating status data set includes records of operating parameters and historical faults of electrical components in corresponding periods; Performing feature extraction on the extreme weather data set to obtain an extreme weather feature set, and performing feature extraction on the electrical component operating status data set to obtain an electrical component operating feature set; Associating and fusing the extreme weather feature set and the electrical component operation feature set to generate an associated feature sequence; Calling a pre-trained fault prediction model to perform fault risk prediction on the associated feature sequence to generate a fault probability distribution sequence of the electrical component; determining fault prediction results of electrical components under different extreme weather conditions according to the fault probability distribution sequence; The step of extracting features from the extreme weather data set to obtain an extreme weather feature set and extracting features from the electrical component operating state data set to obtain an electrical component operating feature set includes: Dividing the occurrence period records in the extreme weather data set into time series to obtain a plurality of continuous weather period units, and spatially defining the impact area records in each weather period unit to generate weather impact area features; Analyzing the intensity change trend of the weather-affected regional characteristics, extracting the persistence characteristics and change characteristics of extreme weather in different weather time units, and combining the persistence characteristics and the change characteristics to obtain an extreme weather feature set; Performing state classification on the operating parameter records in the electrical component operating state data set to obtain a state parameter sequence of the electrical component under different operating conditions, performing fault pattern matching on the historical fault records to extract fault type features corresponding to the state parameter sequence; Performing correlation mapping on the state parameter sequence and the fault type feature to generate an electrical component operation feature set including a correspondence between the operation state and the fault type; The step of associating and fusing the extreme weather feature set and the electrical component operation feature set to generate an associated feature sequence includes: Performing time alignment on each weather period unit in the extreme weather feature set and each state segment in the electrical component operation feature set, and determining a correspondence between weather period units and state segments having overlapping time intervals; According to the corresponding relationship, the extreme weather characteristics of the weather period unit and the electrical component operation characteristics of the state segment are matched with feature fields to extract the persistence characteristics and change characteristics of the extreme weather characteristics and the state change characteristics and fault type characteristics of the electrical component operation characteristics; The matched feature fields are expanded in feature dimension. The time component of the continuous feature is calculated by ratio with the duration of the state segment to obtain the time correlation coefficient. The spatial component of the change feature is calculated by distance with the electrical component position information corresponding to the state segment to obtain the spatial correlation coefficient. Using the temporal correlation coefficient and the spatial correlation coefficient as fusion weights, performing weighted fusion on extreme weather characteristics and electrical component operation characteristics to generate preliminary fusion characteristics; Redundant features are eliminated from the preliminary fusion features, repeated feature fields and feature fields with contributions lower than a preset threshold are deleted to obtain key fusion features, which are arranged in chronological order to generate a correlation feature sequence.

2. The method according to claim 1, characterized in that The time series division of the occurrence period records in the extreme weather data set to obtain a plurality of continuous weather period units, the spatial scope of the impact area records in each weather period unit is defined, and the weather impact area characteristics are generated, including: parsing the occurrence period records in the extreme weather data set, extracting the start time mark and the end time mark of each extreme weather event, calculating the time interval between the start time mark and the end time mark, and equally dividing the time interval according to a preset time window length to obtain a plurality of continuous weather period units; Performing geographic information analysis on the impact area records corresponding to each weather period unit, extracting a set of boundary coordinate points of the impact area, and sorting the set of boundary coordinate points in a clockwise direction to generate a closed regional boundary polygon; Calculating the geometric center coordinates and area parameters of the region boundary polygon, determining the spatial center position of the affected region based on the geometric center coordinates, and determining the coverage size of the affected region based on the area parameters; Extracting geographic partition information within the regional boundary polygon, determining the electrical component distribution area within the extreme weather impact area, and generating electrical component density distribution characteristics within the area; The spatial center position, coverage size and electrical component density distribution characteristics are combined to obtain weather impact area characteristics, wherein each weather time period unit corresponds to a weather impact area characteristic.

3. The method according to claim 1, characterized in that The intensity change trend analysis of the weather impact area characteristics is performed to extract the persistence characteristics and change characteristics of extreme weather in different weather time units, and the persistence characteristics and change characteristics are combined to obtain an extreme weather feature set, including: Obtain the coverage size parameter in the weather impact area characteristics corresponding to each weather period unit, calculate the change ratio of the coverage size between adjacent weather period units, and generate a coverage change rate sequence; Performing trend fitting on the coverage change rate sequence to determine the expansion trend direction or contraction trend direction of the extreme weather impact area, and using the expansion trend direction or contraction trend direction as the direction component of the change feature; Counting the duration of extreme weather in each weather period unit, calculating the sum of the durations of consecutive weather period units, and obtaining the cumulative duration of extreme weather, which is used as the time component of the persistence feature; Analyze the changes in the electrical component density distribution characteristics in the weather-affected area characteristics in different weather time periods, extract the peak position movement trajectory of the density distribution, and use the peak position movement trajectory as the spatial component of the change characteristics; The direction component and the space component are combined to obtain a change feature, the time component is used as a continuous feature, and the change feature and the continuous feature are arranged in the time sequence of the weather period unit to generate an extreme weather feature set.

4. The method according to claim 1, wherein The performing state classification on the operating parameter records in the electrical component operating state data set to obtain a state parameter sequence of the electrical component under different operating conditions, performing fault pattern matching on the historical fault records, and extracting fault type features corresponding to the state parameter sequence, includes: parsing the operating parameter records in the electrical component operating status data set to extract multidimensional operating parameters including voltage fluctuation records, current fluctuation records, and temperature change records; Dividing the multi-dimensional operating parameters into state intervals to obtain interval division results corresponding to each dimension; Classifying the operating status of the electrical component according to the interval division result to obtain a combined status identifier including a voltage status label, a current status label, and a temperature status label, and merging continuous operating parameter records with the same combined status identifier into a status segment to generate a status parameter sequence; Parsing the historical fault records, extracting the fault type description and the corresponding fault occurrence time period when the fault occurred, performing time matching on the fault occurrence time period with the state segments in the state parameter sequence, and determining the state segment set corresponding to each fault type; Statistical analysis is performed on the combined state identifiers in the state segment set, and the combined state identifier with the highest occurrence frequency is extracted as the typical state feature of the fault type. The typical state feature and the fault type description are combined to obtain the fault type feature.

5. The method according to claim 1, wherein The associating and mapping the state parameter sequence and the fault type characteristics to generate an electrical component operation characteristic set including a correspondence between the operation state and the fault type includes: Performing a similarity comparison between each state segment in the state parameter sequence and a typical state feature in the fault type feature, and calculating a matching value between the combined state identifier of the state segment and the typical state feature; When the matching value exceeds a preset threshold, an association relationship between the state segment and the corresponding fault type feature is established, and the state segment start time, state segment end time and fault type description in the association relationship are recorded; Analyze the parameter change trends of the associated state segments, extract the change amplitude of the voltage fluctuation record, the change frequency of the current fluctuation record, and the rising rate of the temperature change record within the state segment, and combine these parameters to obtain the state change characteristics; Integrate the fault type description, status segment start time, status segment end time and status change characteristics in the association relationship to generate an association mapping table containing the corresponding relationship between the operating status and the fault type; The association mapping table is arranged in a time sequence of the state segments to form an electrical component operation feature set, wherein each association mapping table entry corresponds to a feature unit in the electrical component operation feature set.

6. The method according to claim 1, wherein The step of time-aligning each weather period unit in the extreme weather feature set and each state segment in the electrical component operation feature set to determine a correspondence between weather period units and state segments having overlapping time intervals includes: Extracting the start timestamp and end timestamp of each weather period unit in the extreme weather feature set to generate a weather period timeline, and extracting the start timestamp and end timestamp of each state segment in the electrical component operation feature set to generate a state segment timeline; Unifying the time bases of the weather period time axis and the state period time axis; Traverse each weather period unit on the weather period timeline, find the state segment on the state segment timeline that overlaps with the time interval of the weather period unit, and calculate the ratio of the length of the overlapping time interval to the total time length of the weather period unit; When the ratio exceeds a preset overlap threshold, determining that there is a correspondence between the weather period unit and the state segment, and recording the weather period unit identifier and the state segment identifier in the correspondence; De-duplicate all determined correspondences, delete duplicate correspondence records, and generate a weather-state correspondence table; The method of using the temporal correlation coefficient and the spatial correlation coefficient as fusion weights to perform weighted fusion on extreme weather characteristics and electrical component operation characteristics to generate preliminary fusion features includes: Convert the persistence features and change features in extreme weather characteristics and the state change features and fault type features in electrical component operation characteristics into numerical feature vectors so that all feature vectors have the same dimension; Calculating a weighted sum of the temporal correlation coefficient and the spatial correlation coefficient to obtain a comprehensive fusion weight, wherein the weight ratio of the temporal correlation coefficient and the spatial correlation coefficient is determined according to a preset rule; The numerical feature vector of the extreme weather characteristics is multiplied by the comprehensive fusion weight to obtain a weighted extreme weather feature vector, and the numerical feature vector of the electrical component operation characteristics is multiplied by the difference between 1 and the comprehensive fusion weight to obtain a weighted electrical component operation feature vector; Adding the weighted extreme weather feature vector and the weighted electrical component operation feature vector element by element to obtain a preliminary fused feature vector; The preliminary fusion feature vector is subjected to feature normalization, and each feature value is mapped into a preset numerical range to generate a preliminary fusion feature.

7. The method according to claim 1, characterized in that The calling of the pre-trained fault prediction model to perform fault risk prediction on the associated feature sequence to generate a fault probability distribution sequence of the electrical component includes: Inputting the associated feature sequence into the feature input layer of the fault prediction model, segmenting the feature sequence in the time dimension to obtain multiple feature subsequences of fixed length; Inputting each feature subsequence into the feature extraction layer of the fault prediction model, performing nonlinear transformation on the feature subsequence through a multi-layer perceptron to extract high-level abstract features; Inputting the high-level abstract features into the time series modeling layer of the fault prediction model, performing time series dependency modeling on the high-level abstract features through a recurrent neural network, extracting the time correlation information between different feature subsequences, and obtaining features processed by time series modeling; Input the features processed by time series modeling into the probability prediction layer of the fault prediction model, perform probability distribution conversion on them through the probability distribution conversion function, and generate the fault probability distribution corresponding to each feature subsequence; The fault probability distributions are arranged in a time sequence of characteristic subsequences to generate a fault probability distribution sequence of the electrical component, wherein each fault probability distribution includes probability values ​​corresponding to different fault types.

8. A computer system, characterized in that: include: a memory storing a computer program; A processor is used to load the computer program to implement the component failure prediction method based on extreme weather as described in any one of claims 1 to 7.

Images