Power distribution system operation state monitoring method and system based on big data analysis

Abstract

The invention provides a big data analysis-based power distribution system operation state monitoring method and system, and the method comprises the steps: firstly collecting the multi-source operation data of a power distribution system to construct a feature association map, calling a pre-training state mapping model to carry out the hierarchical state analysis, and generating the initial state description of each operation unit; and inputting the initial state description into a time sequence state evolution model to obtain an operation state sequence in a continuous time period. Performing abnormal feature tracing analysis on the operation state sequence, and determining an operation link and associated features corresponding to an abnormal state; and finally, generating a power distribution system operation state monitoring report containing the state identifier and the associated characteristics. The method can comprehensively integrate data, accurately describe the operation state, dynamically track the change trend, quickly position abnormity, and effectively improve the operation stability and reliability of the power distribution system.

Description

Technical Field

[0001] This invention relates to the field of power, and more specifically, to a method and system for monitoring the operating status of a power distribution system based on big data analysis. Background Technology

[0002] In power systems, the distribution system, as a crucial link in power transmission and distribution, directly impacts the quality of the entire power supply through its operational stability and reliability. With the continuous development of power systems, the scale of distribution systems is expanding, their structures becoming increasingly complex, and the amount of data generated during operation is exploding. Traditional methods for monitoring the operational status of distribution systems mainly rely on manual inspections and simple data monitoring equipment, which have significant limitations. Manual inspections are not only inefficient but also struggle to obtain real-time and comprehensive operational information, making it difficult to detect potential faults in a timely manner. Simple data monitoring equipment often only acquires data from a single dimension, failing to perform comprehensive analysis and correlation mining of multi-source data, making it difficult to accurately determine the overall operational status of the distribution system. Furthermore, traditional methods lack the ability to dynamically predict operational status and trace anomalies, failing to provide early warnings and take effective preventative measures before faults occur, making it difficult to quickly locate the root cause of problems when faults occur, thus affecting the efficiency of power supply restoration. Therefore, developing an efficient and accurate method for monitoring the operational status of distribution systems is of significant practical importance. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for monitoring the operating status of a power distribution system based on big data analysis, the method comprising: Collect multi-source operation data generated during the operation of the power distribution system, construct a correlation system of power distribution system operation characteristics, and obtain a feature correlation map; The pre-trained state mapping model is invoked to perform hierarchical state parsing on the feature association map, generating initial state descriptions for each operating unit of the power distribution system; The initial state description is input into the time-series state evolution model to generate a sequence of running states over a continuous time period; Perform anomaly feature tracing analysis on the operation status sequence to determine the operation links and related features corresponding to the abnormal states; Based on the operational processes and associated characteristics corresponding to abnormal states, a power distribution system operation status monitoring report containing status identifiers and associated characteristics is generated.

[0004] In another aspect, embodiments of the present invention also provide a power distribution system operation status monitoring system based on big data analysis, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code. The processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-mentioned method.

[0005] Based on the above, this invention constructs a feature association map by collecting multi-source operational data during the operation of the power distribution system. This comprehensively and systematically integrates operational information from different dimensions, breaking through the limitations of traditional methods that rely on single and partial data, and providing a rich data foundation for accurate monitoring of operational status. By calling a pre-trained state mapping model to perform hierarchical state analysis on the feature association map, the potential features and state information behind the data can be deeply mined, generating initial state descriptions for each operating unit, thus achieving a precise characterization of the operational status. Inputting the initial state descriptions into a time-series state evolution model generates a sequence of operational states over continuous periods, enabling dynamic tracking of operational status changes, early prediction of potential anomalies, and enhanced predictive capabilities for operational status. Anomaly feature tracing analysis of the operational state sequence can quickly locate the operational links and associated features corresponding to abnormal states, providing a clear direction for fault investigation and repair. The final generated power distribution system operational status monitoring report, containing state identifiers and associated features, can intuitively and clearly present key information about the operational status, effectively improving the stability and reliability of the power distribution system, reducing the failure rate, and ensuring the continuity of power supply. Attached Figure Description

[0006] Figure 1 This is a schematic diagram of the execution flow of the power distribution system operation status monitoring method based on big data analysis provided in the embodiments of the present invention.

[0007] Figure 2 This is a schematic diagram of exemplary hardware and software components of the power distribution system operation status monitoring system based on big data analysis provided in an embodiment of the present invention. Detailed Implementation

[0008] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a method for monitoring the operation status of a power distribution system based on big data analysis, provided in one embodiment of the present invention. The following is a detailed description of this method for monitoring the operation status of a power distribution system based on big data analysis.

[0009] Step S110: Collect multi-source operation data generated during the operation of the power distribution system, construct a power distribution system operation feature association system, and obtain a feature association map.

[0010] In this embodiment, the multi-source operational data includes voltage, current, power, temperature, and switch status data collected by various sensors in the power distribution system, as well as equipment parameter data and historical operation record data of the power distribution system. During the data collection process, sensitive data involving user privacy, such as user-side electricity consumption time data and electricity load data, are processed using differential privacy technology. Specifically, noise conforming to a Laplace distribution is added to the original data, so that the statistical characteristics of the data remain unchanged while preventing the deduction of specific user privacy information from the data.

[0011] Step S111: Divide the multi-source running data into feature categories to obtain feature subsets of different running dimensions.

[0012] In this embodiment, feature category division is based on the attributes and uses of the features. First, feature attribute information is extracted from the multi-source operational data to determine the description parameters of each feature. Then, a feature classification standard is constructed based on the description parameters to divide the features into different operational dimension categories. The features in the multi-source operational data are grouped according to the operational dimension categories to form initial feature groups. Feature redundancy detection is performed on the initial feature groups to delete duplicate feature data. The distribution density of features within each initial feature group is calculated, and feature groups with low distribution density are merged. The merged feature groups are named and labeled with the operational dimension corresponding to each feature group. The labeled feature groups are used as feature subsets of different operational dimensions, so that the features within each feature subset have the same operational dimension attributes.

[0013] Step S1111: Extract feature attribute information from multi-source running data and determine the description parameters of each feature.

[0014] In this embodiment, for voltage data, the descriptive parameters include the voltage amplitude range, phase change range, and frequency fluctuation range; for current data, the descriptive parameters include the current amplitude range, phase change range, and frequency fluctuation range; for temperature data, the descriptive parameters include the temperature measurement range and measurement accuracy; for switch status data, the descriptive parameters include the switch's open / closed state and state duration; for equipment parameter data, the descriptive parameters include the equipment's rated voltage, rated current, and rated power; and for historical operation record data, the descriptive parameters include operating time, fault occurrence time, and maintenance record time.

[0015] Step S1112: Construct a feature classification standard based on the description parameters, and divide the features into different operational dimension categories.

[0016] In this embodiment, features are categorized into four dimensions based on their descriptive parameters: power quality, equipment status, load characteristics, and environmental impact. The power quality dimension primarily involves parameters related to power quality, such as voltage and current amplitude, phase, and frequency. The equipment status dimension primarily involves parameters related to equipment operating status, such as temperature, insulation resistance, and switch status. The load characteristics dimension primarily involves parameters related to load characteristics, such as load magnitude, rate of change, and peak-to-valley values. The environmental impact dimension primarily involves parameters related to environmental impact, such as temperature, humidity, and wind speed. Following this classification standard, all features in the multi-source operational data are categorized to ensure that each feature belongs to its corresponding operational dimension category.

[0017] Step S1113: Group the features in the multi-source running data according to the running dimension category to form initial feature groups.

[0018] In this embodiment, features belonging to the same category are grouped together based on the determined operational dimension categories. For example, all features related to voltage amplitude, voltage phase, voltage frequency, current amplitude, current phase, and current frequency are grouped into the power quality dimension feature group; all features related to transformer temperature, circuit breaker opening and closing times, and cable insulation resistance are grouped into the equipment status dimension feature group; all features related to load factor, peak-to-valley difference, and load fluctuation are grouped into the load characteristic dimension feature group; and all features related to ambient temperature, humidity, and wind speed are grouped into the environmental impact dimension feature group. This forms the initial feature groups, with each feature group corresponding to an operational dimension category.

[0019] Step S1114: Perform feature redundancy detection on the initial feature group and delete duplicate feature data.

[0020] In this embodiment, feature redundancy detection is performed by comparing the descriptive parameters and data content of the features. For example, for two voltage amplitude features in the power quality dimension feature group, if their descriptive parameters such as measurement time, measurement location, and measurement accuracy are exactly the same, and their data content is also completely identical, then these two features are considered duplicates, and one of them is deleted. Through such detection and deletion operations, redundant data in the feature group is reduced, improving the efficiency of subsequent processing.

[0021] Step S1115: Calculate the distribution density of features within each initial feature group, and merge feature groups with excessively low distribution densities.

[0022] In this embodiment, the distribution density of a feature refers to the frequency of its occurrence in the dataset. For each initial feature group, the number of times each feature appears in the multi-source running data is counted, and then the average number of occurrences of all features within that feature group is calculated, which is used as the distribution density of that feature group. If the distribution density of a feature group is too low, it indicates that the number of features in that group is small, or that the frequency of their occurrence in the dataset is low. Such feature groups are then merged with other feature groups with similar distribution densities. For example, the humidity and wind speed features in the environmental impact dimension feature group have low distribution densities, so they are merged into an environmental meteorological feature group. This increases the size of the feature groups and improves their representativeness.

[0023] Step S1116: Name the merged feature groups and mark the running dimension corresponding to each feature group.

[0024] In this embodiment, for the merged feature groups, a corresponding name identifier is assigned based on the operational dimension category to which the included features belong. For example, the merged environmental meteorological feature group belongs to the environmental impact dimension, so its name is identified as the Environmental Impact Dimension Environmental Meteorological Feature Group; the power quality dimension feature group is identified as the Power Quality Dimension Voltage and Current Feature Group; the equipment status dimension feature group is identified as the Equipment Status Dimension Equipment Parameter Feature Group; and the load characteristic dimension feature group is identified as the Load Characteristic Dimension Load Data Feature Group.

[0025] Step S1117: Use the identified feature groups as feature subsets of different operating dimensions, so that the features in each feature subset have the same operating dimension attribute.

[0026] In this embodiment, after the above-described grouping, redundancy detection, merging, and name identification processes, the resulting feature groups are feature subsets for different operational dimensions. Features within each feature subset belong to the same operational dimension category and have the same operational dimension attributes. For example, features within the environmental impact dimension's environmental meteorology feature group all belong to the environmental impact dimension and have environmental impact-related attributes; similarly, features within the power quality dimension's voltage and current feature group all belong to the power quality dimension and have power quality-related attributes.

[0027] Step S112: Extract the correlation strength information between features within each feature subset and generate a feature correlation matrix.

[0028] In this embodiment, the correlation strength between features is determined by calculating the correlation coefficient between features. For any two features within each feature subset, the correlation coefficient between them is calculated. The coefficient ranges from -1 to 1, with a larger absolute value indicating a stronger correlation between the two features. These correlation coefficients are arranged in the order of the features to generate a feature correlation matrix. Each element in the feature correlation matrix represents the correlation strength between the corresponding two features, with rows and columns corresponding to features in the feature subset.

[0029] Step S113: Based on the association strength information in the feature association matrix, remove weakly associated features and retain the set of strongly associated features.

[0030] In this embodiment, a correlation strength threshold is set. Features in feature pairs whose absolute correlation strength is less than the threshold in the feature correlation matrix are considered weakly correlated features and are removed. For example, in the feature correlation matrix of the power quality dimension feature subset, if the absolute correlation strength between the voltage amplitude feature and another feature is less than the set threshold, then that other feature is removed from the feature subset.

[0031] Step S114: Perform hierarchical classification of the strongly correlated feature set to determine the hierarchical relationship between features.

[0032] In this embodiment, the hierarchical classification process is based on the correlation strength and logical relationship between features. First, the features in the strongly correlated feature set are sorted from high to low according to their correlation strength. Then, based on the logical relationship between features, features with high correlation strength and logically subordinate relationships are grouped into the same level. For example, in the strongly correlated feature set of the power quality dimension, voltage amplitude and current amplitude features have high correlation strength, and both voltage amplitude and current amplitude are basic indicators of power quality, so they are grouped into the first level; voltage phase features and current phase features have high correlation strength, and they are derived indicators based on voltage amplitude and current amplitude, so they are grouped into the second level; frequency features have a certain correlation with both voltage amplitude and current amplitude, and frequency is an important indicator of power quality, so they are grouped into the first level.

[0033] Step S115: Construct a tree-structured feature association system based on hierarchical relationships, and generate a feature association map containing feature hierarchy and association relationships.

[0034] In this embodiment, the tree-structured feature association system uses the top-level feature as the root node, the next level of features as child nodes, and so on. For example, in the feature association system for the power quality dimension, the top-level feature is the comprehensive power quality feature, the first-level features are voltage amplitude, current amplitude, and frequency features, the second-level features are voltage phase and current phase features, and the third-level features are harmonic content features, etc. Each child node is connected to its parent node by association strength; child nodes with higher association strength are more closely connected to their parent nodes. Arranging these features according to their hierarchical relationship generates a feature association graph, which clearly shows the hierarchical relationship and association strength between features.

[0035] Step S116: Perform unique identification processing on the feature nodes in the feature association graph to make the source of the running data corresponding to each feature node traceable.

[0036] In this embodiment, the unique identification process involves assigning a unique identifier to each feature node. Simultaneously, the source of the operational data corresponding to that feature is recorded in the feature node's attribute information, such as the sensor number, the model of the data acquisition device, and the location of the data storage. Thus, when it is necessary to trace the source of the operational data corresponding to a specific feature node, it can be quickly found by querying the feature node's attribute information.

[0037] Step S117: Visualize the topology of the identified feature nodes according to their hierarchical relationships to form a structured feature association graph.

[0038] In this embodiment, the visualization topology construction uses a graphical approach to display the feature association graph. Using professional visualization tools, feature nodes are arranged according to hierarchical relationships, with top-level feature nodes at the top of the graph, and the next level of feature nodes below them, and so on. The connections between feature nodes are represented by lines, with the thickness of the lines indicating the strength of the association; thicker lines indicate a stronger association. Each feature node is also labeled with its unique identifier and corresponding feature name for easy viewing and understanding. This visualization topology construction forms a structured feature association graph, making the relationships between the operating features of the power distribution system more intuitive and clear.

[0039] Step S120: Call the pre-trained state mapping model to perform hierarchical state parsing on the feature association map and generate the initial state description of each operating unit of the power distribution system.

[0040] In this embodiment, the pre-trained state mapping model is built based on deep learning algorithms and includes a feature input layer, a hierarchical parsing module, a unit matching module, and a state semantic transformation module. This model is pre-trained using a large amount of power distribution system operation data, learning the mapping relationship between feature association maps and operating unit states. When the model is invoked, the feature association map is input into the model, which performs hierarchical state parsing to generate initial state descriptions for each operating unit.

[0041] Step S121: Input the feature association graph into the feature input layer of the state mapping model, and perform dimensional transformation on the hierarchical features in the feature association graph.

[0042] In this embodiment, the main function of the feature input layer is to convert the hierarchical features in the feature association graph into dimensions suitable for model processing. The hierarchical features in the feature association graph exist in the form of a tree structure, and each feature node has multiple attribute information, such as feature name, association strength, and hierarchical position. The feature input layer encodes this attribute information and converts it into a vector of fixed dimensions.

[0043] Step S122: The transformed hierarchical features are analyzed layer by layer from top to bottom through the hierarchical parsing module of the state mapping model to extract the state representation information corresponding to each level feature.

[0044] In this embodiment, the hierarchical parsing module adopts a structure combining convolutional neural networks and recurrent neural networks. First, starting with the top-level features of the feature association map, initial parsing is performed on the transformed hierarchical features to extract the core attribute information of the top-level features. Based on the core attribute information of the top-level features, the parsing extends to the next level of features to determine the association method between the next level of features and the top-level features. Attributes are extracted from the next level of features according to the determined association method to generate the feature attribute set of the current level. For the next level of features, based on the attribute set of the previous level's features, the parsing extends to determine the association method between the next level's features and the previous level's features. Attributes are extracted from the next level's features according to the determined association method to generate the feature attribute set of the current level. For each feature in the feature association graph, excluding the top-level feature, the next-level feature, and the next-next-level feature, an extended analysis is performed based on the attribute set of the previous-level feature to determine the association method between the corresponding-level feature and the previous-level feature. Attributes are extracted from the corresponding-level feature according to the determined association method to generate a feature attribute set for the corresponding-level feature. During the analysis, the association strength and association type between each-level feature and the previous-level feature are recorded. The feature attribute set of each level is integrated with the corresponding association strength and association type to generate state representation information containing the feature attributes, association strength, and association type of each level.

[0045] Step S1221: Starting from the top-level features of the feature association graph, perform initial analysis on the transformed hierarchical features and extract the core attribute information of the top-level features.

[0046] In this embodiment, the top-level feature of the feature association map is the comprehensive operation feature of the power distribution system. The transformed hierarchical feature is a matrix composed of multiple vectors. The initial parsing involves processing this matrix to extract the core attribute information from the vectors corresponding to the top-level feature, such as average voltage, average current, average power, and average temperature. By extracting these attribute information, the core attribute information of the top-level feature is obtained.

[0047] Step S1222: Based on the core attribute information of the top-level feature, extend the analysis to the next level feature to determine the association between the next level feature and the top-level feature.

[0048] In this embodiment, the next-level features include voltage features, current features, power features, temperature features, etc. Further analysis involves calculating the correlation coefficient between the vector corresponding to the next-level feature and the core attribute information of the top-level feature, and determining the association method based on the magnitude of the correlation coefficient. The larger the absolute value of the correlation coefficient, the stronger the association between the next-level feature and the top-level feature.

[0049] Step S1223: Extract attributes from the next level features according to the determined association method to generate the feature attribute set of the current level.

[0050] In this embodiment, attributes are extracted from the vectors corresponding to the next-level features based on the association between the next-level features and the top-level features. For example, if the voltage feature is strongly associated with the top-level features, attribute information such as the maximum voltage value, minimum voltage value, and voltage standard deviation are extracted from the vectors corresponding to the voltage feature; if the current feature is moderately associated with the top-level features, attribute information such as the maximum current value and minimum current value are extracted from the vectors corresponding to the current feature. These extracted attribute information are then integrated to generate the feature attribute set for the current level.

[0051] Step S1224: For the next level of features, perform extended analysis based on the attribute set of the previous level of features to determine the association method between the next level of features and the previous level of features. Extract attributes from the next level of features according to the determined association method to generate the feature attribute set of the current level.

[0052] In this embodiment, the next-level features include voltage phase features, current phase features, power factor features, and temperature change rate features. The extended analysis involves calculating the correlation coefficient between the vector corresponding to the next-level feature and the feature attribute set of the previous level, and determining the association method based on the magnitude of the correlation coefficient. According to the determined association method, attributes are extracted from the vector corresponding to the next-level feature to generate the feature attribute set of the current level.

[0053] Step S1225: For each level feature in the feature association graph except for the top-level feature, the next level feature, and the next level feature, perform extended analysis based on the attribute set of the previous level feature of the corresponding level feature to determine the association method between the corresponding level feature and the previous level feature. Extract attributes of the corresponding level feature according to the determined association method to generate the feature attribute set of the corresponding level feature.

[0054] In this embodiment, the hierarchical features other than those mentioned above include harmonic content features, three-phase unbalance features, and equipment insulation resistance features. Extended analysis involves calculating the correlation coefficient between the vectors corresponding to these hierarchical features and the previous hierarchical feature attribute set to determine the association method. Attribute extraction is then performed on the vectors corresponding to these hierarchical features according to the association method to generate the feature attribute set for the corresponding hierarchical feature.

[0055] Step S1226: During the parsing process, record the correlation strength and correlation type between each level feature and the previous level feature.

[0056] In this embodiment, the correlation strength is represented by the magnitude of the correlation coefficient; the larger the absolute value of the correlation coefficient, the stronger the correlation. The correlation type includes linear and nonlinear correlations, which are determined by calculating the type of correlation coefficient between features.

[0057] Step S1227: Integrate the feature attribute set of each level with the corresponding association strength and association type.

[0058] In this embodiment, integration involves arranging the feature attribute sets, association strengths, and association types of each level in hierarchical order to form a complete feature information structure. For example, the core attribute information of the top-level feature is located at the top layer of the structure; the feature attribute sets, association strengths, and association types of the next level feature are located below the top-level feature; the relevant information of the next level feature is located below the next level feature, and so on. Through this integration, a complete structure containing feature information from each level is formed.

[0059] Step S1228: Generate state representation information containing feature attributes, association strength, and association type at each level.

[0060] In this embodiment, the state representation information is obtained by processing the integrated feature information structure. The integrated feature information structure is converted into a standardized format, which includes information such as the attribute name, attribute value, association strength, and association type of each level feature.

[0061] Step S123: Use the unit matching module of the state mapping model to associate and match the state representation information of each level with the preset running unit feature library.

[0062] In this embodiment, the preset operating unit feature library is a database storing feature information of various operating units in the power distribution system, including transformer units, circuit breaker units, cable units, busbar units, and other operating units. The feature information of each operating unit includes unit name, unit model, rated parameters, and operating status parameters. The function of the unit matching module is to compare the status representation information at each level with the feature information in the operating unit feature library to find the operating unit with the highest matching degree. The matching degree is calculated by comparing the similarity between the feature attributes in the status representation information and the feature attributes in the operating unit feature library; the higher the similarity, the higher the matching degree.

[0063] Step S124: Determine the feature combination corresponding to each running unit based on the matching results, and generate a feature list for the running unit.

[0064] In this embodiment, the matching result is a list containing the matching degree between each operating unit and its state representation information. Based on the matching degree, the feature attribute from the state representation information with the highest matching degree is selected for each operating unit to form the corresponding feature combination. For example, the feature combination for a transformer unit includes voltage amplitude characteristics, current amplitude characteristics, and temperature characteristics; the feature combination for a circuit breaker unit includes current amplitude characteristics and switch status characteristics. These feature combinations are arranged according to the name of the operating unit to generate an operating unit feature list.

[0065] Step S125: Perform state semantic transformation processing on the features in the feature list of the running unit to transform the feature information into a describable state language.

[0066] In this embodiment, the state semantic conversion processing employs natural language processing technology to convert numerical values, symbols, and other information in the feature information into natural language descriptions. For example, the numerical value of the voltage amplitude feature is converted into state language such as voltage amplitude within the normal range, voltage amplitude too high, and voltage amplitude too low; the numerical value of the current amplitude feature is converted into state language such as current amplitude within the normal range and current amplitude too high; and the numerical value of the temperature feature is converted into state language such as temperature normal and temperature too high. Through such conversion, the feature information becomes easier to understand and describe.

[0067] Step S126: Combine the feature combination and state language of each running unit to generate an initial state description that includes the running unit identifier and the corresponding state description.

[0068] In this embodiment, the operating unit identifier is a unique identifier for each operating unit. For example, the identifier for a transformer unit is the transformer unit identifier, and the identifier for a circuit breaker unit is the circuit breaker unit identifier. The initial state description integrates the identifier of each operating unit with its corresponding feature combination and state language to form a complete description. For example, the initial state description of a transformer unit is: transformer unit, voltage amplitude is within the normal range, current amplitude is within the normal range, and temperature is normal; the initial state description of a circuit breaker unit is: circuit breaker unit, current amplitude is within the normal range, and the switch state is closed, etc. These initial state descriptions are arranged according to the identifier of the operating unit to generate the initial state description of each operating unit in the power distribution system.

[0069] Step S130: Input the initial state description into the time-series state evolution model to generate a sequence of running states over a continuous time period.

[0070] In this embodiment, the time-series state evolution model is constructed based on time series analysis algorithms and includes a time-series encoding module and a state prediction module. This model can predict the operating state in future time periods by learning from historical operating state data of the power distribution system. When an initial state description is input, the model performs time-series encoding processing, then performs state prediction to generate a sequence of operating states over consecutive time periods.

[0071] Step S131: Perform time dimension marking processing on the initial state description, and add the corresponding collection period identifier to the state description of each running unit.

[0072] In this embodiment, the data acquisition period identifier refers to the time interval for collecting operational data corresponding to the initial state description. The time dimension marking process involves adding this data acquisition period identifier to the initial state description of each operational unit, for example, adding the data acquisition period identifier to the initial state description of the transformer unit.

[0073] Step S132: Input the initial state description with the acquisition period identifier into the time-series coding module of the time-series state evolution model to generate a time-series feature vector.

[0074] In this embodiment, the function of the time-series encoding module is to convert the initial state description with the acquisition period identifier into a time-series feature vector. First, the state language in the initial state description is encoded and converted into a numerical vector; then, the acquisition period identifier is encoded and converted into a time vector; finally, the numerical vector and the time vector are concatenated to generate the time-series feature vector.

[0075] Step S133: Analyze the state evolution trend of the temporal feature vector over a continuous period using the state prediction module of the temporal state evolution model.

[0076] In this embodiment, the state prediction module is constructed using a Long Short-Term Memory (LSTM) network algorithm, which is capable of processing and predicting time series data. The time series feature vector is input into the LTM network, which processes it and learns the temporal patterns and trends within the feature vector. Then, based on the learned patterns and trends, the module predicts the time series feature vectors for future consecutive time periods, generating state evolution trend analysis results for those consecutive time periods.

[0077] Step S1331: Input the time series feature vector into the trend extraction unit of the state prediction module to extract the changing trend features in the time series feature vector.

[0078] In this embodiment, the trend extraction unit uses a sliding window technique to process the time-series feature vector. A sliding window size is set, dividing the time-series feature vector into multiple windows. The feature vector within each window is calculated to extract its trend characteristics, such as the mean, standard deviation, and slope of the feature vector. By extracting these trend characteristics, the trend information in the time-series feature vector is obtained.

[0079] Step S1332: Construct a time-series evolution model based on the changing trend characteristics, and set the evolution analysis window of the model.

[0080] In this embodiment, the time-series evolution model is constructed based on trend characteristics and employs an autoregressive moving average model algorithm. The size of the evolution analysis window is set, which determines the length of historical data used by the model during prediction. The trend characteristics are input into the autoregressive moving average model, which is then trained to learn the time patterns and evolutionary rules within the trend characteristics.

[0081] Step S1333: Within the evolution analysis window, perform sliding analysis on the time series feature vectors to capture the changing patterns of the feature vectors.

[0082] In this embodiment, sliding analysis involves sliding the evolutionary analysis window across the temporal feature vectors, moving it by one time step at a time, and processing the temporal feature vectors within the window. In each sliding analysis, the mean, standard deviation, slope, and other parameters of the temporal feature vectors within the window are calculated to capture the changing patterns of the feature vectors.

[0083] Step S1334: Determine the rate and direction of change of the feature vector in different time periods based on the change pattern.

[0084] In this embodiment, the rate of change is determined by calculating the ratio of the difference between the time series feature vectors within two adjacent time steps to the time interval. The direction of change is determined by the sign of the difference; a positive difference indicates an increasing direction of change, and a negative difference indicates a decreasing direction of change.

[0085] Step S1335: Combine the rate of change and the direction of change to predict the possible range of change of the feature vector in subsequent time periods.

[0086] In this embodiment, the prediction is based on the rate and direction of change to estimate the time-series feature vector for subsequent periods. For example, if the rate of change in the current period is positive and the direction of change is increasing, then the predicted time-series feature vector for subsequent periods will continue to increase, with the increase based on the current rate of change; if the rate of change in the current period is negative and the direction of change is decreasing, then the predicted time-series feature vector for subsequent periods will continue to decrease, with the decrease based on the current rate of change. Furthermore, considering that the rate of change may change, a possible range of variation is given during prediction, rather than a fixed value.

[0087] Step S1336: Compare the possible range of changes with the preset state evolution threshold, and filter out the change trends that conform to the evolution logic.

[0088] In this embodiment, the preset state evolution thresholds are determined based on the operating requirements of the power distribution system and historical data, including voltage thresholds, current thresholds, and temperature thresholds. The predicted possible range of change is compared with these thresholds. If the possible range of change is within the threshold range, the change trend is considered to conform to the evolution logic; if the possible range of change exceeds the threshold range, the change trend is considered to not conform to the evolution logic and is discarded.

[0089] Step S1337: Integrate the filtered change trends to generate the state evolution trend analysis results for each operating unit.

[0090] In this embodiment, integration involves arranging the filtered trends in chronological order to form the state evolution trend analysis results for each operating unit. For example, the state evolution trend analysis results for the transformer unit include voltage, current, and temperature change trends for each sub-period within the future time period; the state evolution trend analysis results for the circuit breaker unit include current and switch status change trends for each sub-period within the future time period. Through this integration, the state evolution trend analysis results for each operating unit are generated.

[0091] Step S134: Generate a predicted state description for each operating unit in subsequent time periods based on the evolution trend analysis results.

[0092] In this embodiment, the predicted state description is obtained by converting the numerical information from the state evolution trend analysis results into natural language descriptions. For example, if the state evolution trend analysis result of a transformer unit shows that the voltage will rise slightly, the current will remain stable, and the temperature will drop slightly in the future sub-period, then the predicted state description for the transformer unit is: voltage rises slightly, current remains stable, and temperature drops slightly in the future sub-period. If the state evolution trend analysis result of a circuit breaker unit shows that the current will increase slightly and the switch state will remain closed in the future sub-period, then the predicted state description for the circuit breaker unit is: current increases slightly and the switch state remains closed in the future sub-period. These predicted state descriptions are arranged according to the identifier of the operating unit and the predicted time period to generate the predicted state description for each operating unit in subsequent time periods.

[0093] Step S135: Arrange the initial state description of the historical data collection period and the predicted state description of the subsequent period in chronological order.

[0094] In this embodiment, the initial state description of the historical acquisition period refers to the already acquired state description of the operating units, while the predicted state description of subsequent periods refers to the state description of future periods generated by the state prediction module. Arranging these state descriptions in chronological order means arranging them according to the order of acquisition or prediction, for example, first arranging the initial state descriptions of the historical acquisition period, and then arranging the predicted state descriptions of subsequent periods. This forms a continuous sequence of state descriptions, reflecting the changing process of the power distribution system's operating state.

[0095] Step S136: Perform a coherence check on the arranged state descriptions and remove state descriptions with logical breaks.

[0096] In this embodiment, the coherence verification process checks whether there are logical contradictions or breaks in the arranged state description sequence. For example, it checks whether the time interval between two adjacent state descriptions is reasonable, whether the parameter changes in the state descriptions conform to physical laws, and whether there are abrupt changes in the running unit states within the state descriptions. If a state description with a logical break is found, such as an excessively long time interval between two adjacent state descriptions or parameter changes in the state description that do not conform to physical laws, then that state description is removed from the sequence. Through this verification process, the coherence and rationality of the state description sequence are ensured.

[0097] Step S137: Integrate the verified state description sequence to generate a running state sequence within a continuous time period.

[0098] In this embodiment, integration involves arranging the verified state description sequence in chronological order to form a complete operating state sequence. Each element in the operating state sequence contains information such as the operating unit identifier, state description, and time identifier. For example, the first element of the operating state sequence might be a transformer unit; during the historical data collection period, the voltage amplitude was within the normal range, the current amplitude was within the normal range, and the temperature was normal. The second element might also be a transformer unit; during the subsequent prediction period, the voltage might slightly increase, the current might remain stable, and the temperature might slightly decrease. Through this integration, an operating state sequence over a continuous time period is generated.

[0099] Step S140: Perform anomaly feature tracing analysis on the running state sequence to determine the running links and associated features corresponding to the abnormal states.

[0100] In this embodiment, the abnormal feature tracing analysis is performed by calling a pre-trained abnormal identification model to process the running state sequence, identify the abnormal state description, and then analyze the features corresponding to the abnormal state description to trace the source and associated features of the abnormal state.

[0101] Step S141: Call the pre-trained anomaly recognition model to perform time-by-time anomaly detection processing on the running state sequence and mark the state descriptions where anomalies exist.

[0102] In this embodiment, the pre-trained anomaly recognition model is built based on machine learning algorithms and includes a feature extraction module and an anomaly detection module. This model is pre-trained using a large amount of data on normal and abnormal operating states of the power distribution system, learning the differences between normal and abnormal states. When the model is invoked, the operating state sequence is input into the model, which processes it time-by-time, extracting feature information from the state description of each time period. This information is then compared with the pre-trained normal state features to determine whether an anomaly exists in the state description for that time period. If an anomaly is found, the state description is marked.

[0103] Step S142: Extract the feature information corresponding to the marked abnormal state description and generate an abnormal feature set.

[0104] In this embodiment, the labeled abnormal state description refers to the state description marked as abnormal by the abnormality identification model. Feature information, such as voltage amplitude, current amplitude, temperature, and switch status, is extracted from these state descriptions and integrated to generate an abnormal feature set. Each element in the abnormal feature set contains information such as feature name, feature value, and abnormality type. For example, voltage amplitude: too high, abnormality type: exceeding rated value; temperature: too high, abnormality type: exceeding warning value, etc.

[0105] Step S143: Perform feature correlation analysis on the abnormal feature set to determine the interaction relationship between abnormal features.

[0106] In this embodiment, feature correlation analysis calculates the correlation coefficient between any two features in the set of anomalous features, and determines their interaction relationship based on the magnitude of the correlation coefficient. For example, it calculates the correlation coefficient between anomaly features in voltage amplitude and current amplitude, and between anomaly features in temperature and power. The larger the absolute value of the correlation coefficient, the stronger the interaction relationship between the two anomalous features. Simultaneously, it analyzes the causal relationships between the anomalous features to determine which anomalous features are causes and which are consequences.

[0107] Step S144: Construct an abnormal feature propagation path model based on the interaction relationship, and trace the initial generation features of the abnormal features.

[0108] In this embodiment, the anomaly propagation path model is a directed graph model, where nodes represent anomalies and edges represent the propagation direction and interaction relationships between anomalies. When constructing this model, the direction and weight of the edges between nodes are determined based on the interaction relationships between anomalies. The direction of the edge represents the propagation direction of the anomaly, and the weight represents the strength of the interaction relationship. For example, if an anomaly in voltage amplitude is the cause of an anomaly in current amplitude, then the direction of the edge pointing from the node with the anomaly in voltage amplitude to the node with the anomaly in current amplitude is positive, and the weight is the absolute value of the correlation coefficient. Through this construction, an anomaly propagation path model is formed. Then, this model is analyzed to trace the initial generating feature of the anomaly, i.e., the earliest appearing anomaly.

[0109] Step S1441: Determine the propagation direction and propagation order between features based on the interaction relationship between abnormal features.

[0110] In this embodiment, the propagation direction is determined based on the causal relationship between abnormal features, with the causal feature pointing towards the result feature. The propagation order is determined based on the chronological order in which the abnormal features appear; the abnormal features that appear first propagate first, and the abnormal features that appear later propagate later. For example, if the voltage amplitude abnormal feature appears first, and then causes the current amplitude abnormal feature to appear, then the propagation direction is from the voltage amplitude abnormal feature to the current amplitude abnormal feature, and the propagation order is voltage amplitude abnormal feature first, current amplitude abnormal feature second.

[0111] Step S1442: Construct an anomaly feature propagation path graph with each anomaly feature as a node and the propagation direction as an edge.

[0112] In this embodiment, the anomaly propagation path graph is a directed graph, where each node represents an anomaly feature, and each edge represents the propagation direction between anomalies. For example, a voltage amplitude anomaly feature node is represented by a voltage anomaly, a current amplitude anomaly feature node by a current anomaly, and a temperature anomaly feature node by a temperature anomaly. If a voltage amplitude anomaly feature leads to a current amplitude anomaly feature, then the direction of the edge from the voltage anomaly node to the current anomaly node is positive; if a current amplitude anomaly feature leads to a temperature anomaly feature, then the direction of the edge from the current anomaly node to the temperature anomaly node is positive. This construction forms the anomaly propagation path graph.

[0113] Step S1443: Perform path traversal analysis on the propagation path graph to identify all possible propagation paths.

[0114] In this embodiment, path traversal analysis employs a depth-first search algorithm to traverse the propagation path graph. Starting from the initial node, the search proceeds along the edges until a node with no outgoing edges is reached. For example, starting from the voltage anomaly node, the search continues along the edges pointing to the current anomaly node, and then from the current anomaly node, the search continues along the edges pointing to the temperature anomaly node. This yields a propagation path: voltage anomaly → current anomaly → temperature anomaly. By fully traversing the propagation path graph, all possible propagation paths are identified.

[0115] Step S1444: Calculate the propagation probability of each propagation path and retain valid propagation paths whose propagation probabilities meet the preset conditions.

[0116] In this embodiment, the propagation probability is determined by calculating the product of the weights of each edge in the propagation path, where the weights represent the strength of the interaction. A preset condition is that the propagation probability is greater than or equal to a set threshold. Propagation paths with a propagation probability greater than or equal to the set threshold are retained as valid propagation paths, while propagation paths with a propagation probability less than the set threshold are discarded.

[0117] Step S1445: Perform reverse tracing analysis on the effective propagation path, tracing back from the terminal abnormal features to the initial features.

[0118] In this embodiment, the reverse tracing analysis starts from the terminal abnormal feature node of the effective propagation path and traces back along the opposite direction of the edges until a node with no incoming edges is reached. For example, if the effective propagation path is voltage anomaly → current anomaly → temperature anomaly, and the terminal abnormal feature node is temperature anomaly, tracing back along the opposite direction of the edges, the current anomaly node is reached first, followed by the voltage anomaly node. The voltage anomaly node has no incoming edges, so its initial feature is voltage anomaly. Through this reverse tracing analysis, the initial feature of each effective propagation path is determined.

[0119] Step S1446: During the tracing process, record the characteristic propagation time and propagation intensity of each path.

[0120] In this embodiment, feature propagation time refers to the time it takes for an abnormal feature to propagate from one node to another, and propagation intensity refers to the change in intensity of the abnormal feature during propagation. For example, in the propagation path of voltage abnormality → current abnormality → temperature abnormality, the time it takes for the voltage abnormality to propagate to the current abnormality is a set duration, and the propagation intensity is a set value; the time it takes for the current abnormality to propagate to the temperature abnormality is a set duration, and the propagation intensity is a set value.

[0121] Step S1447: Determine the earliest appearing abnormal feature based on the propagation time and propagation intensity, and use the earliest appearing abnormal feature as the initial generated feature.

[0122] In this embodiment, the earliest appearing abnormal feature refers to the abnormal feature with the earliest propagation time. By comparing the feature propagation time on each effective propagation path, the abnormal feature with the earliest propagation time is identified and used as the initial generated feature. For example, in the propagation path of voltage abnormality → current abnormality → temperature abnormality, the voltage abnormality has the earliest propagation time, so the initial generated feature is the voltage abnormality.

[0123] Step S145: Compare and match the initially generated features with the feature library of the power distribution system's operating links to determine the corresponding operating links.

[0124] In this embodiment, the power distribution system's operational feature library is a database storing feature information for each operational stage of the power distribution system, including generation, transmission, substation, distribution, and consumption stages. Each operational stage's feature information includes its name, function, involved equipment type, and characteristic parameters. The comparison and matching process compares the initially generated features with the feature information in the operational feature library to find the operational stage with the highest matching degree. The matching degree is determined by comparing the similarity between the feature parameters of the initially generated features and the feature parameters in the operational feature library; the higher the similarity, the higher the matching degree.

[0125] Step S146: Extract all related features of the corresponding operation process. The related features include direct related features and indirect related features.

[0126] In this embodiment, directly related features refer to features directly related to the initial generated feature. For example, if the initial generated feature is an abnormal voltage amplitude, then directly related features include current amplitude features, power features, etc. Indirectly related features refer to features indirectly related to the initial generated feature, such as temperature features, switching status features, etc. When extracting these related features, all features related to the corresponding operating segment are determined based on the information in the operating segment feature library, and then these features are divided into directly related features and indirectly related features.

[0127] Step S147: Prioritize the directly related features and indirectly related features to generate a list of operational processes and related features corresponding to the abnormal state.

[0128] In this embodiment, the priority ranking is determined based on the correlation strength between the associated features and the initially generated features; the higher the correlation strength, the higher the priority. For example, directly associated features have a higher correlation strength with the initially generated features and thus a higher priority than indirectly associated features. Among directly associated features, features with high correlation strength have a higher priority than features with low correlation strength. The ranked associated features are then integrated with the corresponding operational steps to generate a list of operational steps and associated features corresponding to the abnormal state.

[0129] Step S150: Based on the operational links and associated characteristics corresponding to the abnormal state, generate a power distribution system operation status monitoring report containing status identifiers and associated characteristics.

[0130] In this embodiment, the power distribution system operation status monitoring report is a comprehensive description of the power distribution system's operation status, including basic information on abnormal states, corresponding operational processes, associated characteristics, and handling suggestions. When generating the report, the information from the list of operational processes and associated characteristics corresponding to the abnormal states is organized and analyzed to form the various parts of the report.

[0131] For example, step S151: Extract the operational process identifiers and associated feature lists corresponding to the abnormal status, and determine the core content framework of the monitoring report.

[0132] In this embodiment, the operation link identifier refers to the unique identifier of the corresponding operation link. For example, the identifier for the substation link is the substation link identifier, and the identifier for the distribution link is the distribution link identifier, etc. The associated feature list refers to a list containing directly associated features and indirectly associated features. Determining the core content framework involves using this information to determine the main parts of the monitoring report, such as an overview of abnormal conditions, operation link analysis, associated feature analysis, and handling suggestions, etc.

[0133] Step S152: Classify and describe the direct and indirect related features in the list of related features, describing the attributes and functions of the direct related features and the attributes and functions of the indirect related features.

[0134] In this embodiment, the classification description separates directly related features and indirectly related features. For directly related features, the description includes their feature name, feature value, anomaly type, correlation strength with the initially generated feature, and their role in the generation and propagation of the abnormal state. For indirectly related features, the description includes their feature name, feature value, correlation strength with the initially generated feature, and their impact on the abnormal state. For example, the attributes of the directly related feature, current amplitude anomaly, include feature name: current amplitude, feature value: too high, anomaly type: exceeds rated value, and correlation strength: set value. Its function is to serve as a result of the initially generated feature, voltage amplitude anomaly, further leading to temperature anomalies. The attributes of the indirectly related feature, temperature anomaly, include feature name: temperature, feature value: too high, and correlation strength: set value. Its function is to reflect the impact of the current amplitude anomaly on the equipment.

[0135] Step S153: Assign a unique status identifier to each abnormal state, so that the status identifier corresponds one-to-one with the operation link and related characteristics.

[0136] In this embodiment, the status identifier is a unique string. When assigning status identifiers, it is ensured that each abnormal state has a unique identifier, and that this identifier corresponds one-to-one with the corresponding operating link and associated features. For example, the status identifier for an abnormal state with abnormal voltage amplitude is the abnormal state identifier, the corresponding operating link is the substation link, and the associated features are abnormal current amplitude, abnormal temperature, etc.

[0137] Step S154: Organize and arrange the status identifiers, operational process descriptions, and related feature classifications according to the preset report structure.

[0138] In this embodiment, the preset report structure includes a cover, table of contents, an overview of abnormal states, analysis of operational processes, analysis of related features, handling suggestions, and an appendix. The status identifiers, operational process descriptions, and related feature classifications are organized according to this structure. For example, the status identifiers and overview of abnormal states are placed at the beginning of the report, the operational process descriptions and related feature analyses are placed in the middle, and the handling suggestions are placed at the end.

[0139] Step S155: Supplement key analysis node information in the abnormal feature tracing analysis process and improve the report content.

[0140] In this embodiment, key analysis node information refers to node information that is significant in the process of anomaly feature tracing and analysis, such as the initial generation feature, key nodes in the propagation path, and the propagation time and intensity of the anomaly feature. Adding this information can enrich and detail the report, helping readers better understand the generation and propagation process of the anomaly. For example, information on the initial generation feature can be added to the anomaly status overview section of the report, information on key nodes in the propagation path can be added to the operational phase analysis section, and information on the propagation time and intensity of the anomaly feature can be added to the correlation feature analysis section.

[0141] Step S156: Perform a logical coherence check on the report content and adjust the order of the content to achieve logical clarity.

[0142] In this embodiment, the logical coherence check examines whether the report content is arranged in a logical order and whether there are any duplicate or contradictory contents. For example, it checks whether the abnormal status overview section corresponds to the operational process analysis section and the related feature analysis section, and whether the handling suggestions section is based on the previous analysis results. If any logical inconsistencies are found, the order of the content is adjusted to make the logic clearer.

[0143] Step S157: Generate a power distribution system operation status monitoring report that includes status identifiers, operation link descriptions, associated feature classification descriptions, and key analysis node information.

[0144] In this embodiment, report generation involves converting the organized and analyzed information into a standard report format. The report includes status indicators, operational process descriptions, related feature classifications, key analysis node information, and processing recommendations. Each section has a clear title and detailed content. Through this generation process, a complete power distribution system operation status monitoring report is formed.

[0145] Figure 2 The illustration shows exemplary hardware and software components of a power distribution system operation status monitoring system 100 based on big data analysis, which can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used in the power distribution system operation status monitoring system 100 based on big data analysis and to perform the functions in this application.

[0146] The power distribution system operation status monitoring system 100 based on big data analysis can be a general-purpose server or a special-purpose server; both can be used to implement the power distribution system operation status monitoring method based on big data analysis of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the load.

[0147] For example, the power distribution system operation status monitoring system 100 based on big data analytics may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the power distribution system operation status monitoring system 100 based on big data analytics may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The power distribution system operation status monitoring system 100 based on big data analytics also includes an I / O interface 150 between the computer and other input / output devices.

[0148] For ease of explanation, only one processor is described in the power distribution system operation status monitoring system 100 based on big data analysis. However, it should be noted that the power distribution system operation status monitoring system 100 based on big data analysis in this application may also include multiple processors. Therefore, the steps executed by one processor described in this application may also be executed jointly by multiple processors or individually. For example, if the processor of the power distribution system operation status monitoring system 100 based on big data analysis executes steps A and B, it should be understood that steps A and B may also be executed jointly by two different processors or individually by one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor jointly execute steps A and B.

[0149] Furthermore, this embodiment of the invention also provides a readable storage medium, which has computer-executable instructions pre-set in it. When the processor executes the computer-executable instructions, the above-mentioned method for monitoring the operating status of a power distribution system based on big data analysis is implemented.

[0150] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for monitoring the operating status of a power distribution system based on big data analysis, characterized in that, The method includes: Collect multi-source operation data generated during the operation of the power distribution system, construct a correlation system of power distribution system operation characteristics, and obtain a feature correlation map; The pre-trained state mapping model is invoked to perform hierarchical state parsing on the feature association map, generating initial state descriptions for each operating unit of the power distribution system; The initial state description is input into the time-series state evolution model to generate a sequence of running states over a continuous time period; Perform anomaly feature tracing analysis on the operation status sequence to determine the operation links and related features corresponding to the abnormal states; Based on the operational processes and associated characteristics corresponding to abnormal states, a power distribution system operation status monitoring report containing status identifiers and associated characteristics is generated.

2. The method for monitoring the operation status of a power distribution system based on big data analysis according to claim 1, characterized in that, The system collects multi-source operational data generated during the operation of the power distribution system, constructs a feature association system for the power distribution system operation, and obtains a feature association map, including: Multi-source operational data is categorized by feature type to obtain feature subsets for different operational dimensions; Extract the correlation strength information between features within each feature subset to generate a feature correlation matrix; Based on the association strength information in the feature association matrix, weakly associated features are eliminated, and a set of strongly associated features is retained. Hierarchical classification of strongly correlated feature sets is performed to determine the hierarchical relationships between features; Construct a tree-structured feature association system based on hierarchical relationships, and generate a feature association map containing feature hierarchy and association relationships; The feature nodes in the feature association graph are uniquely identified to ensure that the source of the running data corresponding to each feature node is traceable. The identified feature nodes are visualized and topologically constructed according to their hierarchical relationships to form a structured feature association graph.

3. The method for monitoring the operation status of a power distribution system based on big data analysis according to claim 1, characterized in that, The process of calling the pre-trained state mapping model to perform hierarchical state parsing on the feature association graph generates initial state descriptions for each operating unit of the power distribution system, including: The feature association graph is input into the feature input layer of the state mapping model, and the hierarchical features in the feature association graph are subjected to dimensionality transformation. The hierarchical parsing module of the state mapping model performs top-down layer-by-layer parsing on the transformed hierarchical features to extract the state representation information corresponding to each level feature. The unit matching module of the state mapping model is used to associate and match the state representation information of each level with the preset running unit feature library; Based on the matching results, determine the feature combination corresponding to each running unit and generate a list of running unit features; The features in the feature list of the running unit are processed by state semantic transformation, which transforms the feature information into a describable state language; By combining the features of each operating unit with the state language, an initial state description is generated that includes the operating unit identifier and the corresponding state description.

4. The method for monitoring the operation status of a power distribution system based on big data analysis according to claim 1, characterized in that, The step of inputting the initial state description into the time-series state evolution model to generate a sequence of operating states over a continuous time period includes: The initial state description is marked with a time dimension, and a corresponding collection period identifier is added to the state description of each running unit. The initial state description with the acquisition period identifier is input into the temporal coding module of the temporal state evolution model to generate a temporal feature vector; The state prediction module of the time-series state evolution model is used to analyze the state evolution trend of the time-series feature vector over continuous periods. Based on the evolution trend analysis results, a predicted state description for each operating unit in subsequent periods is generated; The initial state descriptions of historical data collection periods and the predicted state descriptions of subsequent periods are arranged in chronological order. Perform a coherence check on the arranged state descriptions and remove state descriptions with logical breaks. The verified state description sequences are integrated to generate a running state sequence over a continuous time period.

5. The method for monitoring the operation status of a power distribution system based on big data analysis according to claim 1, characterized in that, The step of performing anomaly feature tracing analysis on the operating state sequence to determine the operating links and associated features corresponding to the abnormal states includes: The pre-trained anomaly detection model is invoked to perform time-by-time anomaly detection processing on the running state sequence, and the state descriptions with anomalies are marked. Extract the feature information corresponding to the labeled abnormal state descriptions to generate an abnormal feature set; Perform feature correlation analysis on the set of abnormal features to determine the interaction relationships between abnormal features; An anomaly propagation path model is constructed based on interaction relationships to trace the initial generation characteristics of anomalies. The initial generated features are compared and matched with the feature library of the power distribution system's operating links to determine the corresponding operating links; Extract all relevant features related to the corresponding operational process. These features include both direct and indirect features. Prioritize directly related features and indirectly related features to generate a list of operational processes and related features corresponding to abnormal states.

6. The method for monitoring the operation status of a power distribution system based on big data analysis according to claim 2, characterized in that, The process of classifying multi-source operational data into feature categories to obtain feature subsets of different operational dimensions includes: Extract feature attribute information from multi-source operational data and determine the descriptive parameters for each feature; Based on the description parameters, a feature classification standard is constructed to divide the features into different operational dimension categories; The features in the multi-source operational data are grouped according to the operational dimension category to form initial feature groups; Perform feature redundancy detection on the initial feature set and delete duplicate feature data; Calculate the distribution density of features within each initial feature group, and merge feature groups with excessively low distribution densities. The merged feature groups are named and labeled with the corresponding running dimension for each feature group; The identified feature groups are used as feature subsets of different operating dimensions, so that the features in each feature subset have the same operating dimension attributes.

7. The method for monitoring the operation status of a power distribution system based on big data analysis according to claim 3, characterized in that, The hierarchical parsing module of the state mapping model performs top-down, layer-by-layer parsing of the transformed hierarchical features to extract the state representation information corresponding to each level feature, including: Starting from the top-level features of the feature association graph, the transformed hierarchical features are initially analyzed to extract the core attribute information of the top-level features; Based on the core attribute information of the top-level features, the analysis is extended to the next level of features to determine the relationship between the next level of features and the top-level features. According to the determined association method, the attributes of the next level features are extracted to generate the feature attribute set of the current level; For the next level of features, the attribute set of the previous level features is extended and analyzed to determine the association between the next level of features and the previous level features. The attributes of the next level of features are extracted according to the determined association, and the feature attribute set of the current level is generated. For each feature in the feature association graph except for the top-level feature, the next level feature, and the next level feature, the extension analysis is performed based on the attribute set of the previous level feature of the corresponding level feature to determine the association method between the corresponding level feature and the previous level feature. According to the determined association method, the attributes of the corresponding level feature are extracted to generate the feature attribute set of the corresponding level feature. During the analysis process, the correlation strength and correlation type between each level feature and the previous level feature are recorded; Integrate the feature attribute set of each level with the corresponding association strength and association type; Generate state representation information that includes feature attributes, association strength, and association type at each level.

8. The method for monitoring the operation status of a power distribution system based on big data analysis according to claim 4, characterized in that, The state prediction module using the temporal state evolution model performs continuous time-period state evolution trend analysis on the temporal feature vector, including: The time series feature vector is input into the trend extraction unit of the state prediction module to extract the changing trend features in the time series feature vector; A time-series evolution model is constructed based on the characteristics of changing trends, and the evolution analysis window of the model is set. Within the evolutionary analysis window, a sliding analysis is performed on the time-series feature vectors to capture the changing patterns of the feature vectors; Determine the rate and direction of change of the feature vector in different time periods based on the pattern of change; By combining the rate and direction of change, the possible range of change of the feature vector in subsequent time periods can be predicted; The possible range of changes is compared with the preset state evolution threshold to filter out the change trends that conform to the evolution logic; The selected trends are integrated to generate state evolution trend analysis results for each operating unit.

9. The method for monitoring the operation status of a power distribution system based on big data analysis according to claim 5, characterized in that, The aforementioned model for constructing anomaly propagation paths based on interaction relationships, tracing the initial generation characteristics of anomalies, includes: Based on the interaction relationships between abnormal features, determine the propagation direction and propagation order between features; Construct an anomaly propagation path graph with each anomaly feature as a node and the propagation direction as an edge; Perform path traversal analysis on the propagation path graph to identify all possible propagation paths; Calculate the propagation probability of each propagation path and retain valid propagation paths whose propagation probabilities meet preset conditions; Perform reverse tracing analysis on the effective propagation path, tracing back from the terminal abnormal characteristics to the initial characteristics; During the tracing process, the characteristic propagation time and propagation intensity of each path are recorded; The earliest appearing anomalous feature is determined based on the propagation time and propagation intensity, and the earliest appearing anomalous feature is used as the initial generated feature.

10. A power distribution system operation status monitoring system based on big data analysis, characterized in that, The power distribution system operation status monitoring system based on big data analysis includes a processor and a memory, the memory and the processor are connected, the memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the power distribution system operation status monitoring method based on big data analysis as described in any one of claims 1-9.

Images