Power big data adaptive management method and system fusing space-time feature mapping

Abstract

The application relates to the field of electric power data management, and discloses a power big data adaptive management method and system fusing space-time feature mapping, the method comprising the following steps: acquiring real-time operation data streams from multiple power terminal sources, and constructing an original power data set with time sequence labels and equipment identifiers; based on an edge storage and edge calculation architecture, dividing the original power data set into parallel processing units, adopting event-driven index mapping rules to dynamically update indexes, and constructing a multi-dimensional data index cache system with real-time responsiveness; identifying key abnormal trajectories through a time-varying feature nesting mechanism, and using a streaming feature aggregation network to hierarchically measure the data fluctuation degree in the key abnormal trajectories and analyze the entropy value disturbance; combining the structural matching degree between the historical abnormal event evolution graph and the key abnormal trajectories to identify potential security risk nodes; and generating a multi-level response instruction chain based on the risk assessment results. The application has the advantage of improving the safety of power grid operation.

Description

Technical Field

[0001] This invention relates to the field of power data management, specifically to an adaptive management method and system for power big data that integrates spatiotemporal feature mapping. Background Technology

[0002] Currently, power companies commonly use smart meters to collect electricity consumption data from residential communities and upload it to a power big data management platform via concentrators or wireless communication networks. Existing platforms typically rely on relational databases or big data batch processing frameworks such as Hadoop and Spark to periodically summarize and analyze the collected electricity data. However, with the increasing frequency of smart meter data collection and the rapid growth in the number of terminals, the daily data volume accessed by the platform is growing exponentially, and existing systems suffer from significant batch processing latency issues. For example, when a community experiences a sudden surge in power consumption due to aging lines or wiring faults, although smart meters can report abnormal electricity consumption data within minutes, existing big data systems need to wait for all the data to be written to the storage cluster before offline analysis, typically taking at least several hours to complete anomaly detection and trigger an alarm. This delayed detection method not only increases the risk of damage to power grid equipment but may also lead to the failure to detect illegal activities such as electricity theft in a timely manner, seriously affecting the safety and economy of power grid operation. Most existing systems lack real-time streaming computing capabilities, making it impossible to perform online analysis while data is being written. This results in coarse-grained anomaly detection, only able to statistically analyze power consumption fluctuations at the hourly or daily level, and unable to achieve fine-grained anomaly identification at the second or minute level. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides an adaptive management method for power big data that integrates spatiotemporal feature mapping, which has the advantage of improving the safety of power grid operation and solves the problems mentioned in the background technology.

[0004] This invention provides the following technical solution: an adaptive management method for power big data that integrates spatiotemporal feature mapping, comprising the following steps: Real-time operation data streams from multiple power terminals are acquired, and a raw power dataset with time-series labels and device identifiers is constructed by combining a time window reorganization mechanism and a heterogeneous decoding strategy at the protocol layer. Based on the edge-to-grid architecture, the original power dataset is divided into parallel processing units, and event-driven index mapping rules are used for dynamic index updates to build a multi-dimensional data index caching system with real-time responsiveness. Key abnormal trajectories are identified through a time-varying feature nesting mechanism, and the data fluctuation degree in key abnormal trajectories is measured hierarchically and analyzed for entropy perturbation using a streaming feature aggregation network. By combining the structural matching degree between the historical anomaly event evolution map and the key anomaly trajectory, potential security risk nodes are identified, and the risk assessment results are fine-tuned in reverse through a spatiotemporal collaborative correction strategy. Based on the risk assessment results, a multi-level response command chain is generated to drive the power grid dispatch control terminal and the front-end operation and maintenance nodes to carry out hierarchical linkage response.

[0005] Preferably, the process of constructing the original power dataset with time-series labels and device identifiers is as follows: Data streams are collected from power terminals of different voltage levels, equipment types, and communication protocols; Based on the heterogeneous decoding strategy at the protocol layer, multiple communication formats are uniformly extracted; A time window reassembly mechanism is introduced to perform time-series alignment and reconstruction of power data in cases of cross-time periods or data packet loss; By combining the unique identifier of the terminal device, the data stream is labeled with its ownership and pre-attached with an index to construct the original power dataset.

[0006] Preferably, the dynamic index update process using event-driven metric mapping rules is as follows: Construct an event triggering rule set based on the changing trends of key indicators in the original power dataset; When any indicator fluctuates drastically or exceeds a threshold, the index update scheduling mechanism is triggered. Based on the timestamp of the event and the device identifier, the data partitioning indexing strategy is adjusted in real time to establish a multi-level index mapping relationship; A lightweight cache synchronization mechanism is used to trigger the update process of metrics and response indexes.

[0007] Preferably, the process of constructing a multi-dimensional data index caching system with real-time responsiveness is as follows: The index mapping relationship generated under event-driven conditions is cached in the memory of edge nodes, and a two-dimensional index cache structure is constructed based on the time window and device load status. A partition heat awareness mechanism is introduced to dynamically adjust cache priority based on data access frequency and the probability of indicator mutations. A sliding window incremental update algorithm is used to periodically refresh and evict index entries in the cache; A multi-dimensional data index caching system is constructed by combining the concurrency of data query requests and the distribution of access paths.

[0008] Preferably, the process of identifying key abnormal trajectories through a time-varying feature nesting mechanism is as follows: Decouple the time-series data in the cache according to the device dimension and construct a multi-dimensional data structure with time as the main axis and metrics as nested layers; Time series variation detection is performed on each nested layer of indicators to identify evolutionary trajectories with nonlinear mutations, periodic misalignments, or trend drift characteristics; A time-varying feature nesting mechanism is introduced to compare the similarities and differences between trajectories in different time periods and identify key abnormal trajectories.

[0009] Preferably, the process of hierarchically measuring and analyzing the entropy disturbance of data fluctuations in key anomaly trajectories is as follows: Key anomaly trajectories are grouped into multiple levels according to their fluctuation characteristics, including a transient abrupt change layer, a sustained gradual change layer, and a composite anomaly layer; Within each layer, the short-term mean fluctuation range, maximum deviation, and gradient change rate of the indicator values ​​are calculated to assess the intensity of entropy changes and the time span of the indicator distribution in each layer.

[0010] Preferably, the process for identifying nodes with potential security risks is as follows: Extract indicator segments with significant fluctuation intensity and prominent entropy disturbances from key abnormal trajectories, and construct feature vectors for abnormal trajectories; Extract typical abnormal evolution paths from historical abnormal event evolution maps and establish a structural matching model between map nodes and current abnormal trajectory features; By calculating the structural similarity and evolutionary trend deviation between the current abnormal trajectory and historical evolution nodes, potential security risk nodes can be identified.

[0011] Preferably, the process of fine-tuning the risk assessment results through a spatiotemporal collaborative correction strategy is as follows: Analyze the spatial topology and device interconnection relationships of identified potential risk nodes to determine their interference paths with surrounding nodes; Based on the spatiotemporal correlation coefficients between nodes, a risk backpropagation graph is constructed to dynamically compress and expand the risk level boundary. A correction feedback mechanism is introduced to adjust the risk assessment weight coefficients based on actual control feedback and equipment operating status.

[0012] Preferably, the hierarchical linkage response process between the power grid dispatch control terminal and the front-end operation and maintenance nodes is as follows: Based on the revised risk levels, event types are categorized, and a response level index is constructed. The scheduling strategy library is invoked to generate control-end scheduling instructions based on risk type and location, including load transfer, voltage and current limiting; Synchronously access the operation and maintenance strategy library and schedule front-end maintenance nodes to generate operation plans.

[0013] A big data-based power data management system includes: Data acquisition module: Receives real-time operation data streams from multiple power terminals and constructs a raw power dataset with time-series labels and device identifiers through a time window reassembly mechanism and a heterogeneous decoding strategy at the protocol layer; Index building module: The original power dataset is divided into parallel processing units, and event-driven index mapping rules are used for dynamic index updates to build a multi-dimensional data index caching system with real-time response capabilities; Feature Analysis Module: Identifies key abnormal trajectories through a time-varying feature nesting mechanism, and uses a streaming feature aggregation network to perform hierarchical measurement and entropy perturbation analysis on the degree of data fluctuation. Risk Correction Module: By combining the structural matching degree between the historical anomaly evolution map and the key anomaly trajectory, potential safety risk nodes are identified, and the risk assessment results are fine-tuned in reverse based on the spatiotemporal collaborative correction strategy. Response generation module: Generates a multi-level response instruction chain based on risk assessment results, driving the power grid dispatch control terminal and the front-end operation and maintenance nodes to perform hierarchical linkage response.

[0014] In its operation, this invention effectively improves the real-time performance of data access and query response speed through a storage-on-demand architecture and an event-driven index update mechanism, overcoming the problems of high response latency and lagging index updates in traditional batch processing architectures. It introduces a time-varying feature nesting mechanism and a streaming feature aggregation network, enabling the system to keenly identify multiple types of key anomaly trajectories, particularly suitable for sensing and modeling sudden power disturbances, slowly varying load drifts, and trending equipment anomalies. By combining historical anomaly event evolution maps with structural matching analysis and constructing a backpropagation mechanism through a spatiotemporal collaborative correction strategy, it significantly improves the accuracy of risk identification and the dynamic adaptability of risk assessment. Based on the identification results, it generates a hierarchical linkage response command chain, which can efficiently mobilize resources at the scheduling and operation and maintenance ends, improving the timeliness, targeting, and system linkage coordination capabilities of the response, thereby effectively ensuring the safety, stability, and intelligence level of power grid operation. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the method of the present invention; Figure 2 This is a schematic diagram of the system of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1: Please refer to Figure 1 As shown in the embodiment of the present invention, an adaptive management method for power big data that integrates spatiotemporal feature mapping includes the following steps: S1: Obtain real-time operation data streams from multiple power terminals, and construct an original power dataset with time-series labels and device identifiers by combining a time window reorganization mechanism and a heterogeneous decoding strategy at the protocol layer.

[0018] The process of constructing the original power dataset with time-series labels and device identifiers in S1 is as follows: Data streams from power terminals of different voltage levels, equipment types, and communication protocols are collected. Data acquisition agents are deployed in distribution automation systems, substation monitoring systems, and user-side smart terminals to monitor raw data streams output by devices such as low-voltage data acquisition units, feeder terminals, transformer monitoring devices, and smart meters in real time. Data stream acquisition adopts an asynchronous communication mechanism, which aggregates and uploads data through edge gateways or data relay nodes, and labels the device source information according to voltage level, equipment function, and communication protocol type. Based on a heterogeneous decoding strategy at the protocol layer, multiple communication formats are uniformly extracted. A protocol identification model is established to address the differences in communication protocols among various terminal devices, automatically determining the protocol type of received data packets. According to the determination results, the corresponding protocol parsing plugin or decoding template is loaded, and key parameters such as voltage, current, power, and frequency are extracted according to field mapping rules. For protocols with frame structure check bits, the frame header / tail identifiers and data length are checked first to ensure data integrity before parsing. All parsed parameters are uniformly converted into standardized key-value pair structures and stored in an intermediate buffer for timing reconstruction and labeling. A time window reassembly mechanism is introduced to perform time-series alignment and reconstruction of power data in cases of cross-time periods or packet loss. The decoded raw parameter data is grouped and processed by terminal device according to a fixed sliding time window. If incomplete data or breakpoints are found in a certain window (such as two consecutive missing sampling periods), compensation and reassembly logic is triggered: on the one hand, short-term trend prediction interpolation is performed by combining the device's historical data curves; on the other hand, redundant data in the edge gateway's cache is queried to perform data replenishment and packet loss repair. Finally, the data is sorted in ascending order by timestamp to form a complete and continuous single-device time-series data segment, improving the data stream alignment accuracy. By combining the unique identifier of the terminal device, the data stream is labeled with its attribution and pre-attached with an index, thus constructing an original power dataset that has both temporal continuity and device uniqueness.

[0019] S2: Based on the edge-to-memory architecture, the original power dataset is divided into parallel processing units. Event-driven index mapping rules are used for dynamic index updates, and a multi-dimensional data index caching system with real-time responsiveness is constructed.

[0020] The dynamic index update process using event-driven index mapping rules in S2 is as follows: An event triggering rule set is constructed based on the changing trends of key indicators in the original power dataset. Typical operating indicators (including three-phase voltage, three-phase current, active power, reactive power, power factor, and frequency) are selected from the original power dataset. Based on historical data statistics and expert experience, change thresholds for each type of indicator are defined. A trend detection algorithm is introduced to identify abrupt changes, abnormal fluctuations, or continuous drift in the indicators. For each type of indicator change, an event triggering rule is generated, which includes the indicator name, triggering conditions, triggering frequency, and bound index update strategy. Finally, an event triggering rule set is constructed to drive index updates in real time. When any metric experiences drastic fluctuations or exceeds a threshold, the index update scheduling mechanism is triggered. During data ingestion or streaming, an event rule matching operation is performed in real time for each sampled data to determine whether any metric fluctuation trigger condition is met. Once an event trigger rule is met, an event trigger request containing the device number, metric name, trigger time, and fluctuation type is constructed. An asynchronous index update task request is sent to the index scheduling module, and the data index status corresponding to the device in the current cache is suspended to avoid concurrent conflicts. The scheduling mechanism supports event frequency limit control to prevent the same device from triggering too many index reconstruction operations in a short period of time, ensuring system stability. Based on the event timestamp and device identifier, the data partitioning index strategy is adjusted in real time to establish a multi-level index mapping relationship. After an event is triggered, the system reclassifies the corresponding data segment to the high-priority partition within the current processing cycle according to the event timestamp. Using the event trigger time as the anchor point, the three-level index path of device-indicator-time is reconstructed, and the anomaly level and fluctuation category are attached as additional tags. If the event is a cross-device impact type (such as the current change of multiple meters under the same transformer), a linkage index link is constructed to realize the joint index update of the device group. All index structures are synchronously stored in the memory cache and a temporary index snapshot is created in the backend storage. A lightweight cache synchronization mechanism is used to trigger the index update process for metrics and responses. The index update process employs a lightweight cache synchronization framework (such as a memory-level double buffer or a linked list-based write-ahead structure) to ensure that index updates do not block normal data inflow. After each update, a lock-free switch is achieved through "freezing the old index + activating the new index" to ensure the continuity of the query interface. The synchronization mechanism supports asynchronous disk flushing and batch merging strategies, enabling batch writing to backend storage when the system load is low, thereby improving overall write performance. After an index change, the index status change event is immediately reported to the upstream data scheduling module to drive the dynamic adjustment of the subsequent data flow processing path.

[0021] The process of constructing a multi-dimensional data index caching system with real-time responsiveness in S2 is as follows: The event-driven index mapping relationships are cached in the memory of the edge nodes, and a two-dimensional index cache structure is constructed based on the time window and device load status. The index update tasks triggered by the event-driven mechanism are deployed locally in cache containers on the edge computing nodes, and each index mapping relationship is stored in a structured manner according to "device identifier - event timestamp". According to the data generation frequency of different devices and their respective time windows, a two-dimensional index cache matrix is ​​built in memory with time as the row and device as the column. For high-load devices or devices with fluctuation warnings, higher refresh frequency and larger cache capacity are allocated to ensure the real-time integrity and fast access of index data. A partition heat awareness mechanism is introduced to dynamically adjust cache priority based on data access frequency and metric mutation probability. A cache heat monitor is set up to count the number of accesses, updates, and metric mutation triggers for each index partition within a unit of time. The heat value of each partition is calculated by weighting the above metrics, and the priority of that partition in the cache is adjusted accordingly. Partitions with high heat values ​​are given longer cache residency time, higher access prefetch frequency, and lower eviction probability. An asynchronous heat rollback mechanism is also supported. When a partition has not been accessed for a long time or the metric fluctuation stops, its heat value will automatically decay to release cache resources. A sliding window incremental update algorithm is adopted to periodically refresh and evict index entries in the cache. A sliding time window mechanism is introduced into the cache index area, grouping cache entries by time as the main dimension, with data within each time period forming a sliding window. Each sliding operation slides the oldest set of data out of the cache and slides the newest set of data in, ensuring that the cache content always retains the index information of the most recent N time windows. An incremental update strategy is executed on the index entries within the window, updating only entries whose indicator mutation fields, device status, or access frequency have changed, reducing system IO overhead. Sliding-out entries are selectively written to backend storage or directly discarded based on their popularity value and system load, avoiding invalid data from polluting the cache system. A multi-dimensional data index cache system is constructed by combining the concurrency of data query requests and the distribution of access paths.

[0022] S3: Identify key anomalous trajectories through a time-varying feature nesting mechanism, and use a streaming feature aggregation network to perform hierarchical measurement and entropy perturbation analysis on the data fluctuation degree in key anomalous trajectories.

[0023] The process of identifying key abnormal trajectories through the time-varying feature nesting mechanism in S3 is as follows: The time-series data in the cache is decoupled by device dimension, constructing a multi-dimensional data structure with time as the main axis and indicators as nested layers. The original power time-series data in the cache index system is traversed, and various types of data are categorized according to the unique identifier of the device, achieving logical separation at the device dimension. Within each device's data sequence, a main axis vector is constructed using the timestamp as the primary key, encapsulating multiple operating indicators under the same timestamp into nested vector layers. This ultimately forms a nested data structure of device → time series → indicator layers, allowing the operating status of each device to be parsed as a multi-dimensional time vector spanning multiple consecutive frames. This provides a foundation for subsequent time evolution modeling and multi-indicator coupling analysis, exhibiting good scalability and computational stability. Time series variation detection is performed on each nested index layer to identify evolutionary trajectories with nonlinear mutations, periodic misalignments, or trend drift characteristics. A multi-scale sliding window algorithm is used to perform local variation detection on each nested index sequence to determine whether nonlinear mutations, periodic misalignments, or trend drifts exist. For mutation-type trajectories, the change slope, root mean square error, and mutation intensity index are calculated. For periodic misalignment phenomena, autocorrelation function and fast Fourier transform are introduced to analyze periodic offsets. For trend drift-type indices, sliding regression and trend decomposition are applied to extract long-term offset signals. The detection results of each type are labeled as the basis for marking trajectory anomaly types to construct an index-level variation event identifier set. A time-varying feature nesting mechanism is introduced to compare the similarity and differences between trajectories in different time periods to identify key abnormal trajectories. Within a sliding time window, local segments of equipment indicator trajectories are extracted to construct a trajectory vector sequence composed of time periods, indicator combinations, and anomaly labels. Distance measurement and variability calculation are performed on the trajectory vector sequences of different time periods. A time-varying feature nesting structure is constructed to nest and encode the trajectory vectors of the same indicator in different time periods to capture the continuity and suddenness of its time-varying features. Trajectories that simultaneously exhibit high mutability, low similarity, or trend reversal in multiple dimensions are identified and marked as key abnormal trajectories, reflecting important signals of abnormal equipment status, operational fluctuations, or external disturbances, serving as direct input for subsequent risk node identification and response decisions.

[0024] The process of hierarchical measurement and entropy perturbation analysis of data fluctuation in key anomaly trajectories in S3 is as follows: Key anomaly trajectories are grouped into multiple levels based on their fluctuation characteristics, including a transient abrupt change layer, a sustained gradual change layer, and a composite anomaly layer. A sliding window slicing operation is performed on the key anomaly trajectories to extract the trend of indicator changes in each segment. For each trajectory segment, its first-order difference sequence and slope are calculated to determine whether it exhibits abrupt or gradual change characteristics. If the slope is greater than a set abrupt change threshold and the duration is short, it is classified as a transient abrupt change layer. If the slope is small but the duration of fluctuation is long, and the indicator shows a slow upward or downward trend, it is classified as a sustained gradual change layer. If multiple types of anomalies occur and overlap within the same time period (e.g., fluctuation + drift + misalignment), it is identified as a composite anomaly layer. Each grouping result is recorded as an anomaly layer label. Within each layer, the short-term mean fluctuation range, maximum deviation, and gradient rate of change of the indicator values ​​are calculated to assess the intensity of entropy changes and the time span of the indicator distribution in each layer; for the indicator data sequence within each anomaly layer, calculations are performed separately using a sliding window method: Short-term mean volatility: Local volatility is characterized by the ratio of the standard deviation σ to the mean μ within each window (i.e., the coefficient of variation CV = σ / μ). Maximum deviation: Based on historical normal benchmarks or expected value ranges, calculate the maximum exceeding limit or maximum deviation ratio in the trajectory; Gradient rate of change: The rate of change of the index per unit time is calculated by the first derivative. If there are multiple high-frequency slope change points, the acceleration information (second derivative) is further extracted. All measurement results are stored in a structured manner as a hierarchy-indicator-fluctuation feature matrix, providing quantitative support for the determination of anomaly intensity; Each index value in each layer is discretized into equally spaced intervals, and a frequency distribution histogram of the index is constructed. The entropy value of the index value distribution in each layer is calculated to measure the uncertainty or dispersion of the index in that layer. By comparing the entropy value curves in different time periods, a sudden increase in entropy value indicates that the index fluctuates violently or spreads abnormally, while a gradual decrease in entropy value may indicate abnormal convergence. At the same time, the duration of the abnormality in each layer (start-end timestamp) is recorded, and an abnormal disturbance impact index is generated by combining the entropy value disturbance trend.

[0025] S4: By combining the structural matching degree between the historical anomaly evolution map and the key anomaly trajectory, potential security risk nodes are identified, and the risk assessment results are fine-tuned in reverse through a spatiotemporal collaborative correction strategy.

[0026] The process of identifying potential security risk nodes in S4 is as follows: Extract indicator segments with significant fluctuation intensity and prominent entropy disturbances from key anomaly trajectories and construct anomaly trajectory feature vectors. In key anomaly trajectories, based on the entropy disturbance analysis results, locate indicator segments with significantly increased instantaneous volatility and abnormally amplified average deviation. Standardize these indicator segments, including time normalization, amplitude compression, and dimension unification. Extract the statistical features of the indicator segments within a local window using the sliding window method, and then construct anomaly trajectory feature vectors describing the fluctuation structure. Typical abnormal evolution paths are extracted from historical abnormal event evolution maps, and a structural matching model is established between map nodes and current abnormal trajectory features. Based on the historical power abnormal event database, a multi-source event map composed of equipment status, indicator evolution paths, and handling feedback is extracted. Based on the time series nodes in the map, the abnormal type, severity, and indicator pattern corresponding to the time series nodes are labeled. A structural vector based on map node attributes (such as fluctuation feature codes and evolution direction weights) is constructed. A similarity matching model is used to perform structural matching between the current abnormal trajectory vector and the historical map node vector, and a similarity score is assigned to each candidate node. By calculating the structural similarity and evolutionary trend deviation between the current abnormal trajectory and historical evolutionary nodes, potential safety risk nodes are identified. Based on the structural matching, an evolutionary trend deviation analysis mechanism is introduced to calculate the difference in trend direction between the current abnormal trajectory characteristics and subsequent nodes in each historical path (such as the direction of index variation, the direction of mutation slope, and the first-order derivative trend). If the current trajectory is highly consistent with a certain historical evolutionary path in multiple dimensions, and its development direction has obvious risk convergence (i.e., the trend deviation is small and points to high-risk state nodes), then the evolutionary node corresponding to that path is marked as a potential safety risk node.

[0027] The process of fine-tuning the risk assessment results through the spatiotemporal collaborative correction strategy in S4 is as follows: For identified potential risk nodes, spatial topology location and equipment interconnection relationship analysis are performed to determine their interference paths with surrounding nodes. The identified potential safety risk nodes are mapped to the spatial topology diagram of the power grid, and their physical locations, such as bus segments, substation units, and feeder ends, are extracted. Based on equipment connection relationships and topological adjacency matrices, the linkage relationship between the node and its surrounding related equipment (such as upstream transformers, parallel buses, shared circuit loads, etc.) is identified. The physical interference paths and logical coupling relationships between voltage levels are analyzed; for example, shared buses may lead to multi-device state linkage responses. A spatial impact propagation subgraph centered on the node is constructed. Based on the spatiotemporal correlation coefficient between nodes, a risk backpropagation graph is constructed to dynamically compress and expand the risk level boundary. From a temporal perspective, the indicator change curves of the risk node and its topologically adjacent nodes are analyzed, and temporal correlation is calculated using metrics such as Pearson correlation coefficient and mutual information. From a spatial perspective, a spatial influence factor matrix is ​​constructed by combining node physical distance, electrical coupling, and transmission delay. Combining the two, a multi-node spatiotemporal coordination matrix is ​​constructed to generate a risk backpropagation graph, reflecting the possibility and intensity of the risk backpropagation path. For nodes with high correlation and strong influence coupling on the backpropagation path, their initial risk level is dynamically expanded (risk upgraded); for nodes with strong isolation or broken paths, their risk level is compressed (risk downgraded), thereby refining the boundary range of the original assessment results. A corrective feedback mechanism is introduced to adjust the risk assessment weight coefficients based on actual control feedback and equipment operating status. Real-time feedback data from the dispatch control system and maintenance terminals is invoked, including the execution status of control commands (whether successful or delayed) and the equipment operating status after the response (whether stable or transferred to anomaly). The anomaly evolution path predicted in the risk assessment stage is compared with the actual handling results to calculate the risk prediction offset value. If the feedback shows that a node quickly returns to normal after handling, it indicates that the original assessment result was overestimated, and the risk weight of that node should be reduced. If the feedback shows that the handling is ineffective or further deteriorates, its weight needs to be increased and spread to related nodes for linkage correction. Finally, the assessment coefficients of each risk node are dynamically updated by introducing a time decay coefficient and feedback confidence level.

[0028] S5: Generates a multi-level response command chain based on risk assessment results, driving the power grid dispatch control terminal and the front-end operation and maintenance nodes to perform hierarchical linkage response.

[0029] The hierarchical linkage response process between the drive grid dispatch control terminal and the front-end operation and maintenance node in S5 is as follows: Based on the revised risk levels, event types are categorized, and a response level index is constructed. The system calls the scheduling strategy library to generate control-end scheduling instructions based on risk type and location, including load transfer, pressure limiting, and flow limiting; it also synchronously accesses the operation and maintenance strategy library to generate operation plans for the scheduling front-end maintenance nodes.

[0030] Example 2, as Figure 2 As shown, an adaptive management system for power big data that integrates spatiotemporal feature mapping includes: Data acquisition module: Receives real-time operation data streams from multiple power terminals and constructs a raw power dataset with time-series labels and device identifiers through a time window reassembly mechanism and a heterogeneous decoding strategy at the protocol layer; Index building module: The original power dataset is divided into parallel processing units, and event-driven index mapping rules are used for dynamic index updates to build a multi-dimensional data index caching system with real-time response capabilities; Feature Analysis Module: Identifies key abnormal trajectories through a time-varying feature nesting mechanism, and uses a streaming feature aggregation network to perform hierarchical measurement and entropy perturbation analysis on the degree of data fluctuation. Risk Correction Module: By combining the structural matching degree between the historical anomaly evolution map and the key anomaly trajectory, potential safety risk nodes are identified, and the risk assessment results are fine-tuned in reverse based on the spatiotemporal collaborative correction strategy. Response generation module: Generates a multi-level response instruction chain based on risk assessment results, driving the power grid dispatch control terminal and the front-end operation and maintenance nodes to perform hierarchical linkage response.

[0031] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0032] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An adaptive management method for power big data that integrates spatiotemporal feature mapping, characterized in that, include: Real-time operation data streams from multiple power terminals are acquired, and a raw power dataset with time-series labels and device identifiers is constructed by combining a time window reorganization mechanism and a heterogeneous decoding strategy at the protocol layer. Based on the edge-to-grid architecture, the original power dataset is divided into parallel processing units, and event-driven index mapping rules are used for dynamic index updates to build a multi-dimensional data index caching system with real-time responsiveness. Key abnormal trajectories are identified through a time-varying feature nesting mechanism, and the data fluctuation degree in key abnormal trajectories is measured hierarchically and analyzed for entropy perturbation using a streaming feature aggregation network. By combining the structural matching degree between the historical anomaly event evolution map and the key anomaly trajectory, potential security risk nodes are identified, and the risk assessment results are fine-tuned in reverse through a spatiotemporal collaborative correction strategy. Based on the risk assessment results, a multi-level response command chain is generated to drive the power grid dispatch control terminal and the front-end operation and maintenance node to perform hierarchical linkage response. The process of constructing the original power dataset with time-series labels and device identifiers is as follows: Data streams are collected from power terminals of different voltage levels, equipment types, and communication protocols; Based on the heterogeneous decoding strategy at the protocol layer, multiple communication formats are uniformly extracted; A time window reassembly mechanism is introduced to perform time-series alignment and reconstruction of power data in cases of cross-time periods or data packet loss; By combining the unique identifier of the terminal device, the data stream is labeled with its attribution and pre-attached with an index to construct the original power dataset; The dynamic index update process using event-driven metric mapping rules is as follows: Construct an event triggering rule set based on the changing trends of key indicators in the original power dataset; When any indicator fluctuates drastically or exceeds a threshold, the index update scheduling mechanism is triggered. Based on the timestamp of the event and the device identifier, the data partitioning indexing strategy is adjusted in real time to establish a multi-level index mapping relationship; A lightweight cache synchronization mechanism is used to trigger the update process of metrics and response indexes; The process of building a multidimensional data index caching system with real-time responsiveness is as follows: The index mapping relationship generated under event-driven conditions is cached in the memory of edge nodes, and a two-dimensional index cache structure is constructed based on the time window and device load status. A partition heat awareness mechanism is introduced to dynamically adjust cache priority based on data access frequency and the probability of indicator mutations. A sliding window incremental update algorithm is used to periodically refresh and evict index entries in the cache; A multi-dimensional data indexing and caching system is constructed by combining the concurrency of data query requests and the distribution of access paths; The process of hierarchical measurement and entropy perturbation analysis of the data fluctuation degree in key anomaly trajectories is as follows: Key anomaly trajectories are grouped into multiple levels according to their fluctuation characteristics, including a transient abrupt change layer, a sustained gradual change layer, and a composite anomaly layer. Within each layer, the short-term mean fluctuation range, maximum deviation, and gradient change rate of the indicator values ​​are calculated to assess the intensity of entropy changes and the time span of the indicator distribution in each layer.

2. The adaptive management method for power big data based on spatiotemporal feature mapping according to claim 1, characterized in that, The process of identifying key abnormal trajectories through a time-varying feature nesting mechanism is as follows: Decouple the time-series data in the cache according to the device dimension and construct a multi-dimensional data structure with time as the main axis and metrics as nested layers; Time series variation detection is performed on each nested index layer to identify evolutionary trajectories with nonlinear mutations, periodic misalignments, or trend drift characteristics; A time-varying feature nesting mechanism is introduced to compare the similarities and differences between trajectories in different time periods and identify key abnormal trajectories.

3. The adaptive management method for power big data based on spatiotemporal feature mapping according to claim 1, characterized in that, The process of identifying potential security risk nodes is as follows: Extract indicator segments with significant fluctuation intensity and prominent entropy disturbances from key abnormal trajectories, and construct feature vectors for abnormal trajectories; Extract typical abnormal evolution paths from historical abnormal event evolution maps and establish a structural matching model between map nodes and current abnormal trajectory features; By calculating the structural similarity and evolutionary trend deviation between the current abnormal trajectory and historical evolution nodes, potential security risk nodes can be identified.

4. The adaptive management method for power big data based on spatiotemporal feature mapping according to claim 3, characterized in that, The process of fine-tuning the risk assessment results using a spatiotemporal collaborative correction strategy is as follows: Analyze the spatial topology and device interconnection relationships of identified potential risk nodes to determine their interference paths with surrounding nodes; Based on the spatiotemporal correlation coefficient between nodes, a risk backpropagation graph is constructed to dynamically compress and expand the risk level boundary; A correction feedback mechanism is introduced to adjust the risk assessment weight coefficients based on actual control feedback and equipment operating status.

5. The adaptive management method for power big data based on spatiotemporal feature mapping according to claim 4, characterized in that, The process of hierarchical linkage response between the power grid dispatch control terminal and the front-end operation and maintenance nodes is as follows: Based on the revised risk levels, event types are categorized, and a response level index is constructed. The system calls the scheduling strategy library to generate control-end scheduling instructions based on risk type and location, including load transfer and voltage / current limiting. Synchronously access the operation and maintenance strategy library and schedule the front-end maintenance nodes to generate operation plans.

6. A power big data adaptive management system integrating spatiotemporal feature mapping, applied to the method described in any one of claims 1-5, characterized in that, include: Data acquisition module: Receives real-time operation data streams from multiple power terminals and constructs a raw power dataset with time-series labels and device identifiers through a time window reassembly mechanism and a heterogeneous decoding strategy at the protocol layer; Index building module: The original power dataset is divided into parallel processing units, and event-driven index mapping rules are used for dynamic index updates to build a multi-dimensional data index caching system with real-time response capabilities; Feature Analysis Module: Identifies key abnormal trajectories through a time-varying feature nesting mechanism, and uses a streaming feature aggregation network to perform hierarchical measurement and entropy perturbation analysis on the degree of data fluctuation. Risk Correction Module: By combining the structural matching degree between the historical anomaly evolution map and the key anomaly trajectory, potential safety risk nodes are identified, and the risk assessment results are fine-tuned in reverse based on the spatiotemporal collaborative correction strategy. Response generation module: Generates a multi-level response instruction chain based on risk assessment results, driving the power grid dispatch control terminal and the front-end operation and maintenance nodes to perform hierarchical linkage response.

Images