Fusion analysis method and system for multi-dimensional spatial data of power grid

Abstract

The invention relates to the technical field of power grid data fusion analysis, and discloses a power grid multi-dimensional spatial data fusion analysis method and system, and the method comprises the steps: collecting power grid multi-dimensional spatio-temporal data, and forming a spatio-temporal data standard set; constructing a dynamic weight fusion mechanism according to the spatio-temporal data standard set, obtaining a data quality factor, an aging factor, an importance factor and a correlation factor, and obtaining a fusion weight factor set; constructing a three-dimensional space-time coordinate frame, and mapping each dimension of data in the space-time data standard set to the three-dimensional space-time coordinate frame; fusing the data of each dimension in the three-dimensional space-time coordinate framework based on the fusion weight factor set to obtain a comprehensive association degree value of each space-time coordinate point as a data fusion analysis result; the data fusion analysis result is used for formulating a power grid operation strategy and judging power grid operation abnormity. The method has the outstanding advantages of being high in adaptive capacity, capable of accurately quantifying the comprehensive correlation degree of each coordinate point, capable of being applied to power grids of different scales and different operation scenes and the like.

Description

Technical Field

[0001] This invention relates to the field of power grid data fusion and analysis technology, specifically to a method and system for multi-dimensional spatial data fusion and analysis of power grids. Background Technology

[0002] With the deepening of smart grid construction, especially in key application scenarios such as grid operation status monitoring, fault early warning, and load forecasting, the need for accurate analysis of the correlation of multi-dimensional spatiotemporal data of the power grid is becoming increasingly urgent. Although traditional data analysis methods have made some progress in single data source processing, simple weight allocation, and basic anomaly detection, they have not systematically solved core problems such as dynamic weight allocation of data quality and timeliness, effective identification of deep correlations between multi-dimensional spatiotemporal data, and the propagation law of anomaly patterns in the spatiotemporal dimension. They are difficult to meet the needs of integrated analysis for power grids of different sizes, different operating scenarios, and different data types.

[0003] Therefore, how to integrate multi-source heterogeneous data, deeply explore spatiotemporal correlations, accurately capture the dynamic changes in the power grid's operating status, and have a strong adaptive capability as a multi-dimensional spatial data fusion analysis method has become an urgent problem to be solved.

[0004] CN120408479A discloses an intelligent fusion method for multi-source data in power systems based on soft-merging DS evidence theory, relating to the technical field of using multi-source heterogeneous data in the power system industry. The method includes acquiring multi-source data and calculating change indicators for the multi-source data; performing data detection on the multi-source data to obtain detection results; aggregating the multi-source data based on the change indicators and the detection results to obtain data features; and fusing the data features to obtain a data fusion result. However, this patent does not perform comprehensive temporal and spatial correlation analysis of each spatiotemporal coordinate point, cannot accurately quantify the temporal and spatial correlation at each coordinate point, cannot be applied to different power grid operation scenarios, and is difficult to apply to large-scale power grids; therefore, its applicability is limited. Summary of the Invention

[0005] To address the shortcomings of existing technologies, such as the inability to accurately quantify the temporal and spatial correlation at each coordinate point and the inability to apply them to power grids of different scales, operating scenarios, and data types, this invention provides a method and system for multi-dimensional spatial data fusion analysis of power grids.

[0006] The present invention adopts the following technical solution.

[0007] This invention discloses a method for multi-dimensional spatial data fusion analysis of power grids, comprising: Collect multidimensional spatiotemporal data of the power grid to form a spatiotemporal data standard set; A dynamic weight fusion mechanism is constructed based on the spatiotemporal data standard set to obtain data quality factors, timeliness factors, importance factors, and relevance factors, thereby obtaining a set of fusion weight factors. A three-dimensional spatiotemporal coordinate framework is constructed, and data of each dimension in the spatiotemporal data standard set are mapped to the three-dimensional spatiotemporal coordinate framework; The data of each dimension in the three-dimensional spatiotemporal coordinate framework are fused based on the fusion weight factor set, and the comprehensive correlation value of each spatiotemporal coordinate point is obtained as the data fusion analysis result; the data fusion analysis result is used to formulate power grid operation strategy and judge power grid operation anomalies.

[0008] More preferably, The multidimensional spatiotemporal data of the power grid includes transformer operation data and environmental monitoring data; The transformer operating data includes the load of each transformer, the operating status of the relay protection device, and the bus voltage; The environmental monitoring data includes temperature, humidity, and wind speed.

[0009] More preferably, The spatiotemporal data standard set includes the timestamp of the data acquisition time, the load of each transformer, the bus voltage, the operating status of the relay protection device, the temperature, humidity and wind speed, as well as the sampling location of the load of each transformer, the bus voltage, the operating status of the relay protection device, the temperature, humidity and wind speed.

[0010] More preferably, The dynamic weight fusion mechanism is constructed based on the spatiotemporal data standard set to obtain data quality factors, timeliness factors, importance factors, and relevance factors, resulting in a fusion weight factor set, including: Construct a dynamic weighting model based on data quality factors, timeliness factors, importance factors, and relevance factors; The weighting factors of each dimension of the spatiotemporal data standard set at each time point are calculated based on the dynamic weighting model. A fusion weight factor set is constructed based on the weight factors of data from each dimension.

[0011] More preferably, The calculation of weight factors for each dimension of data in the spatiotemporal data standard set based on the dynamic weight model is based on the quality factor, timeliness factor, importance factor, relevance factor, data quality weight coefficient, timeliness weight coefficient, importance weight coefficient, and relevance weight coefficient corresponding to each dimension of data in each spatiotemporal data standard set. The data quality weight coefficient, timeliness weight coefficient, importance weight coefficient, and relevance weight coefficient are all dynamic weight adjustment parameters. Each of the aforementioned dynamic weight adjustment parameters is calculated based on its time decay coefficient and the time difference between the current time and the data acquisition time.

[0012] More preferably, The data quality factor is calculated based on the actual number of data points, the number of non-abnormal data points, and the number of data points to be collected for each weight factor data to be calculated within a set first time length. The time factor is calculated based on the time difference between the acquisition time of each weight factor data to be calculated and the current time, as well as the time decay coefficient corresponding to the data type; the data types include transformer operation data and environmental monitoring data. The importance factors are determined based on the data dimensions of each weight factor data to be calculated, and the preset data dimension-importance factor correspondence; the data dimensions include the load of each transformer, the operating status of the relay protection device, the bus voltage, temperature, humidity and wind speed; The correlation factor is calculated based on the correlation coefficient between the time series of each weight factor data to be calculated in the preset first time period and the time series of the data in the preset second time period, as well as the corresponding mutual information value.

[0013] More preferably, The three-dimensional spatiotemporal coordinate framework includes longitude coordinates, latitude coordinates, and time coordinates; The step of mapping each dimension of data in the spatiotemporal data standard set to the three-dimensional spatiotemporal coordinate frame involves mapping the value of each dimension of data in the spatiotemporal data standard set to the corresponding coordinate point in the three-dimensional spatiotemporal coordinate frame and performing three-dimensional interpolation to construct the spatiotemporal distribution function of each dimension of data.

[0014] More preferably, The method of fusing data from each dimension in a three-dimensional spatiotemporal coordinate framework based on a set of fusion weight factors to obtain a comprehensive correlation value for each spatiotemporal coordinate point is to calculate the comprehensive correlation value for each spatiotemporal coordinate point based on the weights of each dimension of data at that coordinate point, the spatiotemporal distribution function, and the spatiotemporal correlation strength between each dimension of data and other dimension of data at that coordinate point. The spatiotemporal correlation strength between each dimension of data and other dimension of data is calculated based on the geographical distance and electrical distance between the sampling locations of each dimension of data and other dimension of data, as well as the sampling time interval between each dimension of data and other dimension of data.

[0015] Another aspect of this invention discloses a power grid multi-dimensional spatial data fusion analysis system based on a power grid multi-dimensional spatial data fusion analysis method, comprising a data acquisition and standard set construction module, a dynamic weight fusion module, a data mapping module, and a fusion analysis result generation module: The data acquisition and standard set construction module is used to collect multi-dimensional spatiotemporal data of the power grid and form a spatiotemporal data standard set. The dynamic weight fusion module is used to construct a dynamic weight fusion mechanism based on the spatiotemporal data standard set, obtain data quality factors, timeliness factors, importance factors and relevance factors, and obtain a fusion weight factor set; The data mapping module is used to construct a three-dimensional spatiotemporal coordinate frame and map data of each dimension in the spatiotemporal data standard set to the three-dimensional spatiotemporal coordinate frame. The fusion analysis result generation module fuses data from each dimension of the three-dimensional spatiotemporal coordinate framework based on the fusion weight factor set, and obtains the comprehensive correlation value of each spatiotemporal coordinate point as the data fusion analysis result; the data fusion analysis result is used to formulate power grid operation strategies and judge power grid operation anomalies.

[0016] Another aspect of this application discloses an electronic device, including a processor and a storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the aforementioned multi-dimensional spatial data fusion analysis method for power grids.

[0017] This application also discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the power grid multi-dimensional spatial data fusion analysis method.

[0018] The beneficial effects of this invention are compared with those of the prior art: This invention addresses the shortcomings of traditional methods in integrating multi-source heterogeneous data by collecting multi-dimensional spatiotemporal data of the power grid to form a spatiotemporal dataset, thereby realizing the unified collection and management of power grid GIS graphic data, equipment operation data, environmental monitoring data, and historical fault data. After preprocessing and standardization, the spatiotemporal dataset is obtained as a standardized spatiotemporal data set with a unified format, which effectively solves the problems of inconsistent data formats and coordinate systems, and provides a standardized data foundation for subsequent analysis. A dynamic weight fusion mechanism is constructed based on the spatiotemporal data standard set to obtain data quality factors, timeliness factors, importance factors, and relevance factors, resulting in a fusion weight factor set. This breaks through the limitations of traditional fixed weight allocation and realizes intelligent dynamic adjustment of weights based on data features, significantly improving the accuracy and adaptability of data fusion. By fusing and analyzing the standard set of historical spatiotemporal data through a set of fusion weight factors, a spatiotemporal fusion correlation analysis mechanism was constructed, and spatiotemporal fusion analysis rules were obtained. This effectively solved the problem of difficulty in identifying deep correlations between multidimensional spatiotemporal data, and realized the accurate quantification of the spatiotemporal distribution characteristics of power grid operation status and the in-depth mining of correlation patterns. Based on the spatiotemporal fusion analysis rules, the power grid operation principles are integrated to obtain the power grid operation strategy. The combination of theoretical analysis and practical operation experience forms a scientific and reasonable power grid operation decision support system, which improves the reliability and practicality of the operation strategy. Based on the power grid operation strategy, a multi-level anomaly detection mechanism is constructed to generate abnormal operating conditions, which solves the problems of low accuracy and high false alarm rate of traditional anomaly detection methods, realizes intelligent anomaly identification based on operating rules, and significantly improves the accuracy and efficiency of anomaly detection. The spatiotemporal fusion analysis rules obtained from the current spatiotemporal data standard set are used to construct the current power grid operation strategy. The anomaly detection mechanism is then used to detect the anomalies and obtain the anomaly detection results. This enables real-time monitoring and dynamic evaluation of the power grid operation status, effectively improving the safety and stability of the power grid operation. The algorithm determines whether the anomaly detection result meets the abnormal operating conditions. If the anomaly detection result meets the abnormal operating conditions, the fusion weight factor set is re-acquired to adjust the spatiotemporal fusion analysis rules until the anomaly detection result no longer meets the abnormal operating conditions. An adaptive optimization mechanism is established, which solves the problem of the lack of self-adjustment ability in traditional methods and realizes the continuous optimization and performance improvement of the analysis algorithm. By quantifying anomaly detection results through adaptive visualization algorithms and conducting feedback learning, a closed-loop learning system is constructed, which effectively solves the problems of fixed algorithm parameters and inability to optimize based on actual results, and achieves continuous improvement in system performance and continuous enhancement of intelligence level. By using WebGL technology to achieve real-time 3D rendering, the feedback learning results are presented in a visual way, which solves the problems of limited information and poor interactivity in traditional 2D display methods. It realizes intuitive display and real-time interaction of multi-dimensional data, significantly improving user experience and decision-making efficiency.

[0019] This invention can integrate multi-source heterogeneous data, deeply mine spatiotemporal correlations, and accurately capture the dynamic changes in the power grid's operating status. It has outstanding advantages such as strong adaptability, accurate quantification of the comprehensive correlation at each coordinate point, and applicability to power grids of different scales, different operating scenarios, and different data types. The data fusion analysis results obtained by this invention can be used to formulate power grid operation strategies and identify power grid operation anomalies. Attached Figure Description

[0020] Figure 1This is a schematic diagram of the multi-dimensional spatial data fusion and analysis process for power grids; Figure 2 This is a flowchart illustrating a method for multi-dimensional spatial data fusion analysis of a power grid provided in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of a multi-dimensional spatial data fusion and analysis system for power grids provided in Embodiment 6 of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, other embodiments obtained by those skilled in the art without creative effort are all within the protection scope of this invention.

[0022] This application discloses a method for multi-dimensional spatial data fusion analysis of power grids. (See appendix) Figure 1 ,include: Collect multidimensional spatiotemporal data of the power grid to form a spatiotemporal data standard set; The multidimensional spatiotemporal data of the power grid includes transformer operation data and environmental monitoring data; The transformer operation data includes the load of each transformer, the status of each standby transformer (started, not started), the operating status of relay protection devices (operated, not operated), the status of power distribution equipment (switch position, protection device status), the status of communication equipment, the status of auxiliary equipment (cooling system status (started, not started), lighting system status (on, off)) and bus voltage.

[0023] The environmental monitoring data includes temperature, humidity, and wind speed.

[0024] The spatiotemporal data standard set includes the timestamp of the data acquisition time, the load of each transformer, the status of each standby transformer, the switch position, the status of protection devices, the status of communication equipment, the status of the cooling system (started, not started), the status of the lighting system (on, off), the bus voltage, the operating status of the relay protection device, temperature, humidity, and wind speed, as well as the sampling location corresponding to each data dimension of the transformer load, bus voltage, relay protection device operating status, temperature, humidity, and wind speed. The data dimensions include the load of each transformer (different transformer loads represent different data dimensions), the status of each standby transformer (different standby transformer loads represent different data dimensions), the switch position (different switch positions represent different data dimensions), the status of protection devices (different switch positions represent different data dimensions), the status of communication equipment (different communication device statuses represent different data dimensions), the status of the cooling system (started, not started) (different cooling system statuses represent different data dimensions), the status of the lighting system (on, off) (different lighting system statuses represent different data dimensions), the operating status of the relay protection device (different relay protection device operating statuses represent different data dimensions), the bus voltage, temperature, humidity, and wind speed. A dynamic weighted fusion mechanism is constructed based on a spatiotemporal data standard set to obtain data quality factors, timeliness factors, importance factors, and relevance factors, resulting in a fusion weight factor set, including: A dynamic weighting model based on data quality factors, timeliness factors, importance factors, and relevance factors is constructed. The weight fusion mechanism is presented in the form of a formula, which is to obtain the weight of each dimension of data in each spatiotemporal data standard set by weighting the quality factors, timeliness factors, importance factors, and relevance factors of each dimension of data in each spatiotemporal data standard set.

[0025] The weighting factors of each dimension of data in the spatiotemporal data standard set at each time point are calculated based on the dynamic weighting model. Specifically, the weighting factors of each data point in the spatiotemporal data standard set are calculated based on the quality factor, timeliness factor, importance factor, relevance factor, data quality weight coefficient, timeliness weight coefficient, importance weight coefficient, and relevance weight coefficient corresponding to each data point in the spatiotemporal data standard set. The data quality weight coefficient, timeliness weight coefficient, importance weight coefficient, and relevance weight coefficient are all dynamic weighting adjustment parameters. The weight factor for each data point in each spatiotemporal data standard set is calculated using the following formula: ; in, represents the dynamic weight of the d-th dimension data in the i-th spatiotemporal data standard set at the current time; i represents the i-th spatiotemporal data standard set, i∈[1,Ns]; Ns represents the total number of spatiotemporal data standard sets up to the current time; t represents the sampling time of the i-th spatiotemporal data standard set; Indicates the current time. d Data quality weighting coefficients for dimensional data; This represents the i-th spatiotemporal data standard set at the current time. d Data quality factor for dimensional data; Indicates the current time. d Timeliness weighting coefficient for dimensional data; This represents the i-th spatiotemporal data standard set at the current time. d The timeliness factor of dimensional data; Indicates the current time. d Importance weighting coefficients for dimensional data; This represents the i-th spatiotemporal data standard set at the current time. d Importance factors of dimensional data; Indicates the current time. d The correlation weight coefficient of dimensional data; This represents the i-th spatiotemporal data standard set at the current time. d The correlation factor of dimensional data; The data quality factor is calculated based on the actual number of data points, the number of non-abnormal data points (number of non-abnormal data points = number of actual data points - number of abnormal data points), the number of inconsistent data points, the total number of verification points, and the number of data points to be collected for each weight factor data to be calculated within a set first time length (the data quality factor for each weight factor data to be calculated is obtained accordingly); preferably, the first time length is 1 hour. The calculation formula is as follows: ; in: The integrity weight is preferably set to 0.4. For the current moment d Integrity score of dimensional data; For accuracy weighting, its preferred value is 0.3; For the current moment d Accuracy rating of dimensional data; The consistency weight is preferably set to 0.3. For the current moment d Consistency score of dimensional data.

[0026] The formula for calculating the integrity score is as follows: ; The actual number of data points is the number of data points within the set first time period. d The actual number of data points in the dimension data; the number of data points to be collected is the first data point within the set first time length. dThe number of data points to be collected for dimensional data.

[0027] The accuracy score is calculated using the following formula: ; Among them, the number of non-abnormal data points is the first number within the set first time length. d The actual number of data points in the dimension data minus the number of abnormal data points; those skilled in the art should know how to identify abnormal data points, which will not be elaborated here.

[0028] The formula for calculating the consistency score is as follows: ; The number of inconsistent data points is the first time point within the set first time length. d The number of inconsistent data points in the dimension data; the total number of validation points is the number of validation points performed within the set first time period. d The total number of data points in a dimension; the inconsistent data points refer to data points that, after data verification, are determined not to meet the logical relationship of the data in that dimension. For example, when checking the power balance relationship, if the input power is not equal to the output power plus loss, it is recorded as an inconsistent data point to ensure the consistency between the input power and the output power plus loss. When verifying the voltage level matching relationship, if the voltage ratio between the upper and lower levels is unreasonable, it is recorded as an inconsistent data point to ensure the reasonable ratio between the upper and lower levels.

[0029] The timeliness factor is calculated based on the time difference between the acquisition time and the current time for each weight factor data to be calculated, and the time decay coefficient corresponding to the data type (the timeliness factor of each weight factor data to be calculated is obtained accordingly); the data types include transformer operation data and environmental monitoring data; the timeliness factor The calculation formula is as follows: ; in, It is a natural constant; For the first d The time decay coefficient of dimensional data; Let be the time difference between the current time and the data acquisition time of the i-th spatiotemporal data standard set; when the i-th time difference is... d When the dimensional data is transformer operating data, / minute; when the d When the dimensional data is environmental monitoring data, / minute; The importance factor is determined based on the data dimension of each weight factor data to be calculated, and a preset data dimension-importance factor correspondence (determining the importance factor of each weight factor data to be calculated); the data dimension-importance factor correspondence includes: each transformer load, bus voltage, and the load at the starting end of each branch line on the outgoing side of each transformer all correspond to an importance factor of 1.0 (i.e., the first...). d When the dimension data is transformer load, bus voltage, or the load at the starting end of each branch line on the outgoing side of each transformer, The importance factor for the relay protection device's operating status (on, off), power distribution equipment status (switch position, protection device status), and communication equipment status is 0.8 (i.e., the first importance factor). d When the dimension data is the operating status of relay protection devices, the status of power distribution equipment, or the status of communication equipment, 0.8) The status of auxiliary equipment (cooling system status (started, not started), lighting system status (on, off)), temperature, humidity, and wind speed all correspond to an importance factor of 0.6 (i.e., the first importance factor). d When the dimension data is auxiliary equipment status, temperature, humidity, or wind speed, The importance factor for each of the following is 0.4 (i.e., the first importance factor is 0.4): standby equipment information (standby transformer status (started, not started), standby line information (used, not used)). d When the dimension data is backup device information, 4).

[0030] The correlation factor is calculated based on the correlation coefficient between the time series of each weight factor data to be calculated in a preset first time period and the time series of that data in a preset second time period, and the mutual information value between the time series of each weight factor data to be calculated in the first time period and the time series of that data in the preset second time period (the correlation factor of each weight factor data to be calculated is obtained accordingly). Preferably, for The first time period is [the acquisition time of the i-th spatiotemporal data standard set - the set second time length, the acquisition time of the i-th spatiotemporal data standard set], and [the current time - the set second time length, the current time]. The correlation factor is obtained by weighting the time series of the data to be weighted in the first preset time period and the time series of the data to be weighted in the second preset time period with the correlation coefficient and mutual information value according to a weight of 6:4. The second time length is preferably 0.5 to 1 minute.

[0031] Each of the aforementioned dynamic weight adjustment parameters is calculated based on its initial set value, time decay coefficient, and the time difference between the current time and the data acquisition time of the data corresponding to that dynamic weight adjustment parameter, as shown in the following formula: ; in, Let be the weight ratio of the k-th dynamic weight adjustment parameter corresponding to the i-th spatiotemporal data standard set. This is the initial setting value for the k-th dynamic weight adjustment parameter. This value is determined based on the operating scenario at the time of data acquisition. The operating scenarios include normal operation mode, emergency response mode, maintenance and repair mode, and historical analysis mode. When the data acquisition time is in normal operation mode, the preferred initial setting values ​​for the data quality weight coefficient are 0.25-0.35, the timeliness weight coefficient is 0.20-0.30, the importance weight coefficient is 0.20-0.30, and the correlation weight coefficient is 0.15-0.25. When the data acquisition time is in emergency response mode, the preferred initial setting values ​​for the timeliness weight coefficient and the importance weight coefficient are increased to 0.4. The preferred initial values ​​for the data quality weight coefficient and the correlation factor weight coefficient are both reduced to 0.1. When the data collection time is in maintenance mode, the preferred initial values ​​for the data quality weight coefficient are increased to 0.5, the preferred initial values ​​for the timeliness weight coefficient and the importance weight coefficient are both set to 0.2, and the preferred initial values ​​for the correlation weight coefficient are both set to 0.1. When the data collection time is in historical analysis mode, the preferred initial values ​​for the correlation weight coefficient are increased to 0.4, the preferred initial values ​​for the timeliness weight coefficient are reduced to 0.1, the preferred initial values ​​for the data quality weight coefficient and the importance weight coefficient are both set to 0.3. This represents the time decay coefficient of the k-th dynamic weight adjustment parameter; when the k-th dynamic weight adjustment parameter is the data quality weight coefficient... =0.02 / minute, when the k-th dynamic weight adjustment parameter is the timeliness weight coefficient, =0.2 / minute, when the k-th dynamic weight adjustment parameter is the importance weight coefficient, =0.001 / minute, when the k-th dynamic weight adjustment parameter is the correlation weight coefficient, =0.05 / minute; The time interval is the time difference between the current moment and the data acquisition moment. It is a natural constant.

[0032] A fusion weight factor set is constructed based on the weight factors of data from each dimension.

[0033] A three-dimensional spatiotemporal coordinate framework is constructed, and data of each dimension in the spatiotemporal data standard set are mapped to the three-dimensional spatiotemporal coordinate framework; The three-dimensional spatiotemporal coordinate framework includes longitude coordinates, latitude coordinates, and time coordinates; The step of mapping each data point in the spatiotemporal data standard set to the three-dimensional spatiotemporal coordinate frame involves mapping the value of each dimension of data to the corresponding coordinate point in the three-dimensional spatiotemporal coordinate frame and performing three-dimensional interpolation to construct a continuous function for each dimension of data. This continuous function is then used as the spatiotemporal distribution function of each dimension of data.

[0034] The data of each dimension in the three-dimensional spatiotemporal coordinate framework are fused based on the fusion weight factor set to obtain the comprehensive correlation value of each spatiotemporal coordinate point. That is, at each spatiotemporal coordinate point, the correlation value is calculated based on the weights of each dimension of data at that coordinate point, the spatiotemporal distribution function, and the spatiotemporal correlation strength between each dimension of data and other dimensions of data at that coordinate point (for example, if there is first-dimensional data at that coordinate point, the spatiotemporal correlation strength between the first-dimensional data and other data at that coordinate point includes the spatiotemporal correlation strength between the first-dimensional data and the second-dimensional data at time...). t The spatiotemporal correlation strength, the first-dimensional data and the third-dimensional data at time... t The spatiotemporal correlation strength, ..., the first dimension data and the second dimension m Dimensional data at time t The spatiotemporal correlation strength is calculated for each spatiotemporal coordinate point, and the correlation strength for other dimensions is calculated similarly. The spatiotemporal correlation strength between each dimension of data and other dimensions of data is calculated based on the geographical distance and electrical distance between the sampling locations of each dimension of data and other dimensions of data, as well as the sampling time interval between each dimension of data and other dimensions of data (the sampling time interval between each dimension of data and other dimensions of data is based on the sampling time of each dimension of data).

[0035] The formula for calculating the overall correlation score is as follows: ; Among them, F This represents the comprehensive correlation value of a spatiotemporal coordinate point with longitude coordinate x, latitude coordinate y, and time coordinate t; d represents the d-th data dimension with longitude coordinate x and latitude coordinate y of the sampling point, d∈[1, ], The total number of data dimensions with longitude coordinates x and dimensional coordinates y for each data sampling point; h represents the h-th data dimension, h∈[1, y]. m ], mThis represents the total number of data dimensions. This represents the dynamic weight of the d-th dimension data in the i-th spatiotemporal data standard set at the current moment; This represents the data of dimension d in a three-dimensional spatiotemporal coordinate frame. The spatiotemporal distribution function value at that location; This indicates that the data in dimension d and the data in dimension h are at time d. t The spatiotemporal correlation strength; the data dimensions include data from various dimensions in transformer operation data, data from various dimensions in environmental monitoring data, and load data.

[0036] The spatiotemporal correlation strength between data in each dimension and other data is calculated based on the spatial correlation weight coefficient, temporal correlation weight coefficient, spatial correlation degree, and temporal correlation degree between the g-th dimension data and the h-th dimension data. The calculation formula is as follows: ; in, Let g be the spatiotemporal correlation strength between the g-th dimension data and the h-th dimension data at time t; The spatial correlation weighting coefficients between the g-th and h-th dimension data are... The time correlation weighting coefficient between the g-th dimension data and the h-th dimension data; The spatial correlation degree between the g-th dimension data and the h-th dimension data; The temporal correlation between the g-th dimension data and the h-th dimension data.

[0037] Spatial correlation weighting coefficient between the g-th dimension data and the h-th dimension data The distance is calculated based on the geographic and electrical distances between the g-th and h-th dimension data, and the calculation formula is as follows: ; in, The geographical distance between the g-th dimension data and the h-th dimension data. The electrical distance between the g-th and h-th dimension data (referring to the electrical connection distance between the devices corresponding to the g-th and h-th dimension data). The time correlation weighting coefficient between the g-th dimension data and the h-th dimension data The data is obtained based on the timestamp of the g-th dimension data, the sampling time interval of the h-th dimension data, and the time decay constant. The calculation formula is as follows:

[0038] in, , h represents the data timestamps of the g-th dimension and the h-th dimension, respectively. and The time interval between h can be set by those skilled in the art, or it can be based on the time lag coefficient used in the time correlation calculation formula. Determined, that is = ; Let be the time decay constant (where ) The preferred value range is 5-30 minutes, which can be adjusted according to the data type and correlation strength requirements; preferably, when both the data timestamp of the g-th dimension and the data of the h-th dimension are transformer operation data. =5 minutes, when the timestamp of the data in the g-th dimension and the data in the h-th dimension are both environmental monitoring data. =15 minutes; if the data timestamp of the g-th dimension and the data of the h-th dimension are of different data types, they can be adjusted by those skilled in the art according to the actual situation. (·) indicates that the data within the parentheses is exponentially raised to the base e of the natural constant.

[0039] Spatial correlation between the g-th dimension data and the h-th dimension data The calculation is based on the geographical distance between the g-th and h-th dimension data, the electrical distance between the g-th and h-th dimension data, and the interaction frequency of substations related to the g-th and h-th dimension data. The calculation formula is as follows:

[0040] in, , , These are the weighting coefficients for geographical distance, electrical distance, and historical interaction frequency (preferred value range: , , + + , The geographical distance between the g-th dimension data and the h-th dimension data. Let g be the electrical distance between the g-th dimension data and the h-th dimension data. The frequency of data interaction between the substations corresponding to the g-th and h-th dimension data within a set third time period, where the data type of either the g-th or h-th dimension data is environmental monitoring data. The value is 0 when neither the data type of the g-th dimension nor the h-th dimension is environmental monitoring data, and when the substations corresponding to the g-th dimension and the h-th dimension are the same. ; This indicates taking the maximum value within the parentheses; This represents the maximum frequency of historical data interaction between the substations corresponding to the g-th dimension data and the h-th dimension data within a set third time period. The third time period is preferably 1 hour.

[0041] The temporal correlation between the g-th dimension data and the h-th dimension data is based on the g-th dimension data sequence at time t and t- The h-th dimension of the data sequence at time h is obtained by calculation, where The time lag coefficient; the temporal correlation between the g-th dimension data and the h-th dimension data. The calculation formula is as follows: ; in, and They are the g-th dimension data sequence at time t and t- The data sequence of the h-th dimension at time step, Indicates calculation and The Pearson correlation coefficient; The time lag factor is used; the data sequence at each moment is preferably the data sequence within the interval (the moment at this time - the fourth time length, the moment at this time), and the fourth time length is preferably 5 seconds.

[0042] Time lag coefficient Determined in the following manner: By calculating different time lag coefficients The correlation coefficient between the next two time series is used to find the maximum correlation coefficient value within a preset time window, and the value corresponding to the maximum correlation coefficient value is then used. As a means to calculate the temporal correlation between the g-th dimension and the h-th dimension data. This is used to determine the optimal temporal correlation between the two data sequences. The preferred value range for the preset time window is 3 hours.

[0043] The comprehensive correlation value is used as the result of data fusion analysis; the data fusion analysis result is used to formulate power grid operation strategies and identify power grid operation anomalies.

[0044] Example 1 A method for multi-dimensional spatial data fusion analysis of power grids, such as Figure 2 As shown, it includes: Step 1: Collect multi-dimensional spatiotemporal data of the power grid to form a spatiotemporal data standard set; 1.1 The data acquisition system collects multidimensional spatiotemporal data of the power grid through multiple data interfaces; after preprocessing and standardization of the spatiotemporal dataset, a standardized spatiotemporal data set with a unified format is obtained; The multidimensional spatiotemporal data of the power grid includes geographic information system (GIS) graphic data, transformer operation data, environmental monitoring data, and historical fault data.

[0045] The GIS graphic data includes geographic information such as the global positioning system coordinates (GPS coordinates) of the substation, the route of the transmission line (determined by the GPS coordinates of the starting point and the end point and the coordinates of key intermediate nodes), and the GPS coordinates of the power distribution equipment (accurate to 6 decimal places). The collected GIS graphic data is stored in GeoJSON format and is updated in real time. The transformer operation data includes transformer load rate, status of each switch in the transformer (closed or open), protection device operation information (including protection device number, operation start time, operation stop time, and operation type), main transformer load, bus voltage, main line current, load at the starting end of each branch line on the outgoing side of each transformer, and standby transformer status (started or not started). Among these, the transformer load rate, status of each switch in the transformer, and protection device operation information are collected from all transformers in the area through a data acquisition and monitoring control system (SCADA system). The area is divided according to voltage level and geographical location, preferably within a radius of 3-5 kilometers for substations with voltage levels of 110kV and above. The main transformer load is collected by current transformers installed on the high-voltage side of the transformer, the bus voltage is collected by voltage transformers installed on the bus at each voltage level, and the main line current is collected by current transformers installed in the outgoing line bays. The sampling interval is 1 minute. The environmental monitoring data includes meteorological information such as temperature, humidity, and wind speed. It can be collected by setting up temperature sensors, humidity sensors, and wind speed sensors outdoors, or directly through a weather station. The sampling interval is 15 minutes. The historical fault data includes historical records such as fault type (short circuit fault, ground fault, overload fault, equipment fault, etc.), occurrence time (UTC timestamp format), and affected area (including the number of users affected and the affected area in square kilometers as the quantification unit).

[0046] 1.2 Data Preprocessing: Establishing a unified spatiotemporal coordinate system, specifically including coordinate system unification, timestamp standardization, data format standardization, and handling of missing values; The aforementioned coordinate system unification involves converting all geographic data to the World Geodetic System 1984 coordinate system (WGS84 coordinate system) to ensure consistency in spatial location. The timestamp standardization refers to the unified adoption of the Coordinated Universal Time (UTC) format, with a precision down to the millisecond level. The data format standardization involves converting data from different sources into a unified JSON format, denoted as a spatiotemporal data standard set. All data at each moment constitutes a spatiotemporal data standard set, which includes timestamps, location information (determined by correlation with substation coordinates, transmission line routes, and power distribution equipment coordinates), numerical values, and quality indicators (determined based on a comprehensive score of data completeness, accuracy, and timeliness, with a score range of 0-1, where 0.9 or above is excellent, 0.7-0.9 is good, 0.5-0.7 is average, and below 0.5 is poor). The missing value handling method uses linear interpolation to fill in the missing values ​​of time series data. For critical equipment data (equipment that directly affects the safe and stable operation of the power grid, such as main transformers, important busbars, and main lines, and is distinguished by the number of users affected when the equipment fails, which exceeds 1,000 or affects important users), a forward filling method is used. The aforementioned spatiotemporal data standard set is constructed by gathering all data at each time point, containing complete data information from all collection points at that time. For example, at time T1, all GIS data, transformer operation data, environmental monitoring data, and historical fault data from all substations are integrated into a complete spatiotemporal data standard set.

[0047] 1.3 Data Quality Monitoring: Establish a real-time data quality monitoring mechanism.

[0048] The data quality monitoring mechanism includes integrity monitoring, accuracy verification, timeliness check, and consistency verification.

[0049] The integrity monitoring refers to the statistical analysis of the data arrival rate of each data source (data acquisition equipment such as substations and monitoring points) within a set time window (preferably 1 hour), and triggering an alarm when the arrival rate is lower than 95%. The accuracy verification is performed by cross-validating multi-source data to identify data points with abnormal values. The timeliness check monitors data delays, and data exceeding a preset time window is marked as expired. The consistency check examines the logical consistency between related data, such as power balance relationships.

[0050] Step 2: Construct a dynamic weight fusion mechanism based on the spatiotemporal data standard set, obtain data quality factors, timeliness factors, importance factors, and relevance factors, and obtain a set of fusion weight factors; 2.1 Construct dynamic weights for each data in the spatiotemporal data standard set, and construct a fusion weight factor set based on the dynamic weights of each data.

[0051] The dynamic weights of each data point in the spatiotemporal data standard set are constructed according to the following method: Constructing a data quality factor Time factor Importance factors correlation factor Dynamic weight model:

[0052] in, Let be the dynamic weight of a certain dimension of data in the i-th spatiotemporal data standard set at the current time; t represents the sampling time of the i-th spatiotemporal data standard set. This represents the data quality factor for this dimension at the current moment. This serves as the timeliness factor for the data in this dimension at the current moment. This represents the importance factor of the data in this dimension at the current moment. This is the correlation factor for the data in this dimension at the current moment; This represents the data quality weighting coefficient for this dimension at the current moment. This represents the timeliness weighting coefficient for the data in this dimension at the current moment. This represents the importance weight coefficient of the data in this dimension at the current moment. The four coefficients are the relevance weight coefficients of the data in this dimension at the current moment. All four coefficients are dynamic weight adjustment parameters, namely, the data quality weight coefficient, the timeliness weight coefficient, the importance weight coefficient, and the relevance weight coefficient.

[0053] The dynamic weight adjustment parameter is obtained through The weight ratios corresponding to the parameters are obtained, where, Let be the weight ratio of the k-th dynamic weight adjustment parameter corresponding to the i-th spatiotemporal data standard set. This is the initial setting value for the k-th dynamic weight adjustment parameter, which is determined according to step 2.2; This represents the time decay coefficient of the k-th dynamic weight adjustment parameter; when the k-th dynamic weight adjustment parameter is the data quality weight coefficient... =0.02 / minute, when the k-th dynamic weight adjustment parameter is the timeliness weight coefficient, =0.2 / minute, when the k-th dynamic weight adjustment parameter is the importance weight coefficient, =0.001 / minute, when the k-th dynamic weight adjustment parameter is the correlation weight coefficient, =0.05 / minute; The time interval is the time difference between the current moment and the data acquisition moment. It is a natural constant; Based on the corresponding weight ratios of the parameters, the four dynamic weight adjustment parameters are fused together to form a set of fused weight factors that respond to changes in data in real time.

[0054] The specific calculation methods for each factor are as follows: Data quality factor Calculation: ; Wherein: the integrity score of this dimension of data at the current moment: ; Accuracy score of this dimension of data at the current moment: ; Consistency score of this dimension of data at the current moment: ; Timeliness factor Calculation:

[0055] in: The time decay coefficient for this dimension of data is set according to the data type: Transformer operating data: / minute; Environmental monitoring data: / minute; Historical statistics: / minute.

[0056] 2.1.1 Data Quality Factor A comprehensive evaluation is conducted based on three dimensions: data integrity, data accuracy, and data consistency.

[0057] The integrity score is calculated by counting the number of missing data points within a statistical time window (preferably 1 hour) to characterize data integrity. The calculation method is to subtract the number of missing data points from the total number of data points, and then divide by the total number of data points. For example, if a substation should collect 60 data points in 1 hour, and the main station actually obtains 58, the integrity score is 96.7%. When the integrity is lower than 95%, the system automatically reduces the weight of that data source by 20%, and increases the weight of other data sources proportionally to keep the total weight at 1.

[0058] The accuracy of data is assessed by comparing it with standard values ​​(equipment ratings or historical normal operating ranges) or data from neighboring equipment (the 3-5 geographically closest devices) to identify outliers. Data points outside the outlier range are recorded as non-outliers, and the proportion of non-outlier data points to the total data points is calculated. Data reasonableness verification rules are established, such as temperature data should be within the range of -40°C to 80°C. Statistical methods are used to identify outliers; data points exceeding three standard deviations are marked as outliers.

[0059] The consistency assessment is performed by verifying the consistency between data through logical relationships. The consistency score is calculated as 1 - |number of inconsistent data points / total number of verification points|. For example, when checking the power balance relationship, if the input power is not equal to the output power plus loss, it is recorded as an inconsistent data point to ensure the consistency between the input power and the output power plus loss. When verifying the voltage level matching relationship, if the voltage ratio between the upper and lower levels is unreasonable, it is recorded as an inconsistent data point to ensure the reasonable ratio between the upper and lower levels. A comprehensive quality score is calculated by weighting completeness, accuracy, and consistency according to a 4:3:3 ratio, resulting in the final data quality factor.

[0060] 2.1.2 The timeliness factor reflects the decay of data value over time, and is obtained by establishing an exponential decay function based on the natural logarithm: ; in, The time decay coefficient, The time decay coefficient is set differently based on the importance and frequency of change of the data type, representing the time difference between the current time and the data acquisition time. For transformer operating data, a larger attenuation coefficient is set. / minute, so that the data weight drops to 37% of its original value after ten minutes, ensuring real-time requirements; For environmental monitoring data, a moderate attenuation coefficient is set. / minute, the weight of the data drops to 37% of its original value after fifty minutes, reflecting its relatively slow adaptation to environmental changes; For historical fault data, set a smaller attenuation coefficient. / minute, the weight of the data only drops to 37% of its original value after 1,000 minutes, thus maintaining the long-term value of historical data.

[0061] The system automatically calculates the timeliness weight based on the difference between the current time and the data collection time, and makes dynamic adjustments accordingly.

[0062] 2.1.3 Importance Factors Classified according to the criticality of power grid operation: Level 1 Important Data ( =1.0), including the main transformer load, bus voltage, main line current, and the load at the starting end of each branch line on the outgoing side of each transformer; Secondary important data ( =0.8), including the status of power distribution equipment (switch position, protection device status) and secondary equipment information (operation status of relay protection devices (on, off), and communication equipment status); Level 3 Important Data ( =0.6), including environmental monitoring data and auxiliary equipment status (cooling system status (started, not started), lighting system status (on, off)); Level 4 Important Data ( =0.4), including historical fault data and backup equipment information (backup transformer status (started, not started), backup line information (used, not used)).

[0063] The primary critical data includes core operating parameters such as main transformer load, bus voltage, and main line current. The importance coefficient is set to 1.0. These data are directly related to the safe and stable operation of the power grid. Any abnormality may lead to a major accident. The system implements the highest priority processing for primary data to ensure the real-time transmission and processing of data. The secondary critical data includes the status of power distribution equipment and information on secondary equipment, with an importance coefficient set at 0.8. These data affect the operation of the local power grid and play an important role in the overall stability. The three levels of critical data include environmental monitoring data, auxiliary equipment status, etc., with an importance coefficient set at 0.6. These data provide auxiliary information for operational analysis and help with prediction and prevention. The four levels of critical data include historical power grid multidimensional spatiotemporal data, backup equipment information, etc., with an importance coefficient set at 0.4. This data is mainly used for long-term analysis and planning and has little impact on real-time operation.

[0064] 2.1.4 The correlation factor quantifies the correlation strength between the same data series in different time periods using various statistical methods: To obtain the linear correlation between the data, the covariance of the two data sequences is calculated by dividing the product of their respective standard deviations. A correlation coefficient close to 1 indicates a strong positive correlation, close to -1 indicates a strong negative correlation, and close to zero indicates no linear correlation. The two data sequences belong to two time periods, which are preferably [data acquisition time - set second time length, data acquisition time] and [current time - set second time length, current time]. The set second time length is preferably 0.5 to 1 minute.

[0065] Nonlinear correlation analysis characterizes the relationship between two data sequences by calculating their mutual information value. A higher mutual information value indicates a stronger dependency between the two data sequences. Specifically, the mutual information entropy method is used to measure the nonlinear correlation between data. The mutual information value quantifies the degree of information sharing by calculating the expected value of the logarithm ratio of the joint probability distribution to the marginal probability distribution; a higher mutual information value indicates a stronger dependency between the two variables. The correlation factor is obtained by weighting the linear correlation coefficient and mutual information according to a 6:4 ratio. When there is a strong negative correlation The negative value represents a combination method that considers both simple linear relationships and complex nonlinear dependency systems, and can automatically adjust the weight ratio based on the correlation patterns of historical data.

[0066] 2.2 The weight fusion strategy management system establishes a configuration and switching mechanism for multiple fusion strategies, and automatically adjusts the weight allocation according to different operating scenarios; the operating scenarios include normal operation mode, emergency response mode, maintenance and repair mode, and historical analysis mode.

[0067] In the normal operation mode, four factors are considered in a balanced manner with relatively even weight distribution. The preferred initial value range for the data quality weight coefficient is 0.25-0.35, the preferred initial value range for the timeliness weight coefficient is 0.20-0.30, the preferred initial value range for the importance weight coefficient is 0.20-0.30, and the preferred initial value range for the correlation weight coefficient is 0.15-0.25. This ensures that the comprehensive analysis effect is applicable to the normal operation of the power grid, the quality of various data is stable, there are no emergencies, the system automatically monitors the operating status, and switches to this mode when there are no abnormal alarms for one consecutive hour. In the emergency response mode, the weights of timeliness and importance factors are significantly increased to respond quickly to emergencies. When an emergency such as a transformer equipment failure or a sudden load change is detected, this mode is switched to. In this mode, the preferred initial value of the timeliness weight coefficient is increased to 0.4, the preferred initial value of the importance weight coefficient is increased to 0.4, the preferred initial value of the data quality weight coefficient is reduced to 0.1, and the preferred initial value of the correlation factor weight coefficient is reduced to 0.1, ensuring that critical information is processed first. In the maintenance and repair mode, the weight of the data quality factor is increased to ensure data reliability during maintenance. When transformer equipment is being repaired, this mode is switched to. During equipment repair, some data sources may be unstable, requiring stricter quality control. The preferred initial value of the data quality weight coefficient is increased to 0.5, the preferred initial value of the timeliness weight coefficient is set to 0.2, the preferred initial value of the importance weight coefficient is set to 0.2, and the preferred initial value of the relevance weight coefficient is set to 0.1. In the historical analysis mode, this mode is switched to when offline data analysis and long-term pattern mining are required. This mode increases the weight of the correlation factor, deeply explores the correlation of historical data, and is used for offline analysis and long-term pattern mining. It has lower requirements for real-time performance. The preferred initial value of the correlation weight coefficient is increased to 0.4, the preferred initial value of the timeliness weight coefficient is decreased to 0.1, the preferred initial value of the data quality weight coefficient is set to 0.3, and the preferred initial value of the importance weight coefficient is set to 0.2.

[0068] 2.3 Dynamically optimize the four weight coefficients in the dynamic weight adjustment parameters: First, set the initial coefficients according to step 2.2. For example, when the data acquisition is in normal operating mode, set the data quality weight coefficient. Timeliness weighting coefficient Importance weight coefficient Relevance weight coefficient The weights of the four factors are set to 0.3, 0.25, 0.25, and 0.2, respectively. Data quality weighting coefficient Timeliness weighting coefficient Importance weight coefficient Relevance weight coefficient Dynamic optimization is performed according to preset weight coefficient optimization methods. This invention provides two optional weight coefficient optimization methods, as shown below: The first method for optimizing weight coefficients involves adjusting the dynamic weight adjustment parameters through... The weight ratios corresponding to the parameters are obtained, where, The weight ratio of the k-th dynamic weight adjustment parameter. This is the initial setting value for the k-th dynamic weight adjustment parameter; Let k be the time decay coefficient of the k-th dynamic weight adjustment parameter. The time interval is the time difference between the current moment and the data acquisition moment. It is a natural constant; The second method for optimizing weight coefficients is to optimize parameters using the gradient descent algorithm based on historical prediction accuracy and anomaly detection performance. Neural network models (such as LSTM, CNN, etc.) can be used. The input is the historical weight configuration (the values ​​of the four weight coefficients) and the corresponding prediction accuracy. The output is the optimized four weight coefficients. The partial derivatives of the prediction error with respect to each parameter are calculated to determine the direction of parameter adjustment. The learning rate is set to control the parameter update step size to avoid over-adjustment that could lead to system instability. Constraint management: Ensure that the sum of the four weight coefficients is always equal to one and that all parameters are non-negative; use the Lagrange multiplier method to handle equality constraints (the sum of the four weight coefficients is always equal to one), and use the projected gradient method to handle inequality constraints (all parameters are non-negative).

[0069] Those skilled in the art can select the weight coefficient optimization method according to actual needs. The first weight coefficient optimization method has better processing effect in most scenarios, and will not be elaborated here.

[0070] Step 3: The historically acquired spatiotemporal data standard set is fused and correlated through the fusion weight factor set to obtain the data fusion analysis results, and a spatiotemporal fusion correlation analysis mechanism is constructed to obtain the spatiotemporal fusion analysis rules; the spatiotemporal fusion correlation analysis mechanism includes the contents of steps 3.1 to 3.6.

[0071] 3.1 Construct a three-dimensional spatiotemporal coordinate framework. The three-dimensional spatiotemporal coordinate framework includes a gridded coordinate system with spatial dimensions (longitude coordinates and latitude coordinates) and time dimensions. Each point in the three-dimensional spatiotemporal coordinate framework corresponds to a set of spatiotemporal coordinates, denoted as a spatiotemporal coordinate point (x, y, t), where x is the longitude coordinate, y is the latitude coordinate, and t is time.

[0072] The spatial dimensions are expressed in latitude and longitude coordinates, with an accuracy of 0.0001 degrees (approximately 10 meters). The time dimension uses minute-level time granularity to ensure the accuracy requirements of spatiotemporal analysis; Based on the aforementioned three-dimensional spatiotemporal coordinate framework, a spatiotemporal distribution function for data in each dimension is constructed; the data in each dimension includes transformer operation data and environmental monitoring data. For each dimension of data, 3D interpolation is performed on the corresponding spatiotemporal coordinates within a 3D spatiotemporal framework to construct a continuous function for each dimension. This continuous function is then used as the spatiotemporal distribution function for that dimension. This represents the data of dimension d in a three-dimensional spatiotemporal coordinate frame. The spatiotemporal distribution function value at location, where, To represent a three-dimensional spatiotemporal coordinate point, the spatial location of the data (longitude x, latitude y) is determined using GPS coordinates from a spatiotemporal data standard set, and t is determined using a timestamp from the spatiotemporal data standard set. Used to determine the spatiotemporal location of interpolation points in three-dimensional interpolation.

[0073] 3.2 The dynamic weights are obtained through the spatiotemporal distribution function, and the weighted summation of the spatiotemporal distribution function values ​​and the spatiotemporal correlation matrix is ​​used to generate a spatiotemporal correlation matrix R that reflects the spatiotemporal characteristics of the power grid during operation; the dynamic weights include spatial correlation weight coefficients and temporal correlation weight coefficients.

[0074] Spatial correlation weighting coefficient between the g-th dimension data and the h-th dimension data Calculated using the following formula:

[0075] in, The geographical distance between the g-th dimension data and the h-th dimension data. The electrical distance between the g-th dimension data and the h-th dimension data (referring to the electrical connection distance between the devices corresponding to the g-th dimension data and the h-th dimension data).

[0076] The time correlation weighting coefficient between the g-th dimension data and the h-th dimension data Calculated using the following formula:

[0077] in, , h represents the data timestamps of the g-th dimension and the h-th dimension, respectively. and The time interval between h can be set by those skilled in the art, or it can be based on the time lag coefficient used in the time correlation calculation formula. Determined, that is = ; Let be the time decay constant (where ) The preferred value range is 5-30 minutes, adjusted according to data type and correlation strength requirements: preferably, when both the data timestamp of the g-th dimension and the data of the h-th dimension are transformer operation data. =5 minutes, when the timestamp of the data in the g-th dimension and the data in the h-th dimension are both environmental monitoring data. =15 minutes, when the timestamp of the data in the g-th dimension and the data in the h-th dimension are both historical fault data. =30 minutes; if the data timestamp of the g-th dimension and the data of the h-th dimension are of different data types, they can be adjusted by those skilled in the art according to the actual situation. (·) indicates that the data within the parentheses is exponentially raised to the base e of the natural constant.

[0078] 3.3 Comprehensive assessment of spatial correlation based on geographical and electrical distance; Spatial correlation between the g-th dimension data and the h-th dimension data The calculation formula is:

[0079] in, , , These are the weighting coefficients for geographical distance, electrical distance, and historical interaction frequency (preferred value range: , , + + , The geographical distance between the g-th dimension data and the h-th dimension data. Let g be the electrical distance between the g-th dimension data and the h-th dimension data. The frequency of data interaction between the substations corresponding to the g-th and h-th dimension data within a set third time period, where the data type of either the g-th or h-th dimension data is environmental monitoring data. The value is 0 when neither the data type of the g-th dimension nor the h-th dimension is environmental monitoring data, and when the substations corresponding to the g-th dimension and the h-th dimension are the same. ; This indicates taking the maximum value within the parentheses; This represents the maximum frequency of historical data interaction between the substations corresponding to the g-th dimension data and the h-th dimension data within a set third time period. The third time period is preferably 1 hour.

[0080] Based on the correlation analysis of time series data, the formula for calculating the time correlation between the g-th dimension data and the h-th dimension data is as follows: ; in, and They are the g-th dimension data sequence at time t and t- The data sequence of the h-th dimension at time step, Indicates calculation and The Pearson correlation coefficient; The time lag factor is used; the data sequence at each moment is preferably the data sequence within the interval (the moment at this time - the fourth time length, the moment at this time), and the fourth time length is preferably 5 seconds.

[0081] Time lag coefficient Determined in the following manner: By calculating different time lag coefficients The correlation coefficient between the next two time series is used to find the maximum correlation coefficient value within a preset time window, and the value corresponding to the maximum correlation coefficient value is then used. As a means to calculate the temporal correlation between the g-th dimension and the h-th dimension data. This is used to determine the optimal temporal correlation between the two data sequences. The preferred value range for the preset time window is 3 hours.

[0082] The spatiotemporal correlation strength between each pair of data at time t is calculated based on the temporal and spatial correlation between data in each dimension. A spatiotemporal correlation matrix R is then constructed based on this correlation strength between each pair of dimensions at time t. ; in: Let g be the spatiotemporal correlation strength between the g-th dimension data and the h-th dimension data at time t; The spatial correlation weighting coefficients between the g-th and h-th dimension data are... The time correlation weighting coefficient between the g-th dimension data and the h-th dimension data; The spatial correlation degree between the g-th dimension data and the h-th dimension data; The temporal correlation between the g-th dimension data and the h-th dimension data.

[0083] The spatiotemporal correlation matrix at time t As shown in the following formula: ; Where m is the total number of data dimensions; This indicates the spatiotemporal correlation strength between the first-dimensional data and the second-dimensional data at time t; This indicates the spatiotemporal correlation strength between the first-dimensional data and the second-dimensional data at time t; This represents the spatiotemporal correlation strength between the first-dimensional data and the m-th-dimensional data at time t; This indicates the spatiotemporal correlation strength between the second-dimensional data and the first-dimensional data at time t; This indicates the spatiotemporal correlation strength between the second-dimensional data and the second-dimensional data at time t; This indicates the spatiotemporal correlation strength between the second-dimensional data and the m-th-dimensional data at time t; This represents the spatiotemporal correlation strength between the m-th dimension data and the 1st dimension data at time t; This represents the spatiotemporal correlation strength between the m-th dimension data and the 2nd dimension data at time t; Indicates the first m Dimensional data and the first m Dimensional data at time t The spatiotemporal correlation strength.

[0084] 3.4 Establish a comprehensive spatiotemporal correlation analysis model.

[0085] Using the acquired set of fusion weight factors, various data items in the spatiotemporal data standard set are fused at a unified spatiotemporal granularity to obtain the data fusion analysis results; the spatiotemporal correlation analysis model calculates the comprehensive correlation degree by weighted summation of data in each dimension, and uses the comprehensive correlation degree value as the data fusion analysis result.

[0086] Multiply the data in each dimension by its corresponding fusion weight factor. The weighting coefficients are dynamically adjusted according to the real-time operating scenario, as per steps 2.2 to 2.3. Spatiotemporal distribution function mapping: The data of each dimension after being multiplied by the fusion weighting factor are mapped to specific locations in a three-dimensional spatiotemporal coordinate frame through spatiotemporal interpolation; The correlation matrix fusion is achieved by combining the correlation strength information between data in each dimension of the spatiotemporal correlation matrix R, using the following formula: ; Among them, F This represents the comprehensive correlation value of a spatiotemporal coordinate point with longitude coordinate x, latitude coordinate y, and time coordinate t; d represents the d-th data dimension with longitude coordinate x and latitude coordinate y of the sampling point, d∈[1, ], The total number of data dimensions with longitude coordinates x and dimensional coordinates y for each data sampling point; h represents the h-th data dimension, h∈[1, y]. m ], m This represents the total number of data dimensions. This represents the dynamic weight of the d-th dimension data in the i-th spatiotemporal data standard set at the current moment; This represents the data of dimension d in a three-dimensional spatiotemporal coordinate frame. The spatiotemporal distribution function value at that location; This indicates that the data in dimension d and the data in dimension h are at time d. t The spatiotemporal correlation strength; the data dimensions include data from various dimensions in transformer operation data, data from various dimensions in environmental monitoring data, and load data.

[0087] First, a specified principle is set. By traversing all spatiotemporal coordinate points (x, y, t) in the three-dimensional spatiotemporal coordinate frame, the comprehensive correlation degree value of each coordinate point is calculated. The comprehensive correlation degree refers to the comprehensive evaluation index calculated by combining spatial correlation degree and temporal correlation degree. Then, the comprehensive correlation degree value of the specified spatiotemporal coordinate point is obtained by summation operation, forming a comprehensive correlation degree dataset. Then, sort the comprehensive correlation dataset in descending order and reorder all spatiotemporal coordinate points according to their numerical values ​​from largest to smallest. Then calculate the number of the top 10% of coordinate points. The specific calculation method is: number of the top 10% = total number of coordinate points × 0.1. When the calculation result is a decimal, round it up. The top 10% of spatiotemporal coordinate points are selected from the sorted dataset. The comprehensive correlation values ​​corresponding to these coordinate points are the target data that the system focuses on analyzing. Establish a dynamic coordinate selection threshold mechanism. When a new spatiotemporal coordinate point is added, the system automatically recalculates and sorts the data to ensure that the selection criteria of the top 10% are always maintained. The threshold update frequency is set to once per hour or to be triggered when the number of newly added coordinate points exceeds 5% of the total. The overall correlation coefficient value is used as the result of data fusion analysis.

[0088] This step enables accurate quantification of data correlation strength at different spatial locations and time points, identification of the spatiotemporal propagation patterns and impact range of power grid operation status, and provides accurate spatiotemporal correlation basic data for anomaly detection and fault location. The fusion results are subjected to deep spatiotemporal correlation analysis to obtain the analysis results, which include spatiotemporal statistical analysis, spatiotemporal correlation analysis and evolution analysis. The analysis results were then correlated with spatiotemporal interference attribute factors for iterative analysis to obtain second-generation analysis results. Specifically, correlation modeling involves overlaying the analysis results with one or more spatiotemporal disturbance attribute factors within a unified three-dimensional spatiotemporal coordinate framework. This includes spatially overlaying and temporally aligning the identified equipment aging hotspots with predicted typhoon wind and rainfall field models. A correlation model is then established to quantify the impact of disturbance factors on the analysis results.

[0089] For example, scenario simulation and iterative analysis are based on multiple simulations of different interference scenarios (i.e., the interference factors).

[0090] In Iteration 1, under the current disturbance factors (such as an approaching typhoon or heavy rain), it is determined whether the load in the original hotspot area will shift, the probability of equipment failure will change, and whether the weak links in the power grid will shift or be aggravated.

[0091] In iteration 2, the parameters of the disturbance factor are changed (e.g., the typhoon path is shifted 20 kilometers northward, or the rainfall is increased by 5%), and the above simulation is repeated.

[0092] Iteration 3 simulates extreme cases (such as a once-in-a-century ice storm) or scenarios with multiple interfering factors coupled (such as thunderstorms accompanied by strong winds).

[0093] In each iteration, the system recalculates the correlation index and applies the fusion weight factor set to observe the dynamic changes in the fusion and analysis results.

[0094] After multiple rounds of iterative simulations, the system no longer outputs a static analysis conclusion, but rather a dynamic set of conclusions containing uncertainty and multiple possibilities—the second-generation analysis results. Specifically, these include: This shows the evolution of power grid risks in the spatiotemporal dimensions under different disturbance scenarios.

[0095] Identify the most likely path of deterioration or evolution of the power grid condition when a specific disturbance occurs (e.g., A substation tripping > B line overload > C area voltage collapse).

[0096] It clearly identifies which interference factors are most sensitive to under the current power grid conditions, as well as the spatial locations and time periods of sensitivity.

[0097] Based on iterative analysis, it was determined that corresponding emergency plans need to be activated when the interference threshold is reached.

[0098] Based on the results of the second-generation analysis, the ultimate goal of the entire mechanism is to construct operable spatiotemporal fusion analysis rules, that is, to transform complex analysis results into automated and intelligent decision support tools that can directly guide power grid planning, operation and maintenance.

[0099] Spatiotemporal fusion analysis rules are a collection of IF (possible)-THEN (outcome)-ACTION (behavior) logics, but their conditional parts deeply integrate insights from spatiotemporal context and second-generation analysis results.

[0100] Based on the dynamic risk map, evolution path, and vulnerability assessment obtained from the second-generation analysis results, the following typical actionable rules can be constructed: Category 1, Warning Rules: If the real-time weather forecast shows that the typhoon center will pass through the grid within the next 6 hours, and the grid is identified as a strong wind sensitive equipment gathering area in the second-generation analysis results, and the predicted wind speed is greater than the threshold, then a special inspection warning work order will be automatically generated, executed and dispatched to the operation and maintenance team, and prompts the key inspection list A, B, and C.

[0101] The warning rules directly triggered the operation and maintenance actions, specifying the exact geographical location and equipment.

[0102] Category Two, Operational Control Rules: If the wildfire risk index shows area P re The risk level is extremely high in the next 3 hours, and the second-generation analysis results indicate that if lines L1 and L2 are out of service under these circumstances, it will lead to a load transfer path N. path If the overload risk is greater than 90%, an optimal power flow adjustment plan will be pre-formulated and recommended to the dispatcher or executed automatically: the initial load on the path will be reduced in advance, and backup power will be activated. Operation control rules directly provide preventative control strategies for power grid dispatch, translating analytical insights into specific operational instructions.

[0103] Category 3: Planning Auxiliary Rules: If, based on multi-year load data and evolutionary analysis of urban expansion factors, area Q... re The annual load growth rate has consistently exceeded 8%, and simulation results from second-generation analysis indicate that the existing main transformer will experience two summer peaks, N path If the -1 check fails, a power grid reinforcement recommendation is generated, and a project proposal is automatically created in the planning system, along with a recommended set of candidate site coordinates. This rule provides a direct, data-driven basis for the medium- and long-term planning and investment decisions of the power grid.

[0104] Category 4, Adaptive Optimization Rules: If iterative analysis reveals that, under cold and damp winter weather, the weight factor W of the first equipment parameter in the existing state estimation model... eq1 It needs to be adjusted to the second equipment parameter weighting factor W eq2Only by updating the corresponding parameters in the set of fusion weight factors can the true state be more accurately reflected, thus enabling the entire spatiotemporal fusion correlation analysis mechanism to have self-learning and self-adaptive capabilities.

[0105] The rules optimize the parameters of the analysis mechanism itself, enabling continuous improvement of the system.

[0106] This forms a closed loop from analytical insights to intelligent decision-making. Iterative analysis of spatiotemporal interference attribute factors is a process of deduction and thinking, which, by introducing complex external variables, enables the system to anticipate various possible future scenarios, thereby obtaining more profound and robust second-generation analysis results.

[0107] Constructing operable spatiotemporal fusion analysis rules transforms the results of thinking into conditioned reflexes and muscle memory. It establishes clear behavioral guidelines, enabling the power grid management system to take automatic, rapid, and accurate actions when faced with specific spatiotemporal signals, ultimately achieving the goal of safe, reliable, and efficient operation of the power grid.

[0108] 3.5 Based on the data fusion analysis results, spatiotemporal fusion analysis rules are automatically generated; the spatiotemporal fusion analysis rules include spatiotemporal statistical analysis rules, spatiotemporal correlation analysis rules, and evolutionary analysis rules.

[0109] The spatiotemporal statistical analysis rules described in 3.5.1 conduct in-depth analysis of the statistical indicators of the mean distribution, variance characteristics, and spatial autocorrelation of the comprehensive correlation degree, quantify the spatiotemporal distribution characteristics and change patterns of the power grid operation status, including spatiotemporal mean distribution sub-rules, spatiotemporal variance characteristic sub-rules, and spatial autocorrelation sub-rules.

[0110] The aforementioned spatiotemporal mean distribution sub-rule, through statistical analysis of the average distribution pattern (normal distribution, skewed distribution, uniform distribution, etc., determined by Kolmogorov-Smirnov test) of the comprehensive correlation degree at each spatiotemporal coordinate point in different spatial regions (divided according to administrative divisions or power grid zones, preferably 5km×5km grid), calculates the average value based on 30 days of historical data, and discovers the spatial clustering characteristics (such as continuous area, clustering density, etc.) of high-value areas (average value exceeds historical average + 1 standard deviation) and low-value areas (average value is lower than historical average - 1 standard deviation), providing a basis for load forecasting and equipment configuration; The aforementioned spatiotemporal variance feature sub-rule quantifies the fluctuation and stability level of the comprehensive correlation in each region at the current moment through real-time variance quantification. Regions with variance greater than the historical variance plus 2 standard deviations are identified as "unstable regions," providing guidance for risk assessment and preventive maintenance. The spatial autocorrelation sub-rule analyzes the similarity of spatiotemporal fusion data between adjacent regions, and is calculated using Moran's index I:

[0111] in, The total number of spatial regions is represented by r; r represents the r-th spatial region, r∈[1, ]; j represents the j-th spatial region, j∈[1, xr and xj are the sum of the comprehensive correlation at each spatiotemporal coordinate point in spatial region r and the sum of the comprehensive correlation at each spatiotemporal coordinate point in spatial region j, respectively. This represents the average comprehensive correlation across all spatial regions. The spatial weight represents the distance between spatial regions r and j, defined as the reciprocal of the distance between the center points of the two spatial regions r and j; the closer the distance, the greater the weight. W represents all spatial weights. The sum; This indicates summing all terms corresponding to variable r; This indicates that the summation is performed on all terms corresponding to variable j.

[0112] A similarity value greater than 0.3 is considered a "spatial cluster with similar operational characteristics," providing support for the formulation of regional coordinated control strategies. Those skilled in the art can determine the scope of load transfer based on the "spatial cluster with similar operational characteristics" when a substation corresponding to a load experiences a fault, or when the power grid is not operating normally. Furthermore, they can determine the selectable regions for load transfer based on spatiotemporal variance characteristic sub-rules (selecting regions with variance less than a set variance threshold). The power grid operating status includes normal operating status, warning status, abnormal status, and fault status.

[0113] The spatiotemporal correlation rules described in 3.5.2 are used to describe the mutual influence and propagation patterns of various equipment, regions, and system levels in the spatiotemporal dimension of the power grid, and to establish a complete correlation mapping system, including propagation delay sub-rules, load transfer sub-rules, and seasonal correlation sub-rules.

[0114] The propagation delay sub-rule refers to the spatiotemporal distribution function of transformer operation data, environmental monitoring data, and load data based on the data fusion analysis results. Combined with the spatiotemporal correlation matrix Spatial correlation weight coefficient ws Time-related weighting coefficient wt This study analyzes the spatiotemporal propagation characteristics of abnormal states in the power grid, such as substation equipment failures, transmission line failures, changes in load hotspots, and voltage fluctuation propagation. These characteristics include propagation path (along electrical connection paths and geographically proximate paths), propagation speed (fault propagation speed is typically 60%-80% of the electromagnetic wave speed), and impact range (quantified by the number of affected devices and geographical area, in units of number of units and square kilometers). Weighting factors are then integrated into the analysis. We weighted the data across all dimensions, used graph theory algorithms to analyze the propagation path, and established a directed graph structure for fault propagation. Nodes represent substation equipment, edges represent propagation paths, and weights represent propagation time. The spatiotemporal coordinates of the top 10% of the data with the highest overall correlation were used to identify key propagation nodes.

[0115] The load transfer sub-rule is based on the spatiotemporal coordinate points of the top 10% of the comprehensive correlation results in the data fusion analysis, utilizing the spatiotemporal distribution function of data in each dimension of the three-dimensional spatiotemporal coordinate framework. Identify periodic patterns of power grid operation at different time scales, including daily load curves, weekly load patterns, and seasonal variations, using a spatiotemporal correlation matrix. The analysis of the periodicity of time series by time correlation weighting coefficients can be achieved using the SARIMA model, with the model input being the fused weighting factors. The weighted historical load data and the comprehensive correlation F(x,y,t) are used to output periodic pattern parameters, combined with the spatial correlation Rs. Relationship with time Rt The comprehensive evaluation provides a reference for the formulation of operation plans and the optimization of resource allocation; The load correlation sub-rule is based on the comprehensive correlation degree value in the data fusion analysis results, utilizing the spatiotemporal correlation matrix. By analyzing the correlation strength information among data from various dimensions, we can determine the correlation and transmission mechanism of load changes in different regions, and use the spatial correlation degree Rs Relationship with time Rt Combining the spatiotemporal distribution function in the three-dimensional spatiotemporal coordinate framework A spatiotemporal network model of load transmission is established. This model is constructed based on the results of data fusion analysis. Through dynamic weight adjustment and key analysis of the top 10% comprehensive correlation coordinate points, the load transmission mechanism is accurately modeled.

[0116] The evolution analysis rules described in 3.5.3 are used to detect and track the dynamic changes of the spatiotemporal fusion analysis results in real time, and to obtain the trend prediction and anomaly diffusion simulation analysis of the power grid operation status, including trend prediction sub-rules, anomaly diffusion sub-rules and periodicity sub-rules.

[0117] The trend prediction sub-rule is based on the changing trend of historical data fusion analysis results to establish a prediction model for future operating status. It can adopt an LSTM neural network model, with the input being a historical spatiotemporal fusion data sequence at a certain position (x,y) and the output being the predicted value at a future time. The aforementioned abnormal diffusion sub-rule establishes an impact range assessment model for abnormal events based on the diffusion mechanism (diffusion along electrical paths, diffusion near geographical proximity) and propagation law (exponential decay law, linear propagation law) of abnormal state types (equipment overload, voltage exceeding limits, frequency deviation, etc.) in the spatiotemporal dimension. It can adopt a diffusion equation model, with the initial location of the abnormal event and the abnormal state type (the abnormal state type is presented in data form after binary encoding) as input, and the output as the impact radius (in meters) and the impact degree (the impact degree includes level one, level two, and level three, with level one indicating the greatest impact degree and level three indicating the least impact degree), providing a scientific basis for emergency response. The periodic sub-rules identify periodic patterns in the power grid's operating status at different time scales, including daily load curves, weekly load patterns, and seasonal changes. They can be implemented using Fourier transform methods, with historical operating data time series as input and periodic components and frequency characteristics as outputs, providing a reference for operation planning and resource allocation optimization.

[0118] 3.6 By establishing a rule priority and applicable condition discrimination mechanism, the scenarios, priority order and automatic triggering conditions of three different types of rules are clarified to ensure the effectiveness and practicality of the rule set in complex power grid environments.

[0119] Priority ranking: Spatiotemporal statistical analysis rules (priority 1) > Evolutionary analysis rules (priority 2) > Spatiotemporal correlation analysis rules (priority 3); Application scenarios: Spatiotemporal statistical analysis rules are suitable for daily operation monitoring, evolutionary analysis rules are suitable for fault early warning and emergency response, and spatiotemporal correlation analysis rules are suitable for in-depth analysis and pattern mining; Automatic triggering conditions: Spatiotemporal statistical analysis rules are automatically triggered when data is updated, evolutionary analysis rules are triggered when anomalies are detected, and spatiotemporal correlation analysis rules are triggered manually or on a scheduled basis when in-depth analysis is required.

[0120] Step 4: Obtain the power grid operation strategy by integrating the power grid operation principle based on spatiotemporal fusion analysis rules; wherein, the power grid operation strategy is obtained by integrating the power grid operation principle based on spatiotemporal fusion analysis rules, and abnormal power grid operation conditions are generated through a multi-level anomaly detection mechanism constructed by the power grid operation strategy; 4.1 A power grid state prediction model is constructed based on multidimensional spatiotemporal data of the power grid. The power grid state prediction model takes multidimensional spatiotemporal data of the power grid as input and predicts the power grid operating state in the next time period (preferably 15 minutes). The multidimensional spatiotemporal data of the power grid includes transformer operating data, environmental monitoring data, and historical fault data. The power grid operating state includes normal operating state, warning state, abnormal state, and fault state. These operating states are divided according to the threshold range of key operating parameters. Those skilled in the art should know how to set each threshold, which will not be elaborated here. State prediction is performed based on the power grid state prediction model, and prediction results are obtained. Obtain real-time operating status data at the current moment; The prediction deviation (calculated using mean square error) is obtained by comparing the degree of agreement between the prediction results and the real-time operating status data. The prediction bias is fed back into the power grid state prediction model for iterative training to output high-precision prediction results; Based on high-precision prediction results, the causal chain and spatial correlation of time series are obtained, the dependence of state data is quantified, and a spatiotemporal correlation network covering the entire power grid is constructed. Specifically, the temporal causality analysis employed the Granger causality test, and the test results... The calculation formula is:

[0121] in, For the sum of squared residuals of the constrained model, For the sum of squared residuals of the unconstrained model, The lag order is... The sample size is given; those skilled in the art should know how to obtain the above data, and will not elaborate further here.

[0122] The spatial correlation degree calculation is the same as in step 3.3, as shown in the following formula:

[0123] in, , , These are the weighting coefficients for geographical distance, electrical distance, and historical interaction frequency (preferred value range: , , + + , The geographical distance between the g-th dimension data and the h-th dimension data. Let g be the electrical distance between the g-th dimension data and the h-th dimension data. The frequency of data interaction between the substations corresponding to the g-th and h-th dimension data within a set third time period, where the data type of either the g-th or h-th dimension data is environmental monitoring data. The value is 0 when neither the data type of the g-th dimension nor the h-th dimension is environmental monitoring data, and when the substations corresponding to the g-th dimension and the h-th dimension are the same. ; This indicates taking the maximum value within the parentheses; This represents the maximum frequency of historical data interaction between the substations corresponding to the g-th dimension data and the h-th dimension data within a set third time period. The third time period is preferably 1 hour.

[0124] The degree of interdependence of state data is quantified by mutual information, and the calculation formula is as follows:

[0125] in, This represents the mutual information value between the state variables of node q and node n; q represents the q-th node, q∈[1, ..., n]. ]; n represents the nth node, n∈[1, ]; This represents the total number of nodes, where nodes refer to substation equipment, consistent with the propagation delay sub-rule. , Let be the state variables of node q and the state variables of node n, respectively. Let be the joint probability density of the state variables of node q and node n. , Let be the marginal probability density of the state variables of node q and node n; This indicates summing all terms corresponding to variable q; This indicates that the summation is performed on all terms corresponding to variable n.

[0126] Based on the spatiotemporal fusion analysis rules in step 3.5, the spatiotemporal statistical analysis rules, spatiotemporal correlation analysis rules, and evolutionary analysis rules are applied to the power grid operation state prediction and decision-making process, as follows: Spatiotemporal statistical analysis rules use spatiotemporal mean distribution sub-rules to identify the spatial distribution characteristics of power grid operating parameters, use spatiotemporal variance characteristic sub-rules to quantify the operational stability of each region, and use spatial autocorrelation sub-rules to analyze the cooperative operational characteristics of adjacent regions. These analysis results are used to formulate regional operation strategies and load allocation schemes. The spatiotemporal correlation analysis rules establish a fault propagation path prediction model by using the propagation delay sub-rule, optimize the long-term operation plan by using the seasonal correlation sub-rule, and construct the inter-regional load transmission mechanism by using the load correlation sub-rule. These analysis results are used to formulate preventive maintenance plans and emergency response strategies. The evolutionary analysis rules establish a forward-looking prediction model of the power grid operation status based on the trend prediction sub-rules, use the anomaly diffusion sub-rules to assess the impact range of potential risks, and optimize the operation and scheduling plan through the periodic sub-rules. These analysis results are used to formulate medium- and long-term development plans and real-time scheduling decisions. By comprehensively operating the spatiotemporal fusion analysis rules, a comprehensive analysis result including spatial distribution characteristics, temporal evolution trends, and correlation propagation mechanisms is obtained. This result is used to guide the formulation of power grid operation strategies, optimize resource allocation schemes, and improve system operation efficiency and reliability. A comprehensive power grid operation system is formed based on the operation strategy and decision-making results. The comprehensive power grid operation system includes normal operation mode, emergency response mode and fault handling mode.

[0127] 4.2 Predict the power grid operation status based on the spatiotemporal fusion analysis rules.

[0128] 4.2.1 First, the power grid operating status estimate is updated.

[0129] To update the power grid operating state predicted by the power grid state prediction model based on Kalman gain, it is necessary to pre-set or initialize the state transition matrix, observation matrix, process noise covariance matrix, observation noise covariance matrix, and initial state estimate. Those skilled in the art should know how to set these parameters, so they will not be elaborated here.

[0130] Specifically, firstly, the optimal gain is calculated based on the prior error covariance matrix, the observation matrix, and the observation noise covariance matrix to ensure that the estimation error is minimized; The prior error covariance matrix is ​​updated based on the Kalman gain and the observation matrix to obtain the posterior error covariance matrix, which provides uncertainty information for the prediction of the next time step. The power grid operating state is used as a priori state estimate. The priori state estimate is fused with the observed values ​​based on Kalman gain to obtain the posterior state estimate.

[0131] 4.2.2 Perform adaptive noise estimation and dynamically adjust the noise covariance based on historical data.

[0132] The process noise covariance in the Kalman filter is adaptively adjusted by dynamically updating the process noise covariance in the process noise model using a weighted averaging method that combines the product of historical noise covariance and the current state prediction residual. The weighting coefficients control the fusion ratio of historical information (historical noise covariance) and current information (the product of the current state prediction residual), and can be set by those skilled in the art according to actual conditions; this will not be elaborated here. Specifically, the process noise covariance at the current time (time z) is selected from the most recent 100 historical data points. The calculation formula is:

[0133] Wherein, ρ is the weighting coefficient (preferably ranging from 0.8 to 0.95). Let E[·] be the process noise covariance at time z-1, and E[·] denote the expected value within the brackets; vz is the predicted residual for the current state. The transpose of the residual is predicted for the current state; the iteration stops when the noise covariance of vz changes by less than 0.01 over 10 consecutive iterations, and the initial value is preferably in the range of 0.1-1.0; The observation noise covariance in the Kalman filter is adaptively adjusted. Based on the statistical characteristics of the observation residuals, the observation noise covariance matrix is ​​dynamically adjusted, and a weighted average is used to balance historical noise characteristics and current observation error characteristics. The weights can be set by those skilled in the art according to the actual situation, and will not be elaborated here. In this invention, prediction residual and prediction bias have the same meaning.

[0134] This adaptive mechanism can automatically adjust the filtering parameters according to changes in the power grid operating environment, improve the algorithm's adaptability to different operating conditions, and ensure good filtering performance under various noise environments.

[0135] 4.2.3 Multi-model adaptive estimation handles multiple operating modes.

[0136] According to step 4.1, the power grid operating status includes normal operating status, warning status, abnormal status, and fault status. This invention constructs a prediction model for each of the power grid operating statuses (including normal operating status model, fault mode model, etc., with the current system status and observation data as inputs, and the outputs being the probability of normal operating status (range from 0 to 1), the probability of fault status (range from 0 to 1), etc.).

[0137] Based on multiple models, mode probability updates are performed. Specifically, the posterior probability of each operating state is calculated based on the likelihood function value and mode transition probability matrix of each model (including normal operation model, fault mode model, load change model, etc., with the current system state and observation data as inputs and the probability of power grid operating state as outputs). Normalization is used to ensure that the probability sums to one. 4.2.4 Perform abnormal state detection.

[0138] The predicted operating state probabilities from various models (including normal operation models, failure mode models, and load mutation models) are stored as vectors, denoted as state vectors. Anomaly detection is performed by monitoring the statistical characteristics of these state vectors. Specifically, when the L2 norm of the state vector exceeds the dynamic norm threshold calculated based on the state covariance matrix, the system automatically triggers the anomaly detection mechanism, initiating corresponding fault diagnosis procedures (such as equipment inspection procedures and load transfer procedures) and emergency response procedures (such as emergency power outage procedures and backup equipment activation procedures). The dynamic norm threshold... The calculation formula is: ; in, The baseline threshold is set at 2.0-3.0. This is the adjustment coefficient for the covariance matrix (preferably ranging from 1.5 to 2.5). Let be the state covariance matrix at time z. is the trace of the covariance matrix.

[0139] 4.3 Time series forecasting based on the SARIMA seasonal autoregressive integral moving average model: Time series forecasting is performed using the SARIMAX model (Seasonal Autoregressive Integral Moving Average Exogenous Variable Model), which is an extension of the basic ARIMA autoregressive integral moving average model. Specific improvements include: The SARIMA model adds a seasonal parameter s to the basic ARIMA(P,D,Q) model to obtain the SARIMAX model (P,D,Q,s), where P is the seasonal autoregressive order, D is the seasonal differencing order, Q is the seasonal moving average order, and s is the seasonal cycle length. This is achieved through seasonal differencing... The sD and seasonal autoregressive moving average terms capture multiple periodic patterns of power grid load, including daily (s=24), weekly (s=168), and monthly (s=720) cycles. Multiple periodicity processing: In view of the characteristic that the power grid load has multiple periodicities of daily, weekly and monthly, a multiple seasonal decomposition method is adopted. First, the daily periodicity modeling is performed by SARIMA, then the residual sequence is modeled by weekly periodicity, and finally the secondary residual is modeled by monthly periodicity, so as to achieve accurate capture of complex periodicity patterns. Exogenous variable integration: External factors such as meteorological data (temperature, humidity, wind speed, etc.), holiday information, and economic activity indicators are incorporated into the model as explanatory variables to form the SARIMAX model structure.

[0140] Specifically, the seasonal component captures the daily, weekly, and monthly periodic variations of the power grid load by introducing seasonal differences and seasonal autoregressive and moving average terms. Exogenous variable integration incorporates external factors such as ambient temperature, humidity, and holiday markers into the model as exogenous variables, thereby improving prediction accuracy. The model parameters are adaptively adjusted dynamically based on the prediction error to ensure the model's adaptability to changes in the power grid's operating status. The forecast results outputs load forecast curves for the next 24 hours, equipment operating status predictions, and potential risk point identification, providing a scientific basis for operation scheduling.

[0141] 4.4 Construct a spatiotemporal correlation network to establish a correlation network covering the entire power grid area, including: Using the spatiotemporal fusion analysis rule set established in step 3.5, the system automatically executes the following analysis process: The spatiotemporal statistical analysis rules identify high-value and low-value regions of power grid operation through spatiotemporal mean distribution sub-rules, generating regional operation characteristic distribution maps; quantify the fluctuation degree of operation parameters in each region through spatiotemporal variance characteristic sub-rules, identify unstable regions and generate stability assessment reports; and calculate the similarity of operation parameters between adjacent regions through spatial autocorrelation sub-rules, identifying spatial clusters with similar operation characteristics, providing data support for the formulation of regional coordinated control strategies. The spatiotemporal correlation analysis rules analyze the spatiotemporal propagation characteristics of faults or abnormal states in the power grid through propagation delay sub-rules, establish a spatiotemporal model of fault propagation, identify key propagation nodes and propagation paths, and provide decision-making basis for emergency response and fault isolation. Through seasonal correlation sub-rules, the rules identify periodic patterns of power grid operation status, output periodic pattern parameters, and provide reference for operation plan formulation and resource allocation optimization. Through load correlation sub-rules, the rules analyze the correlation and transmission mechanism of load changes in different regions, establish a spatiotemporal network model of load transmission, and provide support for load scheduling and demand response strategy formulation. The evolutionary analysis rules establish a predictive model of future operating states through trend prediction sub-rules, outputting predicted values ​​for future moments to provide a basis for the formulation of forward-looking operating strategies; they establish an impact range assessment model for abnormal events through anomaly diffusion sub-rules, outputting the impact range and degree to provide a scientific basis for determining the scope of emergency response and resource allocation; and they identify periodic patterns of power grid operating states through periodicity sub-rules, outputting periodic components and frequency characteristics to provide a reference for optimizing operating plans and formulating equipment maintenance plans. A spatiotemporal correlation network is constructed based on the results of rule application: Node definition: Key equipment such as substations, power plants, and important load centers are used as network nodes, and node attributes include geographical coordinates, electrical parameters, and operating status; The correlation strength between nodes is calculated by combining electrical distance, correlation strength, and historical interaction frequency, and is used as the edge weight between the nodes. The formula for calculating the edge weight between two nodes is as follows:

[0142] in, This represents the edge weight between node u and node v; , , These are the weight coefficients for each node (preferred value range: , , , Let u be the electrical distance between node u and node v. Let represent the correlation strength between node u and node v (using the Pearson correlation coefficient). The historical interaction frequency between the substations corresponding to node u and node v; Dynamic network topology update: Adjusting the network structure in real time according to changes in power grid operation mode, including updating node status, recalculating edge weights, and correcting topology relationships; Spatiotemporal correlation network application output: Generates a spatiotemporal correlation map of the entire power grid, identifies key nodes and weak links, and provides comprehensive decision support for power grid planning, operation optimization and risk prevention and control. Specifically, it includes key propagation path identification results, regional coordinated control strategy suggestions, load dispatch optimization schemes and emergency response plans.

[0143] This interconnected network can accurately reflect the mutual influence between different parts of the power grid, providing a networked analytical basis for fault propagation analysis, load transfer strategies, and coordinated control.

[0144] 4.5 Based on the prediction results and correlation analysis in step 4.2, formulate specific power grid operation strategies: 4.5.1 Establish a load allocation optimization strategy, specifically, to achieve optimal load allocation based on a multi-objective optimization model; Objective function design: With the optimization objectives of minimizing network loss, voltage deviation, and operational deviation, a comprehensive objective function is constructed using a weighted summation method.

[0145] in, This means minimizing F; For network active power loss, This is the active power loss weighting coefficient. This refers to the voltage deviation (the difference between the voltage in the operating strategy and the rated voltage). This represents the sum of squares of the voltage deviations at each voltage sampling point. This is the voltage deviation weighting coefficient; The difference between the power in the power deviation operation strategy and the predicted power (the predicted power can be obtained from existing power prediction models or step 4.3). Power deviation weighting coefficient (preferred value: , , The weighting coefficients are determined using a multi-objective optimization algorithm based on historical operating data. Constraint settings: Power balance constraint: Ensure that the total power generation of the system equals the total power consumption of the load plus the active power loss of the network; Equipment capacity constraints: Ensure that all equipment (transformers, lines, generators, etc.) operates within its rated capacity range; Voltage quality constraints: Maintain the voltage of each node within the allowable range. Under normal operation, the allowable range of voltage for each node is 0.95pu-1.05pu, and under emergency conditions, the allowable range is 0.90pu-1.10pu. Solution algorithm implementation: An improved particle swarm optimization algorithm is used to solve multi-constraint optimization problems. The global optimal solution is found through the cooperative search of the particle swarm. This optimization strategy can achieve optimal load allocation under various operational constraints, thereby improving the economy and safety of power grid operation.

[0146] Based on the spatiotemporal fusion analysis rule set established in step 3.5, the system automatically executes the rule matching and application process: The rule matching mechanism automatically matches applicable analysis rules based on the current power grid operating status and spatiotemporal data characteristics. The matching priority is executed according to the priority level set in step 3.6, i.e., spatiotemporal statistical analysis rules (priority 1) > evolution analysis rules (priority 2) > spatiotemporal correlation analysis rules (priority 3). The rule execution result indicates that a regional operation characteristic report is generated after the spatiotemporal statistical analysis rule is executed. This report includes a high-value area distribution map, unstable area identification results, and spatial cluster analysis results. This report is used to guide daily operation monitoring and equipment configuration optimization. After the evolutionary analysis rules are executed, trend prediction reports, anomaly spread risk assessment reports, and periodic operation mode reports are generated. These reports are used for fault early warning, emergency response decision-making, and operation plan formulation. After the spatiotemporal correlation analysis rules are executed, a propagation path analysis report, a seasonal correlation pattern report, and a load correlation network diagram are generated. These results are used to deeply analyze the power grid operation patterns, formulate regional coordinated control strategies, and optimize load dispatching schemes. Rule fusion and comprehensive decision-making integrates the execution results of the three types of rules and generates comprehensive operational strategy suggestions through a weighted synthesis method. The weight allocation is dynamically adjusted according to the current operational scenario. Under normal operation, the weight of statistical analysis is 0.6, the weight of evolutionary analysis is 0.2, and the weight of correlation analysis is 0.2. Under abnormal conditions, the weight of evolutionary analysis is increased to 0.5, the weight of statistical analysis is 0.3, and the weight of correlation analysis is 0.2. The decision support output is based on the rule fusion results. The system automatically generates specific operational strategy suggestions, including equipment scheduling instructions, load transfer schemes, and maintenance plan adjustment suggestions. These suggestions are directly used to guide the actual operation and decision-making of power grid operators.

[0147] 4.5.2 Develop an intelligent equipment emergency dispatch plan: Based on the operational strategy scheme, a priority ranking mechanism is established. When multiple operational strategy schemes conflict, a decision is made according to the safety priority order to obtain the decision result. The priority matrix construction comprehensively considers factors such as equipment importance, current operating status, and maintenance plan to establish a priority evaluation system for equipment scheduling. The priority matrix is ​​shown in the following formula: P(a,b)=w1×importance(a)+w2×status(b)+w3×maintenance(a,b); Where a represents the a-th transformer device, a∈[1, ]; Let be the total number of transformer devices; b represents the b-th power grid operating state, b∈[1, 2, 3]. ]; The total number of transformer equipment; importance(a) is the importance score of equipment a, status(b) is the score of grid operating status b, and maintenance(a,b) is the maintenance plan impact score of equipment a under grid operating status b. Optimize the scheduling sequence: Take into account the time interval requirements and mutual influence between various operations, and formulate the optimal scheduling time sequence emergency plan. For example, load transfer should be carried out first and equipment maintenance should be carried out to ensure that the load transfer is completed safely before the equipment is shut down; or standby equipment should be put into operation first and faulty equipment should be disconnected to ensure that the power grid supply continuity is not affected. The system employs a rapid response strategy with multiple pre-set fault scenarios. When an anomaly is detected, the corresponding emergency dispatch plan is automatically activated. For example, when a transformer overload anomaly is detected (load rate exceeds 95%), the system automatically activates the load transfer plan, transferring part of the load to a nearby transformer and simultaneously activating the standby transformer. When a line fault anomaly is detected (line current suddenly drops to zero or protection trips), the system automatically activates the standby line activation plan, quickly switching to the standby power supply path to ensure power continuity. This dispatch strategy enables intelligent management and coordinated control of equipment, improving the reliability of power grid operation and emergency response capabilities.

[0148] Step 5: Based on the power grid operation strategy, construct a multi-level anomaly detection mechanism to generate abnormal operating conditions; the multi-level anomaly detection mechanism identifies subtle abnormal changes in abnormal power grid operation through multi-scale anomaly detection; the multi-scale anomaly detection includes point anomaly detection, pattern anomaly detection, and group anomaly detection; The threshold for triggering abnormal operating conditions is set according to a multi-discrimination mechanism, including multiple discrimination thresholds such as point abnormality threshold, continuous abnormal duration threshold, and abnormal node number threshold. In steps 1 to 4, the current power grid operation strategy is constructed based on the spatiotemporal fusion analysis rules obtained from the current spatiotemporal data standard set. This step runs the power grid operation strategy and detects the execution process of the operation strategy through a multi-level anomaly detection mechanism to obtain anomaly detection results.

[0149] 5.1.1 Establish a multi-level anomaly detection mechanism. This involves acquiring a spatiotemporal data standard set generated during the current power grid operation in real time, and performing multi-scale anomaly detection based on the spatiotemporal fusion analysis rules of the spatiotemporal data standard set. The multi-scale anomaly detection includes point anomaly detection, pattern anomaly detection, and group anomaly detection, specifically: Point anomaly detection identifies numerical anomalies in a single data point at a specific moment, applicable to scenarios such as sudden changes in transformer equipment parameters and instantaneous sensor failures. Those skilled in the art should know how to implement point anomaly detection. The following point anomaly detection methods are merely two preferred embodiments provided by this invention; those skilled in the art can construct point anomaly detection methods according to actual conditions.

[0150] One method for point anomaly detection is as follows: Point anomaly detection is based on the assumption that anomalies are more easily isolated. It constructs multiple binary trees (based on historical feature data) by randomly selecting features (chosen from core operating parameters such as voltage deviation, load rate change, frequency fluctuation, and power factor anomalies) and split points (feature data values ​​used to divide the selected feature data dataset into two subsets). Data on these features is collected in real-time from each transformer, resulting in multiple data points. The average path length of each data point is calculated, and a point anomaly score is calculated based on the average path length. Data points with a point anomaly score less than a point anomaly threshold are considered to have a point anomaly. The calculation formula is:

[0151] in, Indicates the first One sample, ∈[1, ]; Indicates sample The total length of paths in all binary trees; For the sample Average path length in all binary trees The total number of samples is The standardization factor for time is calculated based on the harmonic number: ,in The total number of samples is The harmonic number at time, in this invention the preferred value of the harmonic number is (Euler's constant) The total number of samples; When selecting features, the selection can be random, or a method for selecting features can be developed by those skilled in the art based on the actual situation. In a preferred embodiment of the present invention, the selection can be performed according to the following method: Prioritize key features sensitive to the power grid's operating status. These key features include voltage deviation between adjacent time points, load rate changes within 5 minutes, frequency fluctuations within 1 minute, and power factor anomalies (defined as a power factor below 0.9). Multiple binary trees can be constructed for each of these key features, and point anomalies corresponding to each feature can be identified. Alternatively, one key feature can be selected from the above key features to construct multiple binary trees, and the point anomalies corresponding to that feature can be identified using the following method: The most discriminative features are selected based on mutual information and analysis of variance. The analysis of variance uses the F-test to improve the accuracy of anomaly detection. Tree construction optimization: When constructing isolated trees, a weighted random selection strategy is adopted. The probability of a feature being selected is adjusted according to the feature importance weight (calculated through information gain), and the split value is randomly selected within the feature value range. The point anomaly threshold can be adjusted by those skilled in the art according to the actual situation, or it can be dynamically adjusted according to the following methods: The point anomaly threshold is dynamically adjusted based on historical anomaly detection results, as shown in the following formula: Point anomaly threshold = baseline point anomaly threshold × (1 + ×Standard deviation of key features+ ×Adaptive factors for key features); The key features include core operating parameters such as voltage deviation between adjacent time points, load rate changes within 5 minutes, frequency fluctuations within 1 minute, and a power factor below 0.9 indicating an abnormal power factor; the baseline abnormal threshold is 0.5. The standard deviation of the feature is the weight; For feature adaptive weights; preferably, The value range is 0.1-0.3. The value range is 0.05-0.15, which can adapt to changes in the power grid operation status; the adaptive factor of the key feature can be set by those skilled in the art according to the actual situation, and its preferred value range is 30% of the mean value of the key feature.

[0152] Another method for point anomaly detection is as follows: A real-time monitoring method based on dynamic statistical thresholds is adopted. For each feature data (including voltage deviation between adjacent time points, load rate change within 5 minutes, and frequency fluctuation within 1 minute), the system maintains a sliding time window (e.g., the most recent 30 sampling points), calculates the mean and standard deviation of the data within the window in real time, and calculates the deviation of the current data point by applying Z-score standardization or the improved median absolute deviation (MAD) method. A feature data point is judged as an anomaly when it meets the following condition: |X ob,z - |>K×

[0153] Among them, X ob,z Let z be the observed value of the feature data at the current time, and K be the anomaly detection sensitivity coefficient. This coefficient can be set by those skilled in the art, or it can be dynamically adjusted according to the false alarm rate of historical data (its preferred value range is 2.5 to 3.5). This represents the mean of the feature data within the current time window. This represents the standard deviation of the feature data within the current time window.

[0154] For key parameters with known normal ranges (such as voltage and frequency), they can be directly compared with preset safe operating upper and lower limits to achieve millisecond-level rapid interception. The output generates a preliminary abnormal event containing the abnormal measurement point ID, abnormal value, deviation degree, and timestamp.

[0155] Pattern anomaly detection is used to identify abnormal patterns or trend changes in data over time series, and is applicable to gradual degradation of equipment performance, early signs of complex faults, etc. Those skilled in the art should know how to implement pattern anomaly detection. The following pattern anomaly detection methods are only two preferred embodiments provided by this invention. Those skilled in the art can choose according to actual conditions, or use other methods for pattern anomaly detection.

[0156] One method for detecting pattern anomalies is as follows: Temporal pattern recognition utilizes deep learning models based on LSTM (Long Short-Term Memory) or Transformer to jointly model the multi-dimensional operating parameters (load, temperature, vibration, etc.) of key equipment (such as transformers and power lines) in a temporal sequence. The model learns from historical normal data to predict the power grid's operating state at the next moment. When the duration of a continuous difference between the predicted and actual power grid operating states exceeds an adaptive continuous anomaly duration threshold, it is determined to be a pattern anomaly. This continuous anomaly duration threshold can be set by those skilled in the art, with a preferred value range of 10 seconds.

[0157] Another method for detecting pattern anomalies is as follows: In feature trend analysis, the slope, curvature, periodicity, and other characteristics of the data within the sliding window are calculated, and the matching degree is calculated with a "health feature template library" learned from historical normal data. If the matching degree is lower than a threshold, it indicates that the operating pattern has deviated.

[0158] In the reconstruction error analysis using Principal Component Analysis (PCA), PCA dimensionality reduction is performed on multiple highly correlated measurement point data, and the original data is reconstructed using principal components. Under normal conditions, the reconstruction error is small; however, when the system operates abnormally, the correlation is disrupted, leading to a significant increase in the reconstruction error, which is thus identified as a mode anomaly.

[0159] The output generates a pattern of abnormal events, including the abnormal device / area, the abnormal mode type (such as continuous overheating, increased vibration), confidence level, and duration. Group anomaly detection identifies collective abnormal behavior exhibited by multiple spatially or logically related devices or areas in a power grid. It is applicable to regional faults, cascading overloads caused by power flow shifts, and systemic oscillations. Those skilled in the art should know how to implement group anomaly detection. The following group anomaly detection methods are merely three preferred embodiments provided by this invention; those skilled in the art can construct group anomaly detection methods according to actual conditions.

[0160] The first method for detecting group anomalies is based on topology propagation analysis using graph neural networks (GNNs). GNN topology propagation analysis abstracts the power grid topology into a graph structure, where nodes represent transformer equipment and edges represent electrical connections or control relationships. The GNN learns the propagation pattern of node state parameters (such as power and voltage) on the topology under normal conditions (used to predict the state of each node in the next moment based on historical node states). When the propagation pattern formed by real-time data on the graph differs significantly from the learned normal pattern (by comparing the predicted node states with the actual node states, a node is marked as an anomaly when the predicted state differs from the actual state; a significant difference is considered when the number of anomaly nodes exceeds a threshold), a group anomaly is identified.

[0161] The second method for detecting group anomalies is as follows: Spatial clustering is performed based on the geographical location and electrical distance of the devices. For devices within the same cluster, the consistency of their state changes is examined. Under normal circumstances, the states of devices within the same cluster should have high consistency. If the states of some devices deviate from the "collective behavior" of most devices within the cluster, then a cluster anomaly is determined to have occurred.

[0162] The third method for detecting group anomalies is as follows: An improved Apriori or FP-Growth algorithm is used to uncover strong correlation rules between transformer equipment alarms and transformer equipment status parameters (such as power and voltage). When multiple correlation rules representing systemic risks appear simultaneously in real-time data (e.g., "Line A overload" and "Bus B voltage drop" and "Protection device C alarm"), a group of abnormal alarms are triggered even if a single event does not exceed the threshold.

[0163] Finally, the output generates a group of abnormal events that includes the scope of the abnormal group (device list or region), a description of the abnormal collective behavior (such as collective overload, voltage cascading drop), an estimated impact range, and potential propagation paths.

[0164] This multi-scale detection mechanism, through progressive analysis at three levels—point, pattern, and population—can comprehensively cover various types of anomalies from micro to macro, significantly improving the completeness and accuracy of anomaly detection.

[0165] 5.1.2 A multi-level anomaly detection mechanism is used to detect and evaluate the execution of the current power grid operation strategy, resulting in an event anomaly score; among which, For each anomalous event detected at the three levels mentioned above, a normalized event anomaly score is calculated. The event anomaly score... A s The calculation can be performed using the following formula: ; in, The point anomaly weight is set to max (the normalized value of the comprehensive correlation value of the data point at the current time, which is the minimum preset point anomaly weight). The normalized value of the comprehensive correlation value of the data point at the current time is the sum of the comprehensive correlation value of the data point / the sum of the comprehensive correlation values ​​of all data points. The preferred value of the minimum preset point anomaly weight is 0.2 to 0.35. This represents the point anomaly gain coefficient, and its preferred value is 3 to 4. Indicates the number of point anomalies; This indicates the maximum number of anomalies in historical data points; max() indicates taking the maximum value within the parentheses. The abnormal pattern weight is preferably set between 0.25 and 0.4. Indicates the duration of the current abnormal pattern; Indicates the maximum duration of historical pattern anomalies; The mode anomaly gain coefficient is represented, and its preferred value is 3.5 to 4.5; The group anomaly weight is preferably set between 0.4 and 0.55. The coefficient represents the presence of group anomalies, which is 1 when group anomalies exist and 0 when group anomalies do not exist. The sum of the point anomaly weight, the pattern anomaly weight, and the group anomaly weight is 1 (when the point anomaly weight is large, the pattern anomaly weight and the group anomaly weight can be reduced proportionally).

[0166] The event anomaly score can also be calculated by combining multiple dimensions such as the statistical significance of the anomaly, the degree of pattern deviation, the number / level of affected devices, and the duration. Those skilled in the art should know how to construct this formula, so it will not be elaborated here.

[0167] A multi-layered anomaly detection mechanism is used to monitor and evaluate the execution of the current power grid operation strategy in real time. This mechanism not only evaluates the raw data but also assesses the consistency between the strategy-driven control commands and the actual system response, thereby obtaining a comprehensive event anomaly score.

[0168] The event anomaly scoring results are matched and analyzed with a preset event anomaly scoring threshold to obtain the analysis results; the analysis results include an analysis of the state change trend during the implementation of the power grid operation strategy; the preferred value range of the event anomaly scoring threshold is [0.2, 0.35]. If the event anomaly score is greater than the preset event anomaly score threshold, the current power grid operation strategy is determined to meet the abnormal operation conditions; otherwise, it does not meet the abnormal operation conditions, and the power grid operation strategy continues to operate.

[0169] If the abnormal operation conditions are met, the multi-level abnormal detection results are integrated to generate a final abnormal detection result that includes the abnormality type, severity, and scope of impact, and then proceed to step 6.

[0170] 5.2 Time Series Anomaly Detection Algorithm Combined with Time Series Statistical Characteristics Analysis for Anomaly Fluctuation Detection: When the power grid's operating status deviates from the normal operating mode, it is considered an abnormal power grid operation. Based on the analysis of abnormal power grid operation combined with time series statistical characteristics, a multi-level anomaly detection mechanism is constructed using a dynamic anomaly detection threshold discrimination method. This dynamic anomaly detection threshold discrimination method determines whether an abnormal power grid operation exists based on anomaly criteria. The discrimination criteria, as defined by the dynamic anomaly detection threshold discrimination method, are as follows:

[0171] in, The fluctuation observations are for normal operation mode (the standard deviation calculated based on historical 30-day data is used as the fluctuation value). This represents the current actual fluctuation value (the difference between the data at the current moment and the data at the previous moment). This represents the anomaly detection sensitivity coefficient. Let z be the standard deviation at time z.

[0172] The data used for abnormal fluctuation detection can be the feature data selected in step 5.1, or can be selected by those skilled in the art.

[0173] The presence of anomalies in power grid operation is determined by comparing the deviation between observed fluctuation values ​​under normal operating conditions and the current actual fluctuation values. An anomaly is identified when the absolute value of the deviation exceeds a threshold calculated based on the dynamic standard deviation. The actual operating fluctuation values ​​are obtained from real-time observations. The anomaly detection sensitivity coefficient is dynamically adjusted based on the data type and application scenario, typically ranging from 1.5 to 3.0. The standard deviation at time t is the standard deviation of the real-time fluctuation characteristic values ​​reflected after normalization of all data points in the spatiotemporal standard dataset at time t. This time-series detection method can effectively identify abnormal patterns in time-series data and is particularly suitable for continuous monitoring and anomaly early warning of power grid operating parameters.

[0174] 5.3 Multi-layered detection architecture: Establishing a three-layered progressive detection system. 5.3.1 First layer: Real-time single-point detection; Detection object: Real-time numerical change of a single measuring point; Detection method: Statistical control chart, three-standard-deviation criterion, or sliding window anomaly detection methods; Response time: Achieve millisecond-level real-time detection response; Implementation mechanism: Calculate the standardized deviation between the current value and the mean of historical data within the sliding window. When the deviation exceeds a preset threshold (preferably 2-3 times the standard deviation), it is determined to be a point anomaly.

[0175] This layer of detection can quickly identify sudden anomalies and provide timely early warnings for emergencies.

[0176] 5.3.2 Second Layer: Perform pattern anomaly detection. Detection objects: anomaly patterns and associations of multiple related measurement points (determined by electrical connection or geographical proximity, preferably 3-5 measurement points per group). Detection methods: Principal component analysis, cluster analysis, and association rule mining are employed. Response time: Achieve minute-level detection response. Identification mechanism: For multiple related measurement points, principal component analysis is used to reduce data dimensionality while retaining key information features. Then, the isolated forest algorithm is used to calculate anomaly scores. The input to the isolated forest model is the dimensionality-reduced multidimensional feature vector. The smaller the anomaly score, the greater the probability of an anomaly. A score less than 0.5 is identified as a data pattern deviating from the normal pattern and recorded as a pattern anomaly.

[0177] This layer of detection can uncover complex anomaly patterns and potential systemic problems, providing a deeper level of anomaly analysis.

[0178] 5.3.3 Third Layer: Perform group anomaly detection. Detection object: the operating status and global characteristics of the entire power grid system; Detection method: global analysis based on power grid topology and network anomaly propagation model; Response time: realize an hour-level detection and response system; In addition to the three methods mentioned in step 5.1.1, group anomaly detection can also be achieved through the following group anomaly estimation mechanism: Group anomalies are determined by calculating the overall power grid health score and the anomaly propagation risk assessment value. A group anomaly is determined when the health score is lower than the health threshold (preferably 0.8) or the anomaly propagation risk assessment value (preferably 0.6). The formula for calculating the overall power grid health score is as follows: Overall power grid health score = Σ(equipment health status × weight) / total weight; Wherein, Σ(device health × weight) means that the health of each device is multiplied by its corresponding weight and then summed.

[0179] The risk assessment value for abnormal propagation is shown in the following formula: Abnormal transmission risk assessment value = Σ(transmission probability × impact degree); Wherein, Σ(propagation probability × degree of influence) represents the summation of the products of each propagation probability and degree of influence.

[0180] This layer of detection can assess the power grid's operating status from a macroscopic perspective and identify potential risks that may affect system stability.

[0181] 5.4 Establish a four-level anomaly classification and response mechanism; Establish a graded response mechanism for abnormal operating conditions, classifying abnormal events into four levels—minor, general, severe, and urgent—based on anomaly scores, scope of impact, and duration, and corresponding to different early warning and response strategies. Minor anomalies (Level 1): Only point anomalies exist. The system automatically records the anomaly information without manual intervention. General anomaly (Level 2): ​​Only a pattern anomaly exists. The system issues a warning message, and operators are advised to pay attention. Critical Anomaly (Level 3): When only a group anomaly exists, it is determined that the anomaly affects the stable operation of a local system but does not threaten the overall security. The system issues an alarm signal and needs to be handled in a timely manner. Emergency Anomaly (Level 4): When there is a group anomaly, as well as a point anomaly and / or pattern anomaly, it is determined that the overall system security is threatened. The system responds immediately and activates emergency plans (such as emergency load shedding plan and emergency equipment shutdown plan).

[0182] Based on step 5.1.2, determine whether the current power grid operation strategy meets the abnormal operation conditions; If the abnormal operating conditions are not met, the power grid operation strategy shall continue to operate.

[0183] If the abnormal operation conditions are met, the multi-level anomaly detection results are integrated to generate a final anomaly detection result that includes the anomaly type, severity, and scope of impact. Then, return to step 6.

[0184] This grading mechanism can provide differentiated response strategies based on the severity of the anomaly, ensuring the rational allocation and efficient use of resources.

[0185] The system monitors the operation mode of the power grid operation strategy in real time, and triggers the corresponding level of early warning response strategy when the operation status deviates from the normal operation mode.

[0186] The continuous abnormal fluctuation monitoring and abnormal propagation range assessment correspond to the temporal anomaly detection and multi-level detection architecture content in steps 5.2-5.3 of this embodiment.

[0187] The threshold for triggering abnormal operating conditions is set according to a multi-discrimination mechanism, including a single-point abnormality threshold, a continuous abnormality duration threshold, and a propagation range threshold.

[0188] Step 6: When the anomaly detection result meets the abnormal operating conditions, re-acquire the fusion weight factor set to adjust the spatiotemporal fusion analysis rules until the anomaly detection result no longer meets the abnormal operating conditions. The process of re-acquiring the fusion weight factor set can generally be done according to step 2. When the second weight coefficient optimization method is selected in step 2.3, the fusion weight factor set can also be re-acquired according to the following method.

[0189] 6.1 Anomaly Detection Result Analysis When the system detects an anomaly, it automatically performs in-depth analysis: First, anomaly type identification is performed. Through feature analysis and pattern matching, it is determined whether the anomaly belongs to data anomaly (corresponding to step 5.3.1 point anomaly detection), equipment anomaly (corresponding to step 5.3.2 pattern anomaly detection), or system anomaly (corresponding to step 5.3.3 group anomaly detection). Then, the scope of the anomaly is assessed. Based on the power grid topology and correlation analysis, and combined with the propagation delay sub-rule in the spatiotemporal correlation rules in step 3.5, the scope of the anomaly is taken as the geographical range affected, and the number of devices within the scope of the anomaly is taken as the number of devices affected. Finally, the severity is quantified by comprehensively considering the duration of the anomaly, the degree of deviation, and the scope of impact, establishing a quantitative assessment system. The assessment formula is as follows: Severity = w ×duration + w ×Deviation degree+w × Scope of impact; in, , , These are the weighting coefficients for duration, degree of deviation, and scope of influence, respectively.

[0190] This analysis mechanism can provide accurate feedback information for subsequent weight adjustments and strategy optimization.

[0191] 6.2 Adaptively adjust the fusion weights based on the anomaly analysis results: Assess the type, severity, and scope of impact of current anomaly detection results, and analyze the root causes of the anomalies; Based on the root cause, identify the abnormal characteristics and historical adjustment experience to determine the type and magnitude of the weight factors that need to be readjusted. Retrieve the values ​​of data quality factor, timeliness factor, importance factor, and relevance factor; Based on the updated factor values, re-obtain the fusion weight factor set:

[0192] in, To determine the dynamic weight of a certain dimension of data in the reacquired spatiotemporal data standard set at the current moment, As a new ; This refers to the data quality factor of the data in this dimension at the current moment, which is re-acquired according to step 2.1. This refers to the timeliness factor of the data in this dimension, which is re-acquired at the current moment based on step 2.1. This refers to the importance factor of the data in this dimension at the current moment, which is re-acquired based on step 2.1. This refers to the correlation factor of the data in this dimension at the current moment, which is re-acquired according to step 2.1. The data quality weight coefficients for this dimension of data at the current moment are to be re-acquired. The timeliness weighting coefficient for the current dimension of the data to be reacquired at the current moment. The importance weight coefficients for this dimension of data at the current moment, which are then reacquired. The correlation weight coefficients for the data of this dimension at the current moment are retrieved again. , , , These are all dynamically adjusted weight parameters that have been re-obtained to adapt to the current abnormal situation.

[0193] The newly obtained dynamic weight adjustment parameters are obtained by adding the product of the learning rate and the error gradient to each current dynamic weight adjustment parameter. The weights for each type of data are processed in this way. The learning rate is the learning rate of the neural network model in step 2.3, calculated using the following formula:

[0194] in, This represents any current dynamic weight adjustment parameter (which can be...). , , , ); This indicates that any of the dynamic weight adjustment parameters have been re-obtained; The learning rate (the more severe the anomaly, the higher the learning rate). The error gradient is calculated using the following formula:

[0195] in, This represents the dynamic weight of a certain dimension of data in the i-th spatiotemporal data standard set; the learning rate is dynamically adjusted according to the severity of the anomaly, with a value ranging from 0.01 to 0.1, and the adjustment process needs to meet the constraint that the sum of the weights is 1; this dynamic adjustment mechanism can continuously optimize the weight configuration based on the actual operating effect, thereby improving the accuracy and adaptability of data fusion.

[0196] 6.3 Achieve continuous improvement of weights through iterative optimization: By reconstructing the spatiotemporal correlation analysis model using the updated fusion weight factor set, weighted summation of data from each dimension is performed, taking into account the influence of the spatiotemporal correlation matrix, resulting in a comprehensive data fusion analysis. Iterative optimization continues until the anomaly detection results meet the convergence criterion, which is five consecutive detection cycles without any anomalies or an anomaly score below a preset threshold. This iterative mechanism ensures continuous improvement and stable operation of the system performance, specifically: ; Among them, F This represents the comprehensive correlation value of a spatiotemporal coordinate point with longitude coordinate x, latitude coordinate y, and time coordinate t; d represents the d-th data dimension with longitude coordinate x and latitude coordinate y of the sampling point, d∈[1, ], The total number of data dimensions with longitude coordinates x and dimensional coordinates y for each data sampling point; h represents the h-th data dimension, h∈[1, y]. m ], m This represents the total number of data dimensions. This represents the dynamic weight of the d-th dimension data in the i-th spatiotemporal data standard set at the current moment; This represents the data of dimension d in a three-dimensional spatiotemporal coordinate frame. The spatiotemporal distribution function value at that location; This indicates that the data in dimension d and the data in dimension h are at time d. t The spatiotemporal correlation strength; the data dimensions include data from various dimensions in transformer operation data, data from various dimensions in environmental monitoring data, and load data.

[0197] Step 7: Quantify the anomaly detection results using an adaptive visualization algorithm to obtain the visualization results; 7.1 Adaptive visualization algorithms automatically select the most suitable visualization method based on data characteristics: The system automatically selects the highest-scoring visualization scheme by evaluating the suitability of different visualization types for the current data. The evaluation process comprehensively considers factors such as data type matching, user preference settings, and display performance constraints.

[0198] 7.1.1 The data feature analysis system automatically analyzes the basic characteristics of the data: Dimensional identification: Identifying different dimensional features of three-dimensional spatiotemporal evolution data within a three-dimensional spatiotemporal coordinate framework; Distribution characteristics: Based on the three-dimensional spatiotemporal evolution data, identify whether the data conforms to a normal distribution, skewed distribution, or multimodal distribution. Methods such as the Shapiro-Wilk test and the Kolmogorov-Smirnov test can be used to distinguish them. Spatiotemporal characteristics: Determine whether the data is static (change range less than 5%), dynamic (change range greater than 5%), periodic (using the results of the seasonal association sub-rule in step 3.5 spatiotemporal association rules), or trend-based; 7.1.2 Visualization types include heatmaps, contour maps, 3D scatter plots, time series plots, etc. The scoring function comprehensively considers three dimensions: data suitability (calculated through a data type and visualization type matching matrix), user preference (using the visualization types most frequently used by users in the last 30 days), and performance constraints (rendering time <100ms, memory usage <500MB). It calculates the comprehensive score of each visualization type through a weighted summation method, where the weight coefficients are adjusted according to the application scenario and user needs. The weight of data suitability is 0.4-0.6, the weight of user preference is 0.2-0.4, and the weight of performance constraints is 0.1-0.3. Generally, data suitability has the largest weight, and performance constraints have the smallest weight.

[0199] This adaptive mechanism can automatically select the most suitable visualization method based on different data characteristics and application needs, thereby improving the effectiveness of information display and user experience.

[0200] 7.2 WebGL, a network graphics library, is used for 3D rendering; WebGL technology is used to achieve high-performance 3D visualization.

[0201] 7.2.1 Construct the scene.

[0202] The geographic base map is loaded with geographic information of the power grid coverage area (including topographic maps, administrative division maps, road network maps, etc.). The equipment models include 3D substation and 3D transmission line models, which are scaled according to the ratio of the geographic base map to the actual scale, and the scaled equipment models are placed in the corresponding positions on the geographic base map. Map operational data to colors, sizes, and animations. For example, voltage data is mapped to colors (green for normal, yellow for warning, and red for abnormal), load data is mapped to device size, and fault status is mapped to flashing animations.

[0203] 7.2.2 Rendering optimization using Level of Detail (LOD) technology: The model's level of detail is dynamically adjusted based on the virtual camera's viewing distance. Frust culling: Only renders objects within the virtual camera's field of view (3D substations, power transmission line models, etc.); Use instantiation rendering technology to render objects of the same type in batches; Using shaders for parallel data processing accelerates computation by enabling graphics processing units (GPUs). 7.2.3 Construct interactive functions, including zoom and roam, object selection, timeline control, and multi-view switching.

[0204] The zoom and roam feature uses mouse wheel and drag technology to support zooming, panning, and rotating of the scene. The object selection is performed using X-ray inspection technology; click to obtain detailed equipment information. The time axis control uses a time slider control to replay the spatiotemporal evolution process of historical data; The multi-view switching is achieved by using a view switching button, which supports multiple observation angles such as top view, side view, and perspective.

[0205] Example 2 A method for multi-dimensional spatial data fusion analysis of power grids includes: Step 1: Collect multidimensional spatiotemporal data of the power grid to form a spatiotemporal dataset; Specifically, step 1 involves deploying various sensors (such as current sensors, voltage sensors, temperature sensors, meteorological sensors, GPS positioning sensors, etc.) to report multi-dimensional spatiotemporal data of the power grid in real time via an edge gateway or communication network. This multi-dimensional spatiotemporal data includes transformer operation data and environmental monitoring data. The transformer operation data includes the load of each transformer, the operating status of relay protection devices, and bus voltage. The environmental monitoring data includes temperature, humidity, and wind speed. Specifically, current sensors collect main line current data, voltage sensors collect bus voltage data, temperature sensors collect ambient temperature data, meteorological sensors collect meteorological data such as humidity and wind speed, and GPS positioning sensors collect the geographical coordinates of the equipment. Other required sensors can be configured by those skilled in the art and will not be elaborated here.

[0206] Data is labeled using timestamps and geographic coordinates to generate a three-dimensional or multi-dimensional raw dataset in the form of "space × time × attribute".

[0207] Step 2: After preprocessing and standardizing the spatiotemporal dataset, a standardized spatiotemporal data set with a unified format is obtained; Centralized preprocessing and standardization of spatiotemporal data yields a standard set of spatiotemporal data, which specifically includes data cleaning, format conversion, and data reduction.

[0208] The data cleaning process involves removing outliers, filling in missing values, and eliminating duplicate records.

[0209] The format conversion refers to unifying the time format (such as the Internet time format standard RFC3339), coordinate reference system, and units (such as kilovolts and megawatts).

[0210] The data reduction involves adjusting the spatiotemporal resolution or extracting features when necessary to form a unified structured table or multidimensional array.

[0211] This reduces the requirements for downstream algorithms to be compatible with heterogeneous data, improves processing efficiency and accuracy, ensures data consistency in the analysis chain, and reduces error propagation.

[0212] The spatiotemporal data standard set includes the timestamp of the data acquisition time, the load of each transformer, the bus voltage, the operating status of the relay protection device, the temperature, humidity and wind speed, as well as the sampling location of the load of each transformer, the bus voltage, the operating status of the relay protection device, the temperature, humidity and wind speed.

[0213] Step 3: Construct a dynamic weight fusion mechanism based on the spatiotemporal data standard set, obtain data quality factors, timeliness factors, importance factors, and relevance factors, and obtain a set of fusion weight factors; A dynamic weight fusion mechanism is constructed to obtain a set of weight factors. By designing a multi-dimensional scoring function, data quality factors (completeness, accuracy, and consistency), timeliness factors (data up-to-dateness), importance factors (data's impact on operations), and correlation factors (data's relevance to other indicators) are calculated separately. The fusion weights of each factor are generated automatically or semi-automatically. This dynamically reflects changes in data status and business needs, making the fusion results more robust and targeted; avoiding human design biases; and improving model adaptability.

[0214] Specifically, the dynamic weight fusion mechanism constructed based on the spatiotemporal data standard set obtains data quality factors, timeliness factors, importance factors, and relevance factors to obtain a fusion weight factor set, including: Construct a dynamic weighting model based on data quality factors, timeliness factors, importance factors, and relevance factors; The weighting factors of each dimension of data in the spatiotemporal data standard set at each time point are calculated based on the dynamic weighting model. Specifically, the weighting factors of each dimension of data in each spatiotemporal data standard set are calculated based on the quality factor, timeliness factor, importance factor, relevance factor, data quality weight coefficient, timeliness weight coefficient, importance weight coefficient, and relevance weight coefficient corresponding to each dimension of data in each spatiotemporal data standard set. The data quality weight coefficient, timeliness weight coefficient, importance weight coefficient, and relevance weight coefficient are all dynamic weighting adjustment parameters. The data quality factor is calculated based on the actual number of data points, the number of non-abnormal data points, and the number of data points to be collected for each weight factor data to be calculated within a set first time length. The time factor is calculated based on the time difference between the acquisition time of each weight factor data to be calculated and the current time, as well as the time decay coefficient corresponding to the data type; the data types include transformer operation data and environmental monitoring data. The importance factors are determined based on the data dimensions of each weight factor data to be calculated, and the preset data dimension-importance factor correspondence; the data dimensions include the load of each transformer, the operating status of the relay protection device, the bus voltage, temperature, humidity and wind speed; The correlation factor is calculated based on the correlation coefficient between the time series of each weight factor data to be calculated in the preset first time period and the time series of the data in the preset second time period, as well as the corresponding mutual information value.

[0215] Each of the aforementioned dynamic weight adjustment parameters is calculated based on its time decay coefficient and the time difference between the current time and the data acquisition time.

[0216] A fusion weight factor set is constructed based on the weight factors of data from each dimension.

[0217] Step 4: The historically acquired spatiotemporal data standard set is fused and correlated through the fusion weight factor set to obtain the data fusion analysis results, and a spatiotemporal fusion correlation analysis mechanism is constructed to obtain the spatiotemporal fusion analysis rules; By fusing spatiotemporal data and constructing a correlation analysis mechanism, historical standardized data is weighted and averaged according to spatiotemporal neighborhoods and system topology to uncover spatiotemporal evolution patterns (such as changes in hotspot areas and time-dependent fault modes) and abstract them into analytical rules (thresholds, prediction models, and state transition diagrams). This results in an interpretable spatiotemporal correlation rule base, supporting fault early warning and risk assessment; enabling automatic capture of complex coupling relationships; and improving the safety of power grid operation.

[0218] 4.1 Input the historical spatiotemporal data standard set, spatial topological features, and fusion weight factor set; The historical spatiotemporal data standard set includes multi-dimensional time-series data such as transformer load rate, bus voltage, line current, ambient temperature and humidity, and fault records, forming an input sequence with a time window length of T; The spatial topology features are an adjacency matrix A constructed based on GIS coordinates and electrical connection relationships, representing the spatial proximity and electrical association between devices; The fusion weighting factor is a weight vector composed of data quality factor, timeliness factor, importance factor, and relevance factor.

[0219] 4.2 Construct a three-dimensional spatiotemporal coordinate framework and map the data of each dimension in the spatiotemporal data standard set to the three-dimensional spatiotemporal coordinate framework; the three-dimensional spatiotemporal coordinate framework includes longitude coordinates, latitude coordinates and time coordinates; The step of mapping each dimension of data in the spatiotemporal data standard set to the three-dimensional spatiotemporal coordinate frame involves mapping the value of each dimension of data in the spatiotemporal data standard set to the corresponding coordinate point in the three-dimensional spatiotemporal coordinate frame and performing three-dimensional interpolation to construct the spatiotemporal distribution function of each dimension of data.

[0220] 4.3 Based on the fusion weight factor set, the data of each dimension in the three-dimensional spatiotemporal coordinate framework are fused to obtain the comprehensive correlation value of each spatiotemporal coordinate point as the data fusion analysis result.

[0221] At each spatiotemporal coordinate point, the comprehensive correlation value of each spatiotemporal coordinate point is calculated based on the weights of each dimension of data at that coordinate point, the spatiotemporal distribution function, and the spatiotemporal correlation strength between each dimension of data and other dimension data. The spatiotemporal correlation strength between each dimension of data and other dimension data is calculated based on the geographical distance and electrical distance between the sampling locations of each dimension of data and other dimension data, as well as the sampling time interval between each dimension of data and other dimension data.

[0222] The overall correlation coefficient value is used as the result of data fusion analysis.

[0223] The top 10% of the comprehensive correlation values ​​were selected as the target data for in-depth analysis.

[0224] 4.4 Construct a spatiotemporal association rule generation model, wherein the spatiotemporal association rule generation model is a neural network architecture used to generate spatiotemporal association rules.

[0225] The spatiotemporal association rule generation model adopts a spatiotemporal graph convolutional neural network (ST-GCN) architecture, including the following layers: input layer, spatiotemporal convolutional layer, attention mechanism layer, and output layer: The input layer inputs the time series of data for each dimension at the spatiotemporal coordinate points corresponding to the top 10% comprehensive correlation values ​​and the top 10% comprehensive correlation values ​​within the fifth time length; the fifth time length is preferably 3 hours.

[0226] The spatiotemporal convolutional layer combines a graph convolutional network (GCN) and a long short-term memory network (LSTM). GCN captures spatial dependencies, while LSTM captures time-series features. The attention mechanism layer introduces a spatiotemporal attention mechanism to dynamically adjust the weights of different spatiotemporal locations. Those skilled in the art should know how to acquire and apply the spatiotemporal attention mechanism, which will not be elaborated here. The output layer generates spatiotemporal association rules, including anomaly thresholds, state transition probability matrices, and risk propagation paths.

[0227] 4.3 Training process of spatiotemporal association rule generation model: The spatiotemporal association rule generation model employs a multi-task learning loss function, as shown in the following equation: L=L pred +λ anomaly L anomaly +λ topology L topology ; Among them, L pred To predict the loss, λ anomaly For anomaly detection weights; L anomaly For anomaly detection loss, λ topology To preserve weights for the topology; L topology Preserve loss for topology; The Adaptive Moment Estimator Optimizer (Adam) algorithm was used, with a learning rate of 0.001 and a batch size of 64. Overfitting is prevented by using Dropout (dropout rate, preferably 0.2) and L2 regularization (weight decay, preferably 0.0001).

[0228] By automatically learning complex spatiotemporal correlation patterns through neural networks, an interpretable spatiotemporal correlation rule base is formed, supporting fault early warning and risk assessment; and realizing the automatic capture of complex coupling relationships, thereby improving the safety of power grid operation.

[0229] Step 5: Integrate the power grid operation principles based on spatiotemporal fusion analysis rules to obtain the power grid operation strategy; Based on the rule-integrated power grid operation principle, an operation strategy is generated. By combining spatiotemporal fusion analysis rules with the operation principles of power grid power flow calculation, load forecasting, and equipment limits, a strategy engine (rules + optimization model) is constructed. It outputs scheduling suggestions, such as reconstructing power flow, switching power supply branches, and dynamically adjusting voltage levels, to support intelligent front-line scheduling decisions, reduce human-machine collaboration costs, respond promptly to complex spatiotemporal changes, and improve power supply quality and stability.

[0230] Step 6: Based on the power grid operation strategy, construct a multi-level anomaly detection mechanism to generate abnormal operating conditions; construct the current power grid operation strategy by using the spatiotemporal fusion analysis rules obtained from the current spatiotemporal data standard set, and then detect the anomaly through the multi-level anomaly detection mechanism to obtain the anomaly detection results; By employing multi-layered anomaly detection and operational strategy verification, a multi-layered detection architecture is constructed, encompassing metadata and business metrics: a data layer (format anomalies / data packet loss anomalies), a metric layer (threshold exceedance anomalies), and a business layer (power flow topology anomalies). The current spatiotemporal standard set and the strategies generated by the fusion rules are simultaneously input into the detection module to determine in real time whether an anomaly alarm is triggered. Through end-to-end monitoring, the root cause of anomalies is quickly located, and the spread of faults is prevented; thus improving the timeliness and accuracy of power grid operation safety early warnings. Those skilled in the art can configure the data layer, metric layer, and business layer according to actual conditions; these details will not be elaborated here.

[0231] In another preferred embodiment, steps 5 and 6 specifically further include: Based on the correlation index and historical state data, state prediction is performed, and the power grid operating state at the next moment is calculated using the system state transition model; Combine the state prediction results to obtain the observation data at the current moment, including real-time data such as sensor measurements and equipment status information; By comparing the state prediction results with the observed data, the deviation between the predicted and observed values ​​is calculated to assess the accuracy of the prediction and identify abnormal situations. Based on the deviation assessment results, system noise characteristics, and observed noise levels, the Kalman gain coefficient is dynamically adjusted to optimize the state estimation accuracy. Based on the optimized gain coefficient, the predicted state and the observed data are optimally fused to obtain the corrected state estimate, thereby achieving accurate tracking of the power grid's operating state. Using the precise tracking results, the temporal relationships of data in each dimension are analyzed to identify the causal chains and propagation paths of data changes; By combining the results of causal chain analysis, the spatial correlation between data from different geographical locations is calculated, and the spatial propagation pattern of power grid operation status is discovered. Based on temporal causality and spatial correlation analysis, the degree of interdependence among multidimensional data is assessed, and a data dependency map is constructed. By integrating the aforementioned causal relationships, spatial correlations, and data dependencies, a spatiotemporal correlation network structure covering the entire power grid domain is formed; Based on the aforementioned spatiotemporal correlation network structure, the periodic patterns in historical operation data are analyzed to identify operation modes at different time scales such as daily, weekly, monthly, and seasonal. By combining periodic pattern analysis, spatial distribution patterns are extracted to discover the correlation characteristics and propagation patterns of power grid operation status in different regions; Using the analysis results of the aforementioned operating mode and propagation mode, the spatiotemporal characteristics of abnormal operating modes are identified, and a spatiotemporal propagation model and impact range assessment mechanism for abnormal events are established. Based on the comparative analysis of normal operation mode and abnormal operation mode, a discrimination feature set for normal operation mode and abnormal mode is constructed to provide a benchmark reference for subsequent anomaly detection.

[0232] Step 7: Determine whether the anomaly detection results meet the abnormal operating conditions. If the anomaly detection results meet the abnormal operating conditions, then re-acquire the fusion weight factor set to adjust the spatiotemporal fusion analysis rules until the anomaly detection results no longer meet the abnormal operating conditions. Through adaptive feedback learning and weight / rule iteration, if the anomaly detection result triggers anomaly conditions, step 3 weight calculation is re-triggered, and then the rules and strategies in steps 4 and 5 are adjusted, using reinforcement learning or online gradient descent methods.

[0233] Step 8: Quantify the anomaly detection results using an adaptive visualization algorithm to perform feedback learning and obtain the feedback learning results; By continuously optimizing weights and rules based on anomaly detection results and actual operational feedback, the strategy achieves closed-loop adaptiveness, enabling the model to continuously improve as business scenarios evolve; thereby increasing the anomaly recognition rate and reducing false positives and false negatives.

[0234] Step 9: Implement real-time 3D rendering using WebGL technology to present the feedback learning results in a visual way.

[0235] WebGL real-time 3D rendering visualization transforms data such as power grid topology, power flow simulation, anomaly heatmaps, and strategy execution results into interactive 3D scenes. It supports zooming, roaming, timeline dragging, alarm highlighting, and solution comparison. It intuitively presents complex spatiotemporal situations, helping maintenance personnel quickly understand the current operational status and risks. Visual feedback drives continuous optimization of models and strategies, significantly improving the efficiency of maintenance decision-making.

[0236] Example 3 A method for multi-dimensional spatial data fusion analysis of power grids is proposed to achieve rapid fault location in power grids based on multi-dimensional spatial data fusion.

[0237] The application scenario involves a large-scale power outage in a regional power grid, affecting multiple voltage levels and spanning different geographical areas. Traditional fault location mainly relies on step-by-step investigation, which is time-consuming and makes it difficult to quickly determine the scope of impact.

[0238] This embodiment employs a multi-dimensional spatial data fusion analysis method for power grids according to the present invention to achieve rapid fault location in power grids based on multi-dimensional spatial data fusion, including: Step 1: Collect multi-dimensional spatiotemporal data of the power grid to form a spatiotemporal data standard set; The multidimensional spatiotemporal data of the power grid includes transformer operation data and environmental monitoring data; The transformer operation data includes the load of each transformer, the status of each standby transformer (started, not started), the operating status of relay protection devices (operated, not operated), the status of power distribution equipment (switch position, protection device status), the status of communication equipment, the status of auxiliary equipment (cooling system status (started, not started), lighting system status (on, off)) and bus voltage.

[0239] The environmental monitoring data includes temperature, humidity, and wind speed.

[0240] The spatiotemporal data standard set includes the timestamp of the data acquisition time, the load of each transformer, the status of each standby transformer, the switch position, the status of protection devices, the status of communication equipment, the status of the cooling system (started, not started), the status of the lighting system (on, off), the bus voltage, the operating status of the relay protection device, temperature, humidity, and wind speed, as well as the sampling location corresponding to each data dimension of the transformer load, bus voltage, relay protection device operating status, temperature, humidity, and wind speed. The data dimensions include the load of each transformer (different transformer loads represent different data dimensions), the status of each standby transformer (different standby transformer loads represent different data dimensions), the switch position (different switch positions represent different data dimensions), the status of protection devices (different switch positions represent different data dimensions), the status of communication equipment (different communication device statuses represent different data dimensions), the status of the cooling system (started, not started) (different cooling system statuses represent different data dimensions), the status of the lighting system (on, off) (different lighting system statuses represent different data dimensions), the operating status of the relay protection device (different relay protection device operating statuses represent different data dimensions), the bus voltage, temperature, humidity, and wind speed. Step 2: Construct a dynamic weighted fusion mechanism based on the spatiotemporal data standard set, obtain data quality factors, timeliness factors, importance factors, and relevance factors, and obtain a set of fusion weight factors, including: A dynamic weighting model based on data quality factors, timeliness factors, importance factors, and relevance factors is constructed. The weight fusion mechanism is presented in the form of a formula, which is to obtain the weight of each dimension of data in each spatiotemporal data standard set by weighting the quality factors, timeliness factors, importance factors, and relevance factors of each dimension of data in each spatiotemporal data standard set.

[0241] The weighting factors of each dimension of data in the spatiotemporal data standard set at each time point are calculated based on the dynamic weighting model. Specifically, the weighting factors of each data point in the spatiotemporal data standard set are calculated based on the quality factor, timeliness factor, importance factor, relevance factor, data quality weight coefficient, timeliness weight coefficient, importance weight coefficient, and relevance weight coefficient corresponding to each data point in the spatiotemporal data standard set. The data quality weight coefficient, timeliness weight coefficient, importance weight coefficient, and relevance weight coefficient are all dynamic weighting adjustment parameters. The weight factor for each data point in each spatiotemporal data standard set is calculated using the following formula: ; in, represents the dynamic weight of the d-th dimension data in the i-th spatiotemporal data standard set at the current time; i represents the i-th spatiotemporal data standard set, i∈[1,Ns]; Ns represents the total number of spatiotemporal data standard sets up to the current time; t represents the sampling time of the i-th spatiotemporal data standard set; Indicates the current time. d Data quality weighting coefficients for dimensional data; This represents the i-th spatiotemporal data standard set at the current time. d Data quality factor for dimensional data; Indicates the current time. d Timeliness weighting coefficient for dimensional data; This represents the i-th spatiotemporal data standard set at the current time. d The timeliness factor of dimensional data; Indicates the current time. d Importance weighting coefficients for dimensional data; This represents the i-th spatiotemporal data standard set at the current time. d Importance factors of dimensional data; Indicates the current time. d The correlation weight coefficient of dimensional data; This represents the i-th spatiotemporal data standard set at the current time. d The correlation factor of dimensional data; Each of the aforementioned dynamic weight adjustment parameters is calculated based on its initial setting value, time decay coefficient, and the time difference between the current moment and the data acquisition moment of the data corresponding to the dynamic weight adjustment parameter. The preferred initial setting value of the timeliness weight coefficient is increased to 0.4, the preferred initial setting value of the importance weight coefficient is increased to 0.4, the preferred initial setting value of the data quality weight coefficient is decreased to 0.1, and the preferred initial setting value of the correlation factor weight coefficient is decreased to 0.1. This weight configuration replaces the dynamic weight under normal operating conditions and is specifically used for data fusion analysis under fault conditions.

[0242] A fusion weight factor set is constructed based on the weight factors of data from each dimension.

[0243] Step 3: Construct a three-dimensional spatiotemporal coordinate framework, mapping the data of each dimension in the spatiotemporal data standard set to the three-dimensional spatiotemporal coordinate framework; the three-dimensional spatiotemporal coordinate framework includes longitude coordinates, latitude coordinates, and time coordinates; The step of mapping each dimension of data in the spatiotemporal data standard set to the three-dimensional spatiotemporal coordinate frame involves mapping the value of each dimension of data in the spatiotemporal data standard set to the corresponding coordinate point in the three-dimensional spatiotemporal coordinate frame and performing three-dimensional interpolation to construct the spatiotemporal distribution function of each dimension of data.

[0244] Step 4: Based on the fusion weight factor set, the data of each dimension in the three-dimensional spatiotemporal coordinate framework are fused to obtain the comprehensive correlation value of each spatiotemporal coordinate point. That is, at each spatiotemporal coordinate point, the comprehensive correlation value of each spatiotemporal coordinate point is calculated based on the weights of each dimension data at that coordinate point, the spatiotemporal distribution function, and the spatiotemporal correlation strength between each dimension data and other dimension data. The spatiotemporal correlation strength between each dimension data and other dimension data is calculated based on the geographical distance and electrical distance between the sampling locations of each dimension data and other dimension data, as well as the sampling time interval between each dimension data and other dimension data.

[0245] The comprehensive correlation value is used as the result of data fusion analysis.

[0246] Step 5: Based on the data fusion analysis results, automatically generate spatiotemporal fusion analysis rules; the spatiotemporal fusion analysis rules include spatiotemporal statistical analysis rules, spatiotemporal correlation analysis rules, and evolutionary analysis rules; Spatiotemporal statistical analysis rules use spatiotemporal mean distribution sub-rules to identify the spatial distribution characteristics of power grid operating parameters, use spatiotemporal variance characteristic sub-rules to quantify the operational stability of each region, and use spatial autocorrelation sub-rules to analyze the cooperative operational characteristics of adjacent regions. These analysis results are used to formulate regional operation strategies and load allocation schemes. The spatiotemporal correlation analysis rules establish a fault propagation path prediction model by using the propagation delay sub-rule, optimize the long-term operation plan by using the seasonal correlation sub-rule, and construct the inter-regional load transmission mechanism by using the load correlation sub-rule. These analysis results are used to formulate preventive maintenance plans and emergency response strategies. The evolutionary analysis rules establish a forward-looking prediction model of the power grid operation status based on the trend prediction sub-rules, use the anomaly diffusion sub-rules to assess the impact range of potential risks, and optimize the operation and scheduling plan through the periodic sub-rules. These analysis results are used to formulate medium- and long-term development plans and real-time scheduling decisions. Step 6: Perform spatiotemporal correlation analysis based on spatiotemporal fusion analysis rules to provide a basis for rapid fault location in the power grid; Spatial clustering analysis was performed on historical failure points.

[0247] Spatiotemporal correlation analysis is performed, including fault propagation path analysis and spatiotemporal correlation mining.

[0248] The fault propagation path analysis is based on a fault propagation model of power grid topology and geographic space, and uses the abnormal propagation sub-rule in the spatiotemporal correlation rules to predict the fault propagation path. The spatiotemporal correlation mining involves analyzing the correlation between the time series and spatial location of fault occurrences. It employs spatiotemporal mean distribution sub-rules, spatiotemporal variance feature sub-rules, and spatial autocorrelation sub-rules from the spatiotemporal statistical analysis rules to perform spatiotemporal correlation analysis of faults. Applying the method of Embodiment 3 of this invention improves the fault location accuracy from 500 meters to within 50 meters in the traditional method, reduces the fault location time from an average of 45 minutes to 8 minutes, accurately predicts the spatial range of the fault's impact with an accuracy of 95%, allows the repair team to reach the fault point directly, reduces the repair time by 65%, and reduces the power outage time: the average power outage time is reduced from 2.5 hours to 0.8 hours.

[0249] Example 4 A multi-dimensional spatial data fusion analysis method for power grids is proposed to realize spatial load density analysis and equipment layout optimization of distribution networks.

[0250] The application scenario involves an uneven load distribution in a rapidly developing urban power distribution network, with some areas experiencing severe equipment overload while others have insufficient equipment utilization. Optimization of equipment layout based on spatial data analysis is required.

[0251] This embodiment employs a multi-dimensional spatial data fusion analysis method for power grids according to the present invention to achieve spatial load density analysis and equipment layout optimization of distribution networks, including: Step 1: Collect multi-dimensional spatiotemporal data of the power grid to form a spatiotemporal data standard set; The multidimensional spatiotemporal data of the power grid includes transformer operation data and environmental monitoring data; The transformer operation data includes the load of each transformer, the status of each standby transformer (started, not started), the operating status of relay protection devices (operated, not operated), the status of power distribution equipment (switch position, protection device status), the status of communication equipment, the status of auxiliary equipment (cooling system status (started, not started), lighting system status (on, off)) and bus voltage.

[0252] The environmental monitoring data includes temperature, humidity, and wind speed.

[0253] The spatiotemporal data standard set includes the timestamp of the data acquisition time, the load of each transformer, the status of each standby transformer, the switch position, the status of protection devices, the status of communication equipment, the status of the cooling system (started, not started), the status of the lighting system (on, off), the bus voltage, the operating status of the relay protection device, temperature, humidity, and wind speed, as well as the sampling location corresponding to each data dimension of the transformer load, bus voltage, relay protection device operating status, temperature, humidity, and wind speed. The data dimensions include the load of each transformer (different transformer loads represent different data dimensions), the status of each standby transformer (different standby transformer loads represent different data dimensions), the switch position (different switch positions represent different data dimensions), the status of protection devices (different switch positions represent different data dimensions), the status of communication equipment (different communication device statuses represent different data dimensions), the status of the cooling system (started, not started) (different cooling system statuses represent different data dimensions), the status of the lighting system (on, off) (different lighting system statuses represent different data dimensions), the operating status of the relay protection device (different relay protection device operating statuses represent different data dimensions), the bus voltage, temperature, humidity, and wind speed. Step 2: Construct a dynamic weighted fusion mechanism based on the spatiotemporal data standard set, obtain data quality factors, timeliness factors, importance factors, and relevance factors, and obtain a set of fusion weight factors, including: A dynamic weighting model based on data quality factors, timeliness factors, importance factors, and relevance factors is constructed. The weight fusion mechanism is presented in the form of a formula, which is to obtain the weight of each dimension of data in each spatiotemporal data standard set by weighting the quality factors, timeliness factors, importance factors, and relevance factors of each dimension of data in each spatiotemporal data standard set.

[0254] The weighting factors of each dimension of data in the spatiotemporal data standard set at each time point are calculated based on the dynamic weighting model. Specifically, the weighting factors of each data point in the spatiotemporal data standard set are calculated based on the quality factor, timeliness factor, importance factor, relevance factor, data quality weight coefficient, timeliness weight coefficient, importance weight coefficient, and relevance weight coefficient corresponding to each data point in the spatiotemporal data standard set. The data quality weight coefficient, timeliness weight coefficient, importance weight coefficient, and relevance weight coefficient are all dynamic weighting adjustment parameters. The weight factor for each data point in each spatiotemporal data standard set is calculated using the following formula: ; in, represents the dynamic weight of the d-th dimension data in the i-th spatiotemporal data standard set at the current time; i represents the i-th spatiotemporal data standard set, i∈[1,Ns]; Ns represents the total number of spatiotemporal data standard sets up to the current time; t represents the sampling time of the i-th spatiotemporal data standard set; Indicates the current time. d Data quality weighting coefficients for dimensional data; This represents the i-th spatiotemporal data standard set at the current time. d Data quality factor for dimensional data; Indicates the current time. d Timeliness weighting coefficient for dimensional data; This represents the i-th spatiotemporal data standard set at the current time. d The timeliness factor of dimensional data; Indicates the current time. d Importance weighting coefficients for dimensional data; This represents the i-th spatiotemporal data standard set at the current time. d Importance factors of dimensional data; Indicates the current time. d The correlation weight coefficient of dimensional data; This represents the i-th spatiotemporal data standard set at the current time. d The correlation factor of dimensional data; Each of the aforementioned dynamic weight adjustment parameters is calculated based on its initial set value, time decay coefficient, and the time difference between the current moment and the data acquisition moment of the data corresponding to the dynamic weight adjustment parameter. A fusion weight factor set is constructed based on the weight factors of data from each dimension.

[0255] Step 3: Construct a three-dimensional spatiotemporal coordinate framework, mapping the data of each dimension in the spatiotemporal data standard set to the three-dimensional spatiotemporal coordinate framework; the three-dimensional spatiotemporal coordinate framework includes longitude coordinates, latitude coordinates, and time coordinates; The step of mapping each dimension of data in the spatiotemporal data standard set to the three-dimensional spatiotemporal coordinate frame involves mapping the value of each dimension of data in the spatiotemporal data standard set to the corresponding coordinate point in the three-dimensional spatiotemporal coordinate frame and performing three-dimensional interpolation to construct the spatiotemporal distribution function of each dimension of data.

[0256] Step 4: Based on the fusion weight factor set, the data of each dimension in the three-dimensional spatiotemporal coordinate framework are fused to obtain the comprehensive correlation value of each spatiotemporal coordinate point. That is, at each spatiotemporal coordinate point, the comprehensive correlation value of each spatiotemporal coordinate point is calculated based on the weights of each dimension data at that coordinate point, the spatiotemporal distribution function, and the spatiotemporal correlation strength between each dimension data and other dimension data. The spatiotemporal correlation strength between each dimension data and other dimension data is calculated based on the geographical distance and electrical distance between the sampling locations of each dimension data and other dimension data, as well as the sampling time interval between each dimension data and other dimension data.

[0257] The comprehensive correlation value is used as the result of data fusion analysis.

[0258] Step 5: Based on the data fusion analysis results, automatically generate spatiotemporal fusion analysis rules; the spatiotemporal fusion analysis rules include spatiotemporal statistical analysis rules, spatiotemporal correlation analysis rules, and evolutionary analysis rules; Spatiotemporal statistical analysis rules use spatiotemporal mean distribution sub-rules to identify the spatial distribution characteristics of power grid operating parameters, use spatiotemporal variance characteristic sub-rules to quantify the operational stability of each region, and use spatial autocorrelation sub-rules to analyze the cooperative operational characteristics of adjacent regions. These analysis results are used to formulate regional operation strategies and load allocation schemes. The spatiotemporal correlation analysis rules establish a fault propagation path prediction model by using the propagation delay sub-rule, optimize the long-term operation plan by using the seasonal correlation sub-rule, and construct the inter-regional load transmission mechanism by using the load correlation sub-rule. These analysis results are used to formulate preventive maintenance plans and emergency response strategies. The evolutionary analysis rules establish a forward-looking prediction model of the power grid operation status based on the trend prediction sub-rules, use the anomaly diffusion sub-rules to assess the impact range of potential risks, and optimize the operation and scheduling plan through the periodic sub-rules. These analysis results are used to formulate medium- and long-term development plans and real-time scheduling decisions. Step 6: Perform spatial load density modeling and multi-dimensional spatiotemporal correlation analysis based on spatiotemporal fusion analysis rules to provide a basis for distribution network spatial load density analysis and equipment layout optimization; Modeling spatial load density includes: Collect user distribution data, including the precise spatial coordinates of residential, commercial, and industrial users; Collect load characteristic data, including load density and time-varying characteristics of different types of users; Collect urban planning data, including spatial layout information for future development plans and geographic constraint data such as topography, roads, and buildings.

[0259] Multidimensional spatial correlation analysis is conducted, including load spatial clustering analysis, equipment coverage analysis, and spatial growth prediction.

[0260] The load space clustering analysis is based on spatiotemporal statistical analysis rules to identify high load density regions and low load density regions. The device coverage analysis is based on spatiotemporal correlation analysis rules, as well as geographical distance and electrical distance to identify the service range of the device; The spatial growth forecast is based on evolutionary analysis rules combined with urban development planning to predict the spatial growth trend of load.

[0261] Applying the method of Embodiment 4 of the present invention reduces the variance of equipment load rate between regions from 0.25 to 0.08, resulting in a more balanced load distribution. Through spatial optimization, the efficiency of equipment investment is improved by 40%, the average power supply radius is optimized from 1.2 km to 0.8 km, the distribution network line loss rate is reduced from 4.2% to 2.8%, and the equipment footprint is reduced by 30%, significantly improving space utilization efficiency.

[0262] Example 5 A multi-dimensional spatial data fusion analysis method for power grids is proposed to achieve spatial early warning of power grid security risks based on spatiotemporal data fusion.

[0263] The application scenario is that power grid security risks have obvious spatial clustering characteristics and spatiotemporal evolution patterns, and traditional risk assessment methods are difficult to effectively identify the spatial propagation path and impact range of risks.

[0264] This embodiment employs a multi-dimensional spatial data fusion analysis method for power grids according to the present invention to achieve spatial early warning of power grid security risks based on spatiotemporal data fusion, including: Step 1: Collect multi-dimensional spatiotemporal data of the power grid to form a spatiotemporal data standard set; The multidimensional spatiotemporal data of the power grid includes transformer operation data and environmental monitoring data; The transformer operation data includes the load of each transformer, the status of each standby transformer (started, not started), the operating status of relay protection devices (operated, not operated), the status of power distribution equipment (switch position, protection device status), the status of communication equipment, the status of auxiliary equipment (cooling system status (started, not started), lighting system status (on, off)) and bus voltage.

[0265] The environmental monitoring data includes temperature, humidity, and wind speed.

[0266] The spatiotemporal data standard set includes the timestamp of the data acquisition time, the load of each transformer, the status of each standby transformer, the switch position, the status of protection devices, the status of communication equipment, the status of the cooling system (started, not started), the status of the lighting system (on, off), the bus voltage, the operating status of the relay protection device, temperature, humidity, and wind speed, as well as the sampling location corresponding to each data dimension of the transformer load, bus voltage, relay protection device operating status, temperature, humidity, and wind speed. The data dimensions include the load of each transformer (different transformer loads represent different data dimensions), the status of each standby transformer (different standby transformer loads represent different data dimensions), the switch position (different switch positions represent different data dimensions), the status of protection devices (different switch positions represent different data dimensions), the status of communication equipment (different communication device statuses represent different data dimensions), the status of the cooling system (started, not started) (different cooling system statuses represent different data dimensions), the status of the lighting system (on, off) (different lighting system statuses represent different data dimensions), the operating status of the relay protection device (different relay protection device operating statuses represent different data dimensions), the bus voltage, temperature, humidity, and wind speed. Step 2: Construct a dynamic weighted fusion mechanism based on the spatiotemporal data standard set, obtain data quality factors, timeliness factors, importance factors, and relevance factors, and obtain a set of fusion weight factors, including: A dynamic weighting model based on data quality factors, timeliness factors, importance factors, and relevance factors is constructed. The weight fusion mechanism is presented in the form of a formula, which is to obtain the weight of each dimension of data in each spatiotemporal data standard set by weighting the quality factors, timeliness factors, importance factors, and relevance factors of each dimension of data in each spatiotemporal data standard set.

[0267] The weighting factors of each dimension of data in the spatiotemporal data standard set at each time point are calculated based on the dynamic weighting model. Specifically, the weighting factors of each data point in the spatiotemporal data standard set are calculated based on the quality factor, timeliness factor, importance factor, relevance factor, data quality weight coefficient, timeliness weight coefficient, importance weight coefficient, and relevance weight coefficient corresponding to each data point in the spatiotemporal data standard set. The data quality weight coefficient, timeliness weight coefficient, importance weight coefficient, and relevance weight coefficient are all dynamic weighting adjustment parameters. The weight factor for each data point in each spatiotemporal data standard set is calculated using the following formula: ; in, represents the dynamic weight of the d-th dimension data in the i-th spatiotemporal data standard set at the current time; i represents the i-th spatiotemporal data standard set, i∈[1,Ns]; Ns represents the total number of spatiotemporal data standard sets up to the current time; t represents the sampling time of the i-th spatiotemporal data standard set; Indicates the current time. d Data quality weighting coefficients for dimensional data; This represents the i-th spatiotemporal data standard set at the current time. d Data quality factor for dimensional data; Indicates the current time. d Timeliness weighting coefficient for dimensional data; This represents the i-th spatiotemporal data standard set at the current time. d The timeliness factor of dimensional data; Indicates the current time. d Importance weighting coefficients for dimensional data; This represents the i-th spatiotemporal data standard set at the current time. d Importance factors of dimensional data; Indicates the current time. d The correlation weight coefficient of dimensional data; This represents the i-th spatiotemporal data standard set at the current time. d The correlation factor of dimensional data; Each of the aforementioned dynamic weight adjustment parameters is calculated based on its initial set value, time decay coefficient, and the time difference between the current moment and the data acquisition moment of the data corresponding to the dynamic weight adjustment parameter. A fusion weight factor set is constructed based on the weight factors of data from each dimension.

[0268] Step 3: Construct a three-dimensional spatiotemporal coordinate framework, mapping the data of each dimension in the spatiotemporal data standard set to the three-dimensional spatiotemporal coordinate framework; the three-dimensional spatiotemporal coordinate framework includes longitude coordinates, latitude coordinates, and time coordinates; The step of mapping each dimension of data in the spatiotemporal data standard set to the three-dimensional spatiotemporal coordinate frame involves mapping the value of each dimension of data in the spatiotemporal data standard set to the corresponding coordinate point in the three-dimensional spatiotemporal coordinate frame and performing three-dimensional interpolation to construct the spatiotemporal distribution function of each dimension of data.

[0269] Step 4: Based on the fusion weight factor set, the data of each dimension in the three-dimensional spatiotemporal coordinate framework are fused to obtain the comprehensive correlation value of each spatiotemporal coordinate point. That is, at each spatiotemporal coordinate point, the comprehensive correlation value of each spatiotemporal coordinate point is calculated based on the weights of each dimension data at that coordinate point, the spatiotemporal distribution function, and the spatiotemporal correlation strength between each dimension data and other dimension data. The spatiotemporal correlation strength between each dimension data and other dimension data is calculated based on the geographical distance and electrical distance between the sampling locations of each dimension data and other dimension data, as well as the sampling time interval between each dimension data and other dimension data.

[0270] The comprehensive correlation value is used as the result of data fusion analysis.

[0271] Step 5: Based on the data fusion analysis results, automatically generate spatiotemporal fusion analysis rules; the spatiotemporal fusion analysis rules include spatiotemporal statistical analysis rules, spatiotemporal correlation analysis rules, and evolutionary analysis rules; Spatiotemporal statistical analysis rules use spatiotemporal mean distribution sub-rules to identify the spatial distribution characteristics of power grid operating parameters, use spatiotemporal variance characteristic sub-rules to quantify the operational stability of each region, and use spatial autocorrelation sub-rules to analyze the cooperative operational characteristics of adjacent regions. These analysis results are used to formulate regional operation strategies and load allocation schemes. The spatiotemporal correlation analysis rules establish a fault propagation path prediction model by using the propagation delay sub-rule, optimize the long-term operation plan by using the seasonal correlation sub-rule, and construct the inter-regional load transmission mechanism by using the load correlation sub-rule. These analysis results are used to formulate preventive maintenance plans and emergency response strategies. The evolutionary analysis rules establish a forward-looking prediction model of the power grid operation status based on the trend prediction sub-rules, use the anomaly diffusion sub-rules to assess the impact range of potential risks, and optimize the operation and scheduling plan through the periodic sub-rules. These analysis results are used to formulate medium- and long-term development plans and real-time scheduling decisions. Step 6: Apply multi-dimensional spatial data fusion based on spatiotemporal fusion analysis rules to provide a basis for spatial early warning of power grid security risks; Multidimensional spatial data fusion applications include risk spatial element fusion, environmental risk factor acquisition, load importance analysis, network topology analysis, risk spatial diffusion analysis, temporal evolution analysis, and multi-scale spatial analysis. Risk spatial element fusion involves fusing the spatial distribution of equipment health status: equipment aging degree, and fault history. Environmental risk factors include the spatial distribution of lightning density, pollution level, and geological hazards. Load importance analysis analyzes the spatial distribution and importance level of important users and critical facilities. Network topology analysis analyzes the spatial representation and vulnerability of power grid connections based on spatiotemporal correlation analysis rules.

[0272] The risk spatial diffusion analysis is a risk propagation model based on spatiotemporal statistical analysis rules, geographical proximity, and electrical connectivity. The aforementioned time evolution analysis is based on spatiotemporal correlation analysis rules to analyze the dynamic evolution of risk over time in space; The multi-scale spatial analysis includes multi-scale risk assessment from the equipment level, station level to the regional level. Applying the method of Embodiment 5 of this invention improves the accuracy of spatial risk identification from 78% to 94%, advances the risk warning time from post-event analysis to 24 hours in advance, accurately predicts the spatial range of risk impact with a coverage rate of 92%, formulates precise prevention and control measures based on spatial risk distribution, improves resource allocation efficiency by 50%, reduces the incidence of major safety accidents by 60%, significantly improves the level of power grid safety operation, and avoids economic losses of approximately RMB 150 million annually due to risk accidents.

[0273] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

[0274] This application also discloses a power grid multi-dimensional spatial data fusion analysis system based on the aforementioned power grid multi-dimensional spatial data fusion analysis method, including a data acquisition and standard set construction module, a dynamic weight fusion module, a data mapping module, and a fusion analysis result generation module: The data acquisition and standard set construction module is used to collect multi-dimensional spatiotemporal data of the power grid and form a spatiotemporal data standard set. The dynamic weight fusion module is used to construct a dynamic weight fusion mechanism based on the spatiotemporal data standard set, obtain data quality factors, timeliness factors, importance factors and relevance factors, and obtain a fusion weight factor set; The data mapping module is used to construct a three-dimensional spatiotemporal coordinate frame and map data of each dimension in the spatiotemporal data standard set to the three-dimensional spatiotemporal coordinate frame. The fusion analysis result generation module fuses data from each dimension of the three-dimensional spatiotemporal coordinate framework based on the fusion weight factor set, and obtains the comprehensive correlation value of each spatiotemporal coordinate point as the data fusion analysis result; the data fusion analysis result is used to formulate power grid operation strategies, judge power grid operation anomalies, etc.

[0275] Example 6 like Figure 3 As shown, a multi-dimensional spatial data fusion and analysis system for power grids includes: Data acquisition module: used to collect multi-dimensional spatiotemporal data of the power grid and form a spatiotemporal dataset; Data preprocessing module: used to preprocess and standardize spatiotemporal datasets to obtain a standardized set of spatiotemporal data in a unified format; Dynamic weight fusion module: used to construct a dynamic weight fusion mechanism based on the spatiotemporal data standard set, obtain data quality factors, timeliness factors, importance factors and relevance factors, and obtain a set of fusion weight factors; Operation strategy generation module: used to establish power grid operation strategies and abnormal power grid operation conditions based on source spatiotemporal data standard sets based on spatiotemporal datasets; Anomaly detection module: used to acquire the target spatiotemporal data standard set generated during the current operation in real time, perform multi-level anomaly detection according to the spatiotemporal fusion analysis rules of the target spatiotemporal data standard set, and obtain anomaly detection results; Judgment module: used to match and analyze the results of the anomaly detection with the abnormal operating conditions of the power grid. When the anomaly detection results meet the abnormal operating conditions, the fusion weight factor set is re-acquired to adjust the spatiotemporal fusion analysis rules until the anomaly detection results do not meet the abnormal operating conditions. The power grid operation strategy is obtained by integrating the power grid operation principle based on the spatiotemporal fusion analysis rules. The abnormal operating conditions of the power grid are generated through the multi-level anomaly detection mechanism constructed by the power grid operation strategy. Feedback module: Used to quantify anomaly detection results through adaptive visualization algorithms for feedback learning, and obtain feedback learning results; Visualization and Rendering Module: Used to achieve real-time 3D rendering through WebGL technology, presenting the feedback learning results in a visual way.

[0276] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0277] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0278] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0279] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0280] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for multi-dimensional spatial data fusion and analysis of power grids, characterized in that, include: Collect multidimensional spatiotemporal data of the power grid to form a spatiotemporal data standard set; A dynamic weight fusion mechanism is constructed based on the spatiotemporal data standard set to obtain data quality factors, timeliness factors, importance factors, and relevance factors, thereby obtaining a set of fusion weight factors. A three-dimensional spatiotemporal coordinate framework is constructed, and data of each dimension in the spatiotemporal data standard set are mapped to the three-dimensional spatiotemporal coordinate framework; The data of each dimension in the three-dimensional spatiotemporal coordinate framework are fused based on the fusion weight factor set, and the comprehensive correlation value of each spatiotemporal coordinate point is obtained as the data fusion analysis result; the data fusion analysis result is used to formulate power grid operation strategy and judge power grid operation anomalies.

2. The method for multi-dimensional spatial data fusion analysis of power grids according to claim 1, characterized in that: The multidimensional spatiotemporal data of the power grid includes transformer operation data and environmental monitoring data; The transformer operating data includes the load of each transformer, the operating status of the relay protection device, and the bus voltage; The environmental monitoring data includes temperature, humidity, and wind speed.

3. The method for multi-dimensional spatial data fusion and analysis of power grids according to claim 1, characterized in that: The spatiotemporal data standard set includes the timestamp of the data acquisition time, the load of each transformer, the bus voltage, the operating status of the relay protection device, the temperature, humidity and wind speed, as well as the sampling location of the load of each transformer, the bus voltage, the operating status of the relay protection device, the temperature, humidity and wind speed.

4. The method for multi-dimensional spatial data fusion analysis of power grids according to claim 1, characterized in that: The dynamic weight fusion mechanism is constructed based on the spatiotemporal data standard set to obtain data quality factors, timeliness factors, importance factors, and relevance factors, resulting in a fusion weight factor set, including: Construct a dynamic weighting model based on data quality factors, timeliness factors, importance factors, and relevance factors; The weighting factors of each dimension of the spatiotemporal data standard set at each time point are calculated based on the dynamic weighting model. A fusion weight factor set is constructed based on the weight factors of data from each dimension.

5. The method for multi-dimensional spatial data fusion analysis of power grids according to claim 4, characterized in that: The calculation of weight factors for each dimension of data in the spatiotemporal data standard set based on the dynamic weight model is based on the quality factor, timeliness factor, importance factor, relevance factor, data quality weight coefficient, timeliness weight coefficient, importance weight coefficient, and relevance weight coefficient corresponding to each dimension of data in each spatiotemporal data standard set. The data quality weight coefficient, timeliness weight coefficient, importance weight coefficient, and relevance weight coefficient are all dynamic weight adjustment parameters. Each of the aforementioned dynamic weight adjustment parameters is calculated based on the time decay coefficient of the dynamic weight adjustment parameter and the time difference between the current time and the data acquisition time.

6. The method for multi-dimensional spatial data fusion analysis of power grids according to claim 1, 4, or 5, characterized in that: The data quality factor is calculated based on the actual number of data points, the number of non-abnormal data points, and the number of data points to be collected for each weight factor data to be calculated within a set first time length. The time factor is calculated based on the time difference between the acquisition time of each weight factor data to be calculated and the current time, as well as the time decay coefficient corresponding to the data type; the data types include transformer operation data and environmental monitoring data. The importance factors are determined based on the data dimensions of each weight factor data to be calculated, and the preset data dimension-importance factor correspondence; the data dimensions include the load of each transformer, the operating status of the relay protection device, the bus voltage, temperature, humidity and wind speed; The correlation factor is calculated based on the correlation coefficient between the time series of each weight factor data to be calculated in the preset first time period and the time series of the data in the preset second time period, as well as the corresponding mutual information value.

7. The method for multi-dimensional spatial data fusion analysis of power grids according to claim 1, characterized in that: The three-dimensional spatiotemporal coordinate framework includes longitude coordinates, latitude coordinates, and time coordinates; The step of mapping each dimension of data in the spatiotemporal data standard set to the three-dimensional spatiotemporal coordinate frame involves mapping the value of each dimension of data in the spatiotemporal data standard set to the corresponding coordinate point in the three-dimensional spatiotemporal coordinate frame and performing three-dimensional interpolation to construct the spatiotemporal distribution function of each dimension of data.

8. The method for multi-dimensional spatial data fusion analysis of power grids according to claim 7, characterized in that: The method of fusing data from each dimension in a three-dimensional spatiotemporal coordinate framework based on a set of fusion weight factors to obtain a comprehensive correlation value for each spatiotemporal coordinate point is to calculate the comprehensive correlation value for each spatiotemporal coordinate point based on the weights of each dimension of data at that coordinate point, the spatiotemporal distribution function, and the spatiotemporal correlation strength between each dimension of data and other dimension of data at that coordinate point. The spatiotemporal correlation strength between each dimension of data and other dimension of data is calculated based on the geographical distance and electrical distance between the sampling locations of each dimension of data and other dimension of data, as well as the sampling time interval between each dimension of data and other dimension of data.

9. A power grid multi-dimensional spatial data fusion analysis system utilizing the power grid multi-dimensional spatial data fusion analysis method according to any one of claims 1-8, characterized in that, It includes modules for data acquisition and standard set construction, dynamic weight fusion, data mapping, and fusion analysis result generation. The data acquisition and standard set construction module is used to collect multi-dimensional spatiotemporal data of the power grid and form a spatiotemporal data standard set. The dynamic weight fusion module is used to construct a dynamic weight fusion mechanism based on the spatiotemporal data standard set, obtain data quality factors, timeliness factors, importance factors and relevance factors, and obtain a fusion weight factor set; The data mapping module is used to construct a three-dimensional spatiotemporal coordinate frame and map data of each dimension in the spatiotemporal data standard set to the three-dimensional spatiotemporal coordinate frame. The fusion analysis result generation module fuses data from each dimension of the three-dimensional spatiotemporal coordinate framework based on the fusion weight factor set, and obtains the comprehensive correlation value of each spatiotemporal coordinate point as the data fusion analysis result; the data fusion analysis result is used to formulate power grid operation strategies and judge power grid operation anomalies.

10. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the power grid multi-dimensional spatial data fusion analysis method according to any one of claims 1-8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the power grid multi-dimensional spatial data fusion analysis method according to any one of claims 1-8.

Images