A power distribution network health state diagnosis method and system based on big data analysis

Abstract

The application discloses a power distribution network health state diagnosis method and system based on big data analysis, relates to the technical field of power distribution network health evaluation, and comprises the following steps: S1, collecting multi-source health related data and generating a power distribution network original health data set; S2, performing cleaning, verification, dimension unification and space-time alignment to generate a multi-time state power distribution network health feature set; S3, mapping to a power supply grid to generate a grid power distribution network digital model; S4, constructing a power distribution network heterogeneous health atlas; S5, generating an atlas diagnosis result through an improved HGT network; S6, comprehensively evaluating reliability, adaptability, operation economy and new load access safety state to generate a grid health evaluation result; and S7, performing bearing boundary deduction and governance priority sorting to generate decision information. The application adopts multi-time state grid modeling and an improved HGT network to realize accurate diagnosis of the health state of the power distribution network.

Description

Technical Field

[0001] This invention relates to the field of power distribution network health assessment technology, and in particular to a method and system for diagnosing the health status of power distribution networks based on big data analysis. Background Technology

[0002] With the construction of new power systems, the rapid growth of electric vehicle charging facilities, and the large-scale integration of distributed power sources, health status diagnosis technologies for distribution networks, including their operational status, carrying capacity, and weak points, have received widespread attention. Existing distribution network health assessment methods primarily rely on operational monitoring data, equipment ledger data, fault repair records, or manual inspection results to statistically analyze individual indicators. They determine the distribution network's operational status through indicators such as load factor, voltage qualification rate, number of faults, and equipment defect level. However, these methods generally suffer from the following problems in practical applications: Data on distribution network operation, equipment ledgers, topology, fault repair, new load access, and planning and construction are scattered across different business systems. Data encoding, spatial location, and temporal references are inconsistent, making it difficult for existing methods to unify multi-source data onto the same equipment, feeder, and power grid. This results in fragmented data and inconsistent sources in health assessment results. Existing health diagnosis methods often rely on static indicator weighting or manual empirical rules for judgment, lacking continuous modeling of historical, current, and planned states. This makes it difficult to reflect the correlation between long-term equipment degradation trends, current operational risks, and the impact of future new load access, leading to delayed distribution network status diagnosis and an inability to identify areas with insufficient carrying capacity in advance. Furthermore, existing methods do not adequately consider distribution network topology, power supply, operational, and fault propagation relationships. They typically only identify the abnormal state of individual devices, making it difficult to track the risk transmission process between abnormal devices, power supply paths, and new load access points. This results in inaccurate location of weak points and a lack of reliable basis for prioritizing governance, affecting the accuracy of distribution network health status diagnosis and subsequent governance decisions.

[0003] Therefore, how to provide a method and system for diagnosing the health status of power distribution networks based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a method and system for diagnosing the health status of distribution networks based on big data analysis. This invention employs a combination of multi-temporal gridded modeling, a heterogeneous health map of the distribution network, and an improved HGT network. By collecting distribution network operation data, equipment ledger data, topology data, fault repair data, new load access data, and planning and construction data, it constructs distribution network health characteristics covering historical, current, and planned states. Furthermore, it maps equipment status, power supply paths, load access, and potential faults to the power supply grid, achieving a gridded expression of the distribution network's health status. Based on this, it constructs a heterogeneous health map of the distribution network that includes power supply relationships, operational relationships, and fault relationships, and introduces [the necessary technologies] into the improved HGT network. The topology gating mechanism improves upon the traditional HGT network's method of propagating heterogeneous relationships solely based on node and edge types. This allows node health representation to simultaneously reflect state changes under historical, current, and planned states, as well as actual power supply topology constraints and power flow safety boundaries. This enables more accurate identification of equipment malfunctions, power supply path anomalies, load access anomalies, and fault association anomalies. Furthermore, through equipment weakness identification, risk propagation path tracing, load boundary deduction, and governance priority ranking, it generates diagnostic decision-making information that can be used for distribution network operation and maintenance and planning governance. It boasts advantages such as high diagnostic accuracy, clear risk location, strong load boundary quantification capability, and applicability to dynamic diagnosis of distribution network health status under new load access scenarios.

[0005] A method for diagnosing the health status of a power distribution network based on big data analysis according to an embodiment of the present invention includes the following steps: S1. Collect multi-source health correlation data during the operation of the distribution network, and perform unified equipment encoding, spatial calibration and time stamping on the multi-source health correlation data to generate the original health dataset of the distribution network. S2. Perform data cleaning, anomaly verification, unit unification and spatiotemporal alignment on the original health dataset of the distribution network, and organize it hierarchically according to historical state, current state and planned state to generate a multi-temporal distribution network health feature set; S3. Based on the power supply area boundary, distribution network topology and equipment spatial location, map the multi-temporal distribution network health feature set to the power supply grid to generate a gridded distribution network digital model; S4. Extract distribution network equipment nodes and new load access nodes based on the gridded distribution network digital model, and construct a heterogeneous health map of the distribution network based on the power supply relationship, operation relationship and fault relationship between the distribution network equipment nodes and the new load access nodes; S5. Input the heterogeneous health map of the distribution network into the improved HGT network, perform map inference diagnosis on the heterogeneous health map of the distribution network, and generate map diagnosis results; S6. Based on the diagnostic results of the aforementioned map, a comprehensive evaluation is conducted on the reliability status, adaptability status, operational economy status, and security status of new load access in the distribution network, and grid health evaluation results are generated. S7. Based on the grid health assessment results, perform load-bearing boundary deduction and governance priority ranking to generate distribution network health status diagnosis decision information.

[0006] Optionally, S1 specifically includes: Acquire power distribution network operation data, equipment ledger data, topology data, fault repair data, defect and hidden danger data, automation coverage data, new load access data, and planning and construction data to generate multi-source health correlation data; Read the device name, device type, voltage level, feeder, transformer area, power grid, and asset number from the multi-source health association data, and generate a unified device code according to the preset coding rules; Based on the equipment installation location, line path, power supply area boundary and grid division results in the multi-source health association data, spatial location matching is performed on substations, feeders, distribution transformers, switchgear, distributed power sources and new load access points to generate spatial calibration results; The time base is unified based on the collection time, event occurrence time, fault handling time, defect registration time, and planned commissioning time in the multi-source health correlation data, and a time stamp result is generated. The unified encoding of the equipment, the spatial calibration result, and the time stamp result are written into the multi-source health association data to generate the original health dataset of the power distribution network.

[0007] Optionally, S2 specifically includes: Read the equipment unified code, spatial calibration result, time stamp result and health association field from the original health dataset of the power distribution network, clean the data with missing fields, duplicate data, incorrect format data and invalid collected data, and generate a cleaned health dataset. Based on the distribution network topology, equipment rated parameters, operating thresholds, and the time sequence of business records, the cleaned health dataset is subjected to anomaly verification. Abnormal data that does not meet the requirements of topology connection relationship, equipment operating boundary, and time continuity are removed, and a verified health dataset is generated. The voltage, current, load, capacity, number of faults, defect level, automation coverage and scale of new load access in the verified health dataset are standardized in terms of dimensions to generate a standardized health dataset. Using the unified device encoding, the spatial calibration result, and the time stamp result as indexes, the unified health dataset is spatiotemporally aligned to generate a spatiotemporally aligned health dataset. Historical health features are generated based on historical operational data in the spatiotemporal aligned health dataset; current health features are generated based on current operational data in the spatiotemporal aligned health dataset; and planned health features are generated based on planning and construction data in the spatiotemporal aligned health dataset. The historical health features, the current health features, and the planned health features are organized hierarchically to generate a multi-temporal distribution network health feature set.

[0008] Optionally, S3 specifically includes: Read the unified equipment codes, spatial calibration results, time stamp results, historical health features, current health features, and planned health features from the multi-temporal distribution network health feature set, and read the power supply area boundary, the distribution network topology, and the spatial location of the equipment to generate grid mapping basic data; The power supply range of the distribution network is divided into grids according to the power supply area boundary, generating a power supply grid, and the power supply grid is configured with grid number, administrative region attribute, power supply unit attribute and spatial boundary attribute; Based on the spatial location of the equipment, substations, feeders, distribution transformers, switchgear, distributed power sources, and new load access points are matched to the corresponding power supply grids to generate equipment grid affiliation relationships. Based on the power distribution network topology, the topology of the upper and lower level power supply paths, feeder connection paths, distribution transformer access paths and new load access paths in the equipment grid affiliation relationship is checked to generate a grid topology mapping relationship. Based on the time stamp results, the historical health features, the current health features, and the planned health features are respectively bound to the power supply grid, the device grid affiliation relationship, and the grid topology mapping relationship to generate multi-temporal grid health mapping results; Based on the multi-temporal grid health mapping results, a gridded model structure including a power supply grid layer, a power supply path layer, a power distribution equipment layer, and a load access layer is constructed to generate a gridded power distribution network digital model.

[0009] Optionally, S4 specifically includes: Read the power grid layer, power path layer, power distribution equipment layer and load access layer in the gridded distribution network digital model, extract the distribution network equipment nodes and new load access nodes, and generate a heterogeneous node set; Based on the unified equipment code, equipment type, voltage level, power supply grid to which the distribution network equipment node belongs, historical health characteristics, current health characteristics, and planned health characteristics of the equipment node, a set of equipment node attributes is generated; A set of load access node attributes is generated based on the access location, access capacity, access type, power grid to which the new load access node belongs, and access time. Based on the upstream and downstream power supply paths and feeder connection paths in the gridded distribution network digital model, the power supply relationships in the heterogeneous node set are determined; The operational relationships within the heterogeneous node set are determined based on the operational measurement associations, capacity occupancy associations, and planned access associations. Based on the associations of fault records, defects and potential hazards, and risk propagation in the heterogeneous node set, the fault relationships in the heterogeneous node set are determined; The power supply relationship, the operation relationship, and the fault relationship are labeled with relationship types to generate a heterogeneous relationship set; The heterogeneous node set, the device node attribute set, the load access node attribute set, and the heterogeneous relationship set are organized into a graph to generate a heterogeneous health graph of the distribution network.

[0010] Optionally, S5 specifically includes: Read the heterogeneous node set, heterogeneous relationship set, equipment node attribute set, and load access node attribute set from the heterogeneous health map of the power distribution network, and generate map inference input data; Node state data, edge relationship data, temporal feature data, and grid attribute data are extracted from the graph inference input data, and the node state data, edge relationship data, temporal feature data, and grid attribute data are input into the improved HGT network; The improved HGT network is used to perform heterogeneous relationship reasoning on the distribution network equipment nodes and new load access nodes in the heterogeneous health map of the distribution network, and to generate node health representations. Based on the node health representation, abnormal equipment operation, abnormal power supply path, abnormal load access, and abnormal fault association within the power supply grid are identified, and grid health diagnostic features are generated. The risk contribution of the distribution network equipment nodes is ranked according to the node health representation, and the equipment weakness identification results are generated. Based on the set of heterogeneous relationships and the identification results of the equipment weaknesses, the risk transmission direction in the power supply relationship, operation relationship and fault relationship is tracked to generate risk transmission paths; The grid health diagnosis features, the equipment weakness identification results, and the risk transmission path are combined to generate a map diagnosis result.

[0011] Optionally, the step of performing heterogeneous relationship reasoning on distribution network equipment nodes and new load access nodes in the heterogeneous health map of the distribution network by improving the HGT network and generating node health representations specifically involves: The node status data, edge relationship data, temporal characteristic data, and grid attribute data in the heterogeneous health map of the distribution network are input into the improved HGT network to obtain the node health representation; The improved HGT network includes a heterogeneous node embedding layer, a heterogeneous relation encoding layer, a topology-gated inference layer, and a power flow consistency output layer; The node status data, temporal feature data and grid attribute data are read through the heterogeneous node embedding layer to embed node features of distribution network equipment nodes and new load access nodes, thereby generating an initial representation of heterogeneous nodes. The edge relationship data is read through the heterogeneous relationship encoding layer, and the relationship type encoding is performed on the power supply relationship, operation relationship and fault relationship. The heterogeneous relationship representation is generated by combining the initial representation of the heterogeneous nodes. Based on the topology gating mechanism, the topology gating inference layer performs joint inference on the initial representation of heterogeneous nodes, the representation of heterogeneous relationships, the temporal feature data, and the grid attribute data to generate a topology gating node representation. The topology gating mechanism determines the temporal correlation strength between distribution network equipment nodes and new load access nodes based on node operation changes under historical, current, and planned states. It also generates relationship gating coefficients by combining power supply paths, switch on / off states, feeder connection states, load access locations, voltage boundaries, current boundaries, and capacity boundaries. These relationship gating coefficients are used to adjust the propagation weights of different heterogeneous relationships during the health state inference process. The propagation weights are reduced for heterogeneous relationships that are inconsistent with the actual power supply topology, power flow direction, or capacity constraints, while the propagation weights are increased for heterogeneous relationships that have continuous anomalies, sudden load increases, or planned access effects, thus obtaining a topology-gated node representation that conforms to the operating rules of the distribution network. The power flow consistency output layer performs power flow consistency verification and health status mapping on the topology gated node representation, and outputs a node health representation that includes the degree of node anomaly, the impact of risk propagation, and the health status level.

[0012] Optionally, S6 specifically includes: Read the grid health diagnosis features, equipment weakness identification results, and risk transmission paths from the atlas diagnosis results to generate health assessment input data; Based on the health assessment input data, power supply reliability characteristics, equipment anomaly characteristics, fault impact characteristics, and power supply recovery characteristics are extracted to generate reliability status assessment results. Based on the health assessment input data, extract the grid load-bearing characteristics, load growth characteristics, planned access characteristics, and equipment capacity margin characteristics to generate adaptive state assessment results; Based on the health assessment input data, extract line loss characteristics, equipment utilization characteristics, operation and maintenance cost characteristics, and governance input characteristics to generate an operational economic status evaluation result; Based on the health assessment input data, the scale of new load access, access location, access time period and risk transmission path are extracted to generate a new load access security status assessment result; The reliability status evaluation results, the adaptability status evaluation results, the operational economy status evaluation results, and the new load access security status evaluation results are comprehensively evaluated to generate a distribution network grid health score. The health level result is determined based on the distribution network grid health score, and the problem type label is determined based on the reliability status evaluation result, the adaptability status evaluation result, the operation economy status evaluation result, and the new load access security status evaluation result. The grid health score, health level result, problem type label, equipment weakness identification result, and risk transmission path are combined to generate a grid health evaluation result.

[0013] Optionally, S7 specifically includes: Read the distribution network grid health score, health level result, problem type label, equipment weak link identification result and risk transmission path from the grid health assessment result, and generate load-bearing boundary simulation input data; Based on the load-bearing boundary extrapolation input data, the power supply grid to be extrapolated is determined, and the equipment capacity, operating load, voltage status and power supply path corresponding to the power supply grid to be extrapolated are read to generate grid load-bearing analysis data; Based on the grid load analysis data, the access scale of charging pile access load, distributed power supply access load and regular growth load are respectively simulated, access location is verified and operation risk is checked to generate the load capacity boundary. Based on the health level results, the problem type labels, the carrying capacity boundaries, and the risk transmission paths, the grid health level corresponding to the power supply grid to be simulated is determined; Based on the identification results of the weak links in the equipment, the health score of the distribution network grid, the problem type label, and the carrying capacity boundary, the equipment risk level corresponding to the distribution network equipment node is determined. Based on the grid health level, the equipment risk level, the carrying capacity boundary, and the problem type label, the power grid to be simulated is sorted by the urgency of governance, and a governance priority is generated. Based on the grid health level, the equipment risk level, the carrying capacity boundary, and the governance priority, a visual diagnostic report is generated that includes the location of the power grid, the location of abnormal equipment, the risk transmission path, and the governance order. The grid health level, the equipment risk level, the carrying capacity boundary, the governance priority, and the visual diagnostic report are combined to generate power distribution network health status diagnostic decision information.

[0014] A power distribution network health status diagnosis system based on big data analysis according to an embodiment of the present invention includes the following modules: The multi-source data acquisition module is used to collect multi-source health correlation data during the operation of the distribution network, and to perform unified equipment encoding, spatial calibration and time stamping on the multi-source health correlation data to generate the original health dataset of the distribution network. The multi-temporal feature processing module is used to perform data cleaning, anomaly verification, unit unification and spatiotemporal alignment on the original health dataset of the distribution network, and to organize it hierarchically according to historical state, current state and planned state to generate a multi-temporal distribution network health feature set; The gridded modeling module is used to map multi-temporal distribution network health feature sets to the power supply grid based on the power supply area boundary, distribution network topology and equipment spatial location, and generate a gridded distribution network digital model. The heterogeneous health map construction module is used to extract distribution network equipment nodes and new load access nodes based on the gridded distribution network digital model, and to construct a heterogeneous health map of the distribution network based on the power supply relationship, operation relationship and fault relationship between the nodes. The graph reasoning and diagnosis module is used to input the heterogeneous health graph of the distribution network into the improved HGT network, perform graph reasoning and diagnosis on the heterogeneous health graph of the distribution network, and generate graph diagnosis results. The grid health assessment module is used to comprehensively evaluate the reliability status, adaptability status, operational economy status, and safety status of new load access of the distribution network based on the map diagnosis results, and generate grid health assessment results. The diagnostic decision output module is used to extrapolate the carrying capacity boundary and prioritize governance based on the grid health assessment results, and generate diagnostic decision information for the health status of the distribution network.

[0015] The beneficial effects of this invention are as follows: This invention addresses the problems of scattered storage, inconsistent coding, and inconsistent temporal and spatial benchmarks in distribution network operation data, equipment ledger data, topology data, fault repair data, new load access data, and planning and construction data by integrating multi-source health correlation data fusion, multi-temporal gridded modeling, and the cascaded integration of heterogeneous distribution network health maps with improved HGT networks. It employs unified equipment coding, spatial calibration, and time stamping to uniformly map health correlation data from different systems to the power supply grid, achieving a gridded expression of the distribution network's health status. In the feature organization stage, this invention constructs a multi-temporal distribution network health feature set according to historical, current, and planned states, incorporating long-term equipment degradation trends, current operational risks, and the impact of future new load access into the same diagnostic process, avoiding the problem that traditional static index evaluation cannot reflect state evolution. In the map reasoning stage, this invention constructs a feature set including distribution network health features... This paper presents a heterogeneous health map of the distribution network, including power grid equipment nodes, new load access nodes, power supply relationships, operational relationships, and fault relationships. A topology gating mechanism is introduced into the improved HGT network to determine the temporal correlation strength based on node operational changes under historical, current, and planned states. Relationship gating coefficients are generated by combining power supply paths, switch on / off states, feeder connection states, load access locations, voltage boundaries, current boundaries, and capacity boundaries. These coefficients adjust the propagation weights of different heterogeneous relationships during health state inference, improving the adaptability of node health representation to the actual power supply structure, state evolution process, and operational safety boundaries. Ultimately, the paper achieves intelligent output of grid health assessment results, equipment weakness identification results, risk transmission paths, carrying capacity boundaries, and governance priorities, effectively improving the accuracy of distribution network health state diagnosis, the clarity of risk location, and the reliability of new load access governance decisions. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a power distribution network health status diagnosis method based on big data analysis proposed in this invention; Figure 2 This is a schematic diagram of the structure of a power distribution network health status diagnosis system based on big data analysis proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figure 1A method for diagnosing the health status of a power distribution network based on big data analysis includes the following steps: S1. Collect multi-source health correlation data during the operation of the distribution network, perform unified equipment coding, spatial calibration and time stamping on the multi-source health correlation data, and generate the original health dataset of the distribution network. S2. Perform data cleaning, anomaly verification, unit unification and spatiotemporal alignment on the original health dataset of the distribution network, and organize it hierarchically according to historical state, current state and planned state to generate a multi-temporal distribution network health feature set; S3. Based on the power supply area boundary, distribution network topology and equipment spatial location, map the multi-temporal distribution network health feature set to the power supply grid to generate a gridded distribution network digital model; S4. Extract distribution network equipment nodes and new load access nodes based on the gridded distribution network digital model, and construct a heterogeneous health map of the distribution network based on the power supply relationship, operation relationship and fault relationship between the nodes; S5. Input the heterogeneous health map of the distribution network into the improved HGT network, perform map inference diagnosis on the heterogeneous health map of the distribution network, and generate map diagnosis results; S6. Based on the map diagnosis results, conduct a comprehensive evaluation of the reliability status, adaptability status, operational economy status, and safety status of new load access in the distribution network, and generate grid health evaluation results; S7. Based on the grid health assessment results, perform load-bearing boundary deduction and governance priority ranking to generate distribution network health status diagnosis decision information.

[0019] In this embodiment, S1 specifically refers to: Acquire power distribution network operation data, equipment ledger data, topology data, fault repair data, defect and hidden danger data, automation coverage data, new load access data, and planning and construction data to generate multi-source health correlation data; Read the device name, device type, voltage level, feeder, transformer area, power grid, and asset number from the multi-source health correlation data, and generate a unified device code according to the preset coding rules; Based on the equipment installation location, line path, power supply area boundary and grid division results in the multi-source health correlation data, spatial location matching is performed on substations, feeders, distribution transformers, switchgear, distributed power sources and new load access points to generate spatial calibration results; The time base is unified based on the collection time, event occurrence time, fault handling time, defect registration time and planned commissioning time in the multi-source health correlation data, and time stamp results are generated. The unified equipment coding, spatial calibration results, and time stamp results are written into the multi-source health correlation data to generate the original health dataset of the power distribution network. In the specific implementation process, distribution network operation data is accessed through the distribution automation master station, electricity consumption information collection system, and dispatch data interface. The collection cycle for voltage, current, load, and switch status is set to 5 minutes, and the collection cycle for transformer area load data is set to 15 minutes. Equipment ledger data, topology data, and automation coverage data are accessed through the production management system and geographic information system, with a field update cycle of one week. Fault repair data, defect and hidden danger data, new load access data, and planning and construction data are written according to event triggering. The unified equipment code adopts a 24-bit coding structure, where the first 6 bits represent the power supply unit and administrative region, the 7th to 10th bits represent the voltage level and equipment type, the 11th to 18th bits represent the feeder, transformer area, and power supply grid, and the 19th to 24th bits represent the asset sequence number. When the same equipment has different names in different systems, the asset number is used as the primary matching field, and the equipment type, voltage level, feeder, and equipment installation location are used as auxiliary matching fields. When the matching similarity reaches 0.95 or higher, the same unified equipment code is written. Spatial calibration uses a unified coordinate reference, with positioning errors for substations, distribution transformers, switchgear, distributed power sources, and new load access points controlled within 5 meters. Feeder line paths are calibrated using polygonal lines based on pole points, cable branch points, and switch nodes, with the distance between adjacent path points not exceeding 50 meters. Time stamping uses a unified clock source, with operational data time deviations controlled within 30 seconds. Fault repair data, defect and hidden danger data, and planning and construction data are uniformly converted to a standard time format. After completing the unified equipment coding, spatial calibration results, and time stamping results, records with a field completeness rate below 95% are marked for supplementation, and records with a duplication rate above 2% are merged to generate the original health dataset of the distribution network.

[0020] In this embodiment, S2 specifically refers to: Read the equipment unified code, spatial calibration result, time stamp result and health association field from the original health dataset of the distribution network, clean the data with missing field data, duplicate data, incorrect format data and invalid collected data, and generate a cleaned health dataset; Based on the distribution network topology, equipment rated parameters, operating thresholds, and the time sequence of business records, anomaly verification is performed on the cleaned health dataset to remove abnormal data that do not meet the requirements of topology connection, equipment operating boundaries, and time continuity, and a verified health dataset is generated. The voltage, current, load, capacity, number of faults, defect level, automation coverage and scale of new load access in the verified health dataset are standardized in terms of dimensions to generate a standardized health dataset. Using the unified device coding, spatial calibration results, and time stamping results as indexes, the unified dimensional health dataset is spatiotemporally aligned to generate a spatiotemporally aligned health dataset. Historical health features are generated based on historical operational data in the spatiotemporally aligned health dataset; current health features are generated based on current operational data in the spatiotemporally aligned health dataset; and planned health features are generated based on planning and construction data in the spatiotemporally aligned health dataset. Historical health features, current health features, and planned health features are hierarchically organized to generate a multi-temporal distribution network health feature set. In the specific implementation process, data cleaning is carried out using unified equipment coding, spatial calibration results, and time stamping results as the main indexes. Records with a field integrity rate of less than 95% are entered into the supplementary entry queue, and duplicate data are merged according to the unified equipment coding and time stamping results. Anomaly verification is performed according to the rated parameters and operating thresholds of the equipment. Records with voltage deviations exceeding ±15%, current exceeding 120% of the rated current, and load exceeding 130% of the equipment capacity are marked as abnormal data. When unifying the units, voltage is unified to kV, current to A, load and capacity to kW, and defect levels are mapped to levels 1 to 3. The automation coverage rate is converted to a decimal between 0 and 1. Spatiotemporal alignment is based on a 15-minute time granularity. Data with consecutive missing time periods of less than 30 minutes is filled by interpolation between adjacent time periods. Historical health features are generated from data within the past year, current health features are generated from data within the most recent month, and planned health features are generated from planned construction data and new load access data, ultimately forming a multi-temporal distribution network health feature set.

[0021] In this embodiment, S3 specifically refers to: Read the unified equipment codes, spatial calibration results, time stamp results, historical health features, current health features, and planned health features from the multi-temporal distribution network health feature set, and read the power supply area boundary, distribution network topology, and equipment spatial location to generate grid mapping basic data; The power supply range of the distribution network is divided into grids according to the power supply area boundary, generating a power supply grid, and configuring grid number, administrative region attribute, power supply unit attribute and spatial boundary attribute for the power supply grid; Based on the spatial location of the equipment, substations, feeders, distribution transformers, switchgear, distributed power sources and new load access points are matched to the corresponding power supply grids to generate equipment grid affiliation relationships; Based on the distribution network topology, the topology of the equipment grid affiliation relationship is checked for the upper and lower level power supply paths, feeder connection paths, distribution transformer access paths and new load access paths, and grid topology mapping relationship is generated. Based on the time stamping results, the historical health features, current health features, and planned health features are respectively bound to the power supply grid, equipment grid affiliation relationship, and grid topology mapping relationship to generate multi-temporal grid health mapping results; Based on the multi-temporal grid health mapping results, a gridded model structure including a power supply grid layer, a power supply path layer, a power distribution equipment layer, and a load access layer is constructed to generate a gridded distribution network digital model. In the specific implementation process, the power supply area boundary is represented by GIS surface data. The power supply grid is divided according to the power supply range of 10kV feeders and the service range of transformer substations. The area of ​​a single power supply grid is controlled between 0.5km² and 3km². In the core urban area, the area of ​​a single power supply grid is controlled between 0.5km² and 1km². In the rural area, the area of ​​a single power supply grid is controlled between 1km² and 3km². The grid number adopts a 16-digit coding structure, where the first 6 digits represent the administrative region, the 7th to 10th digits represent the power supply unit, the 11th to 13th digits represent the voltage level and grid type, and the 14th to 16th digits represent the grid sequence number. The equipment grid affiliation is determined according to the inclusion relationship between the spatial location of the equipment and the spatial boundary of the power supply grid. The spatial matching error of substations, distribution transformers, switchgear, distributed power sources and new load access points is controlled within 5m. Feeders are segmented and affiliated according to the intersection relationship between the line path and the boundary of the power supply grid. When a single feeder segment crosses multiple power supply grids, it is allocated according to the proportion of line length. Cross-grid segments with a length proportion of less than 5% are not separately affiliated and are merged into the adjacent main power supply grid. The grid topology mapping relationship is checked according to the power supply hierarchy of "substation - feeder - switchgear - distribution transformer - new load access point". The topological distance between adjacent devices in the upper and lower level power supply paths is no more than 2 levels. The feeder connection path is distinguished according to the status of normally open and normally closed switches. The distribution transformer access path and the new load access path are bound according to the actual access point. When the spatial location of the equipment is inconsistent with the topology, the distribution network topology relationship is used as the main judgment basis and the spatial location of the equipment is used as the verification basis. The multi-temporal grid health mapping results are organized into three temporal categories: historical, current, and planned. Historical health features are based on data from the past year, aggregated in monthly windows. Current health features are based on data from the most recent month, aggregated at a 15-minute granularity. Planned health features are based on planned construction data and new load access data, bound according to planned commissioning batches. Each power supply grid must contain at least four types of records: power supply path, distribution equipment, load access, and health features. Power supply grids lacking records of power supply paths or distribution equipment are set as invalid grids. The gridded model structure adopts a four-layer storage structure. The power supply grid layer records the grid number, administrative region attribute, power supply unit attribute, and spatial boundary attribute. The power supply path layer records the upper and lower level power supply paths, feeder connection paths, and distribution transformer access paths. The power distribution equipment layer records the unified equipment code, equipment type, voltage level, and equipment grid affiliation. The load access layer records the new load access points, access capacity, access time, and access location. The four layers are linked through the grid number and the unified equipment code to finally generate a gridded distribution network digital model.

[0022] In this embodiment, S4 specifically refers to: Read the power grid layer, power path layer, power distribution equipment layer and load access layer in the gridded distribution network digital model, extract the distribution network equipment nodes and new load access nodes, and generate a heterogeneous node set; Based on the unified equipment code, equipment type, voltage level, power supply grid to which the equipment belongs, historical health characteristics, current health characteristics, and planned health characteristics of the equipment nodes in the distribution network, generate a set of equipment node attributes; A set of attributes for load access nodes is generated based on the access location, access capacity, access type, power grid to which they belong, and access time corresponding to the new load access nodes. Based on the upstream and downstream power supply paths and feeder connection paths in the gridded distribution network digital model, determine the power supply relationships in the heterogeneous node set; Based on the operational measurement associations, capacity occupancy associations, and planned access associations in the heterogeneous node set, determine the operational relationships within the heterogeneous node set; Based on the associations of fault records, defects and potential hazards, and risk propagation in the heterogeneous node set, the fault relationships in the heterogeneous node set are determined; Label the power supply relationship, operation relationship, and fault relationship by relationship type to generate a heterogeneous relationship set; The heterogeneous node set, the device node attribute set, the load access node attribute set, and the heterogeneous relationship set are organized into a graph to generate a heterogeneous health graph of the distribution network. In the specific implementation process, distribution network equipment nodes are extracted according to substations, feeders, distribution transformers, switching equipment, and distributed power sources. New load access nodes are extracted according to charging pile access points, energy storage access points, and distributed photovoltaic access points. The node number uses the unified equipment code or access point code as the primary key. The equipment node attribute set is set as a 64-dimensional attribute vector, of which the basic equipment attributes account for 16 dimensions, the operating status attributes account for 20 dimensions, the historical health characteristics, current health characteristics, and planned health characteristics account for 24 dimensions, and the power supply grid attribute accounts for 4 dimensions. The load access node attribute set is set as a 32-dimensional attribute vector, with access capacity, access location, access type, power supply grid, and access time all written into the corresponding fields. Power supply relationships are generated according to the upstream and downstream power supply paths and feeder connection paths in the gridded distribution network digital model, with the relationship direction pointing from the upstream power supply equipment to the downstream power receiving equipment; operational relationships are generated according to operational measurement association, capacity occupancy association, and planned access association. An operational relationship is established when the load correlation coefficient between two nodes is not lower than 0.70 within a consecutive month, or the difference in capacity occupancy ratio changes is not greater than 10% for three consecutive times; fault relationships are generated according to fault record association, defect and hidden danger association, and risk propagation association. A fault relationship is established when the time interval between the occurrence of faults in two adjacent equipment levels within the same power supply path is less than 30 minutes. The heterogeneous relationship set is distinguished by relationship type identifiers: power supply relationship is marked as R1, operation relationship is marked as R2, and fault relationship is marked as R3. Each relationship record includes the starting node, ending node, relationship type, relationship direction, time stamp, and the power supply grid to which it belongs. The heterogeneous health map of the distribution network is jointly stored using a node attribute table and an adjacency table. When the number of nodes in a single power supply grid exceeds 500, subgraphs are established according to feeder segments. Subgraphs are connected to each other through feeder connections and nodes of the upper-level substation, ultimately generating the heterogeneous health map of the distribution network.

[0023] In this embodiment, S5 specifically refers to: Read the heterogeneous node set, heterogeneous relationship set, equipment node attribute set, and load access node attribute set from the heterogeneous health map of the distribution network, and generate map inference input data; Node state data, edge relationship data, temporal feature data, and grid attribute data are extracted from the graph inference input data, and then the node state data, edge relationship data, temporal feature data, and grid attribute data are input into the improved HGT network. By improving the HGT network, heterogeneous relationship reasoning is performed on distribution network equipment nodes and new load access nodes in the heterogeneous health map of the distribution network to generate node health representations. Based on node health representation, abnormal equipment operation, abnormal power supply path, abnormal load access, and abnormal fault association within the power grid are identified, and grid health diagnostic features are generated. Based on the node health representation, the risk contribution of distribution network equipment nodes are ranked to generate the equipment weakness identification results; Based on the heterogeneous relationship set and the identification results of equipment weaknesses, the risk transmission direction in power supply relationship, operation relationship and fault relationship is tracked to generate risk transmission path; The grid health diagnosis features, equipment weakness identification results, and risk transmission paths are combined to generate a map-based diagnosis result; In the specific implementation process, the input data for graph inference is organized in batches according to the power supply grid and feeder segments. Each batch contains 256 to 1024 nodes and no more than 5000 heterogeneous relationships. Node status data includes voltage deviation rate, load rate, fault frequency, defect level, and automation coverage status, which are uniformly mapped to a 128-dimensional vector. Edge relationship data is classified and coded according to power supply relationship, operation relationship, and fault relationship. Temporal feature data is divided into time slices according to historical state, current state, and planned state. Grid attribute data includes grid number, power supply unit attribute, spatial boundary attribute, and load access density. After the node health indicator is output, abnormal equipment operation is identified according to the following: voltage deviation exceeding ±7%, load rate exceeding 80%, and number of critical defects not less than 1. Abnormal power supply path is identified according to abnormal switch status, abnormal feeder connection path, and disconnection of upper and lower level power supply paths. Abnormal load access is identified according to the scale of new load access exceeding 70% of available capacity. Abnormal fault association is identified according to the fault interval between two adjacent levels of equipment being less than 30 minutes. Grid health diagnostic features are generated. The results of equipment vulnerability identification are sorted according to the risk contribution level corresponding to the node health representation. Within a single power supply grid, the top 10% of distribution network equipment nodes in terms of risk contribution level are selected as vulnerable equipment nodes. The risk transmission path starts from the vulnerable equipment node and is traced along the power supply relationship, operation relationship, and fault relationship. The maximum tracing depth is set to level 5, and tracing stops when the cumulative risk value of the path is lower than 0.20. Finally, the grid health diagnosis features, equipment vulnerability identification results, and risk transmission paths are bound according to the power supply grid number to generate a map diagnosis result.

[0024] In this embodiment, the heterogeneous relationship inference of distribution network equipment nodes and new load access nodes in the heterogeneous health map of the distribution network is performed by improving the HGT network, and node health representation is generated. Specifically: The node status data, edge relationship data, temporal characteristic data, and grid attribute data in the heterogeneous health map of the distribution network are input into the improved HGT network to obtain the node health representation; The improved HGT network includes a heterogeneous node embedding layer, a heterogeneous relation encoding layer, a topology-gated inference layer, and a power flow consistency output layer; By reading node status data, temporal feature data and grid attribute data through the heterogeneous node embedding layer, node feature embedding is performed on distribution network equipment nodes and new load access nodes to generate initial representations of heterogeneous nodes. By reading edge relationship data through the heterogeneous relationship coding layer, the relationship type of power supply relationship, operation relationship and fault relationship is encoded, and a heterogeneous relationship representation is generated by combining the initial representation of heterogeneous nodes; Based on the topology gating mechanism, the topology-gated inference layer performs joint inference on the initial representation of heterogeneous nodes, the representation of heterogeneous relationships, temporal feature data and grid attribute data to generate topology-gated node representations. The topology gating mechanism determines the temporal correlation strength between distribution network equipment nodes and new load access nodes based on node operation changes under historical, current, and planned states. It also generates relationship gating coefficients by combining power supply paths, switch on / off states, feeder connection states, load access locations, voltage boundaries, current boundaries, and capacity boundaries. The relationship gating coefficients are used to adjust the propagation weights of different heterogeneous relationships in the health state reasoning process. The propagation weights are reduced for heterogeneous relationships that are inconsistent with the actual power supply topology, power flow direction, or capacity constraints, while the propagation weights are increased for heterogeneous relationships that have continuous anomalies, sudden load increases, or planned access effects, thus obtaining a topology-gated node representation that conforms to the operation rules of the distribution network. The topology gated node representation is validated for power flow consistency and mapped to health status through the power flow consistency output layer. The output includes a node health representation that includes the degree of node anomaly, the impact of risk propagation, and the health status level. In the specific implementation process, the improved HGT network inherits the traditional HGT network's processing framework for type-aware encoding of heterogeneous nodes and heterogeneous relationships in its overall structure, including node embedding, relationship encoding, heterogeneous attention calculation, and node representation update process. However, in the distribution network health diagnosis scenario, a topology-gated inference layer is introduced to constrain and enhance the heterogeneous relationship propagation process in the traditional HGT network, replacing the traditional HGT network's message passing method based solely on node type and edge type.

[0025] First, distribution network equipment nodes and new load access nodes are mapped to 128-dimensional node initial vectors, and power supply relationships, operational relationships, and fault relationships are mapped to 64-dimensional relationship vectors. Three types of temporal input features are constructed according to historical state, current state, and planned state. The historical state represents the long-term operational trend of nodes, the current state represents the current voltage, current, load, and fault alarm status, and the planned state represents the potential impact of distributed power sources, electric vehicle charging piles, and energy storage devices after access. Subsequently, the topology gating inference layer generates relationship gating coefficients based on power supply paths, switch on / off states, feeder connection states, and load access locations. The propagation weight is reduced for disconnected switch relationships, cascading power supply relationships, invalid connection relationships, and access relationships that do not meet capacity boundaries; the propagation weight is increased for relationships with continuous heavy loads, frequent voltage exceedances, sudden increases in new loads, or significant planned access impacts. Compared to traditional HGT networks, the improved HGT network does not simply rely on existing semantic relationships in the graph for inference. Instead, it embeds the actual topology connections of the distribution network, temporal state changes, and power flow safety boundaries into the relationship propagation process, making the node health representation more consistent with actual operating mechanisms. Finally, the power flow consistency output layer verifies and corrects the topology gated node representation based on voltage deviation, current overrun, load factor, and capacity margin, and outputs the node anomaly level, risk propagation impact, and health status level, thereby improving the accuracy of weak device identification, anomaly propagation path judgment, and new load access risk diagnosis.

[0026] In this embodiment, S6 specifically refers to: Read the grid health diagnosis features, equipment weakness identification results and risk transmission paths from the map diagnosis results to generate health assessment input data; Based on the health assessment input data, power supply reliability characteristics, equipment anomaly characteristics, fault impact characteristics, and power supply recovery characteristics are extracted to generate reliability status assessment results. Based on the health assessment input data, extract the grid load-bearing characteristics, load growth characteristics, planned access characteristics, and equipment capacity margin characteristics to generate adaptive state assessment results; Based on the health assessment input data, extract line loss characteristics, equipment utilization characteristics, operation and maintenance cost characteristics, and governance input characteristics to generate operational economic status evaluation results; Based on the health assessment input data, extract the scale, location, time period, and risk transmission path of new load access, and generate the security status assessment results of new load access; The reliability status evaluation results, adaptability status evaluation results, operational economy status evaluation results, and new load access security status evaluation results are comprehensively evaluated to generate a distribution network grid health score. The health level is determined based on the distribution network grid health score, and the problem type label is determined based on the reliability status evaluation results, adaptability status evaluation results, operational economy status evaluation results, and new load access security status evaluation results. The grid health score, health level results, problem type labels, equipment weakness identification results, and risk transmission paths of the distribution network are combined to generate grid health evaluation results; In the specific implementation process, the health assessment input data is summarized according to the power supply grid, and the assessment cycle is set to one month. The power supply reliability characteristics include the average number of power outages, the average power outage duration, the number of users affected by the fault, and the fault recovery time. Among them, the records with a fault recovery time of more than 4 hours are marked as reliability risk records. The equipment abnormality characteristics include the number of heavy overloads, the number of voltage over-limits, the number of critical defects, and the number of automation failures. The distribution transformer load rate is recorded as a warning state when it exceeds 80%, and as a high-risk state when it exceeds 90%. The adaptability status evaluation results are generated based on the grid carrying capacity characteristics, load growth characteristics, planned access characteristics, and equipment capacity margin characteristics. Power grids with equipment capacity margins below 20% are marked as access-restricted grids, and power grids with new load access scales exceeding 70% of the current available capacity are marked as planned conflict grids. The operational economic status evaluation results are calculated based on line loss rate, equipment utilization rate, operation and maintenance costs, and governance investment. Power grids with line loss rates exceeding 8% or equipment utilization rates below 30% are included in the economic anomaly list. The safety status evaluation results for new load access are determined based on the scale, location, time period, and risk transmission path of the new load access. The safety risk level is increased when the access point is located downstream of high-risk equipment or the risk transmission path length exceeds level 3. The distribution network grid health score adopts a 100-point system. The health level results are divided into four categories: healthy, concerning, risky, and severe risk, according to the scores of no less than 85 points, 70 to 84 points, 60 to 69 points, and below 60 points. The distribution network grid health score, health level results, problem type labels, equipment weak link identification results, and risk transmission path are combined to generate the grid health evaluation results.

[0027] In this embodiment, S7 specifically refers to: Read the distribution network grid health score, health level result, problem type label, equipment weak link identification result and risk transmission path from the grid health assessment result, and generate the load-bearing boundary simulation input data; The power supply grid to be simulated is determined based on the input data of the load-bearing boundary simulation, and the equipment capacity, operating load, voltage status and power supply path corresponding to the power supply grid to be simulated are read to generate grid load-bearing analysis data; Based on grid load analysis data, the access scale of charging pile access load, distributed power supply access load and regular growth load are respectively simulated, access location is verified and operation risk is checked to generate load capacity boundaries. Based on the health level results, problem type labels, carrying capacity boundaries, and risk transmission paths, determine the grid health level corresponding to the power supply grid to be simulated; Based on the identification results of equipment weak links, distribution network grid health score, problem type label and carrying capacity boundary, determine the equipment risk level corresponding to the distribution network equipment node; Based on the grid health level, equipment risk level, carrying capacity boundary, and problem type label, the power grid to be simulated is ranked according to the urgency of governance, and a governance priority is generated. Based on grid health level, equipment risk level, carrying capacity boundary and governance priority, a visual diagnostic report is generated that includes the location of power grid, location of abnormal equipment, risk transmission path and governance sequence. By combining grid health level, equipment risk level, carrying capacity boundary, governance priority and visual diagnostic report, distribution network health status diagnostic decision information is generated; In the specific implementation process, the input data for the load-bearing boundary simulation is organized according to the power supply grid. Power supply grids with health level results of concern, risk, and severe risk are selected as power supply grids to be simulated. Power supply grids with a distribution network grid health score of less than 70 points are directly included in the key simulation queue. The grid load-bearing analysis data includes distribution transformer capacity, feeder capacity, current maximum operating load, minimum daily voltage, power supply path level, and identification results of equipment weak links. Among them, power supply grids with distribution transformer remaining capacity of less than 20% or feeder load rate of more than 80% are marked as load-bearing limited grids. The carrying capacity boundary is simulated separately for charging pile access load, distributed power supply access load, and regular growth load. The charging pile access load is simulated with four power levels: 7kW, 30kW, 60kW, and 120kW per gun. The distributed power supply access load is checked according to the access capacity accounting for 20%, 40%, 60%, and 80% of the distribution transformer capacity. The regular growth load is simulated according to the growth ratio of 5%, 10%, 15%, and 20%. The access location check is based on the following conditions: the power supply path does not exceed level 3, the voltage deviation at the access point does not exceed ±7%, and the distribution transformer load rate does not exceed 90%. The grid health level is determined comprehensively based on the health level results, problem type labels, carrying capacity boundaries, and risk transmission paths. When the carrying capacity boundary is lower than 80% of the planned access scale, or when the risk transmission path passes through high-risk equipment nodes, the grid health level is downgraded by one level. The equipment risk level is determined according to the identification results of equipment weaknesses and carrying capacity boundaries. Distribution network equipment nodes with a load rate exceeding 90%, critical defects, or located in the top 3 levels of the risk transmission path are marked as high-risk equipment nodes. The governance priorities are sorted according to grid health level, equipment risk level, carrying capacity boundary, and problem type label. The priority of grids with severe risks is set to level one, the priority of grids with risks is set to level two, and the priority of grids of concern is set to level three. The visual diagnostic report uses the power supply grid as the display unit, and marks the location of the power supply grid, the location of abnormal equipment, the risk transmission path, the carrying capacity boundary, and the governance order. Finally, the information is combined to generate the distribution network health status diagnostic decision information.

[0028] refer to Figure 2 A power distribution network health status diagnosis system based on big data analysis includes the following modules: The multi-source data acquisition module is used to collect multi-source health correlation data during the operation of the distribution network, and to perform unified equipment encoding, spatial calibration and time stamping on the multi-source health correlation data to generate the original health dataset of the distribution network. The multi-temporal feature processing module is used to perform data cleaning, anomaly verification, unit unification and spatiotemporal alignment on the original health dataset of the distribution network, and to organize it hierarchically according to historical state, current state and planned state to generate a multi-temporal distribution network health feature set; The gridded modeling module is used to map multi-temporal distribution network health feature sets to the power supply grid based on the power supply area boundary, distribution network topology and equipment spatial location, and generate a gridded distribution network digital model. The heterogeneous health map construction module is used to extract distribution network equipment nodes and new load access nodes based on the gridded distribution network digital model, and to construct a heterogeneous health map of the distribution network based on the power supply relationship, operation relationship and fault relationship between the nodes. The graph reasoning and diagnosis module is used to input the heterogeneous health graph of the distribution network into the improved HGT network, perform graph reasoning and diagnosis on the heterogeneous health graph of the distribution network, and generate graph diagnosis results. The grid health assessment module is used to comprehensively evaluate the reliability, adaptability, operational economy, and security status of the distribution network based on the map diagnosis results, and generate grid health assessment results. The diagnostic decision output module is used to extrapolate the carrying capacity boundary and prioritize governance based on the grid health assessment results, and generate diagnostic decision information for the health status of the distribution network.

[0029] Example 1: To verify the feasibility of this invention in practice, it was applied to the health status diagnosis system of a distribution network in a mixed power supply area of ​​the eastern urban area and the suburban industrial area of ​​a certain city, with a testing period of one month. This area includes old residential communities, commercial complexes, bus charging stations, small industrial parks, and distributed photovoltaic access points. The power supply structure includes two 110kV substations, 24 10kV feeders, 386 distribution transformers, 118 switchgear, 42 new load access points, and 96 power grids. Before implementation, the health assessment of this area mainly relied on manual inspections, static records, and individual operational indicators, which resulted in problems such as delayed detection of potential faults, inaccurate identification of transformer overload, and conservative or overly aggressive judgments on the capacity of new load access points. For example, after some bus charging stations were centrally connected, the load rate of adjacent distribution transformers consistently exceeded 85% during the evening peak hours. However, traditional methods could only address the issue after an alarm occurred, failing to identify risk transmission paths and prioritize remediation efforts in advance.

[0030] In this scenario, data is first integrated from the distribution automation master station, electricity consumption information collection system, production management system, geographic information system, fault repair system, and planning project database to form multi-source health correlation data. Operational measurement data is integrated at 5-minute intervals, transformer load data at 15-minute intervals, and fault repair data, defect and hazard data, and new load access data are integrated using event-triggered methods. The system performs unified equipment coding, spatial calibration, and time stamping on data from different sources to generate the original health dataset of the distribution network. This test integrated approximately 2.31 million valid operational measurement records, 5,268 equipment ledger records, 416 fault repair records, 692 defect and hazard records, 87 new load access records, and 54 planning and construction records.

[0031] Subsequently, the system cleans, verifies anomalies, unifies dimensions, and aligns the original health dataset of the distribution network in time and space, forming a multi-temporal distribution network health feature set. Historical health features are generated from data within the past year, current health features from operational data of the most recent month, and planned health features from planning and construction data and new load access data. During grid-based modeling, the system maps 386 distribution transformers, 118 switching devices, 24 feeders, and 42 new load access points to 96 power supply grids, and constructs a power supply grid layer, a power supply path layer, a distribution equipment layer, and a load access layer. This unifies equipment status, defect records, load changes, and planned access information scattered across different business systems under a single power supply grid.

[0032] After a month of testing, the present invention and the traditional static evaluation method were compared, and the results are shown in Table 1.

[0033] Table 1 Comparison of Application Effects of Distribution Network Health Status Diagnosis As shown in Table 1, this invention outperforms traditional static evaluation methods in terms of diagnostic efficiency, anomaly identification, weak link location, and load capacity boundary calculation. Traditional methods primarily rely on judging the exceedance of single indicators, failing to adequately consider the relationships between grids, devices, and load connection points. Therefore, they often only identify the results without understanding the underlying causes. This invention reflects the changing trends of the power supply grid in historical, current, and planned states through a multi-temporal distribution network health feature set. It expresses various relationships between devices, power supply paths, and new load connection points through a heterogeneous distribution network health map. Furthermore, by improving the health representation of HGT network output nodes, it can simultaneously identify abnormal grids, weak devices, and risk transmission paths.

[0034] In specific application results, the system output a total of 25 abnormal power supply grids, including 4 grids with a severe risk health level, 9 grids with a risk level, and 12 grids with a concern level. After manual review, 23 abnormal power supply grids were confirmed to be correctly identified. The main issues included persistently high transformer load rates, low voltage at the feeder ends, insufficient automation coverage of switching equipment, charging pile access capacity exceeding remaining capacity, and localized voltage fluctuations caused by reverse power flow from distributed photovoltaic systems. The system also output 38 weak points in equipment, including 11 high-load transformers, 7 switching equipment with abnormal automation operation, 5 feeder sections with insufficient load margin, and 15 distribution equipment nodes significantly affected by new load access. For the severe risk grids, after generating a governance priority, the system recommended prioritizing the 4 transformer overload grids and 2 feeder end voltage low paths. Maintenance personnel conducted on-site verification according to the governance priority, confirming that the system's ranking was largely consistent with expert judgment.

[0035] In the map diagnosis phase, the system extracts distribution network equipment nodes and new load access nodes based on the gridded distribution network digital model, establishing a heterogeneous health map of the distribution network that includes power supply relationships, operational relationships, and fault relationships. This results in 546 nodes and 2846 heterogeneous relationships, including 1226 power supply relationships, 1038 operational relationships, and 582 fault relationships. The system inputs the heterogeneous health map of the distribution network into the improved HGT network, performs heterogeneous relationship inference, and generates node health representations. For a 10kV feeder with high load fluctuations, the system not only identifies the high load rate of the three distribution transformers at the end of the feeder but also tracks abnormal automation operations of upstream switchgear, forming a risk transmission path of "feeder—switchgear—distribution transformer—charging pile access point." Compared to traditional static evaluation methods, this invention can simultaneously identify the impact of upstream weak equipment and downstream new load access.

[0036] This embodiment demonstrates that the present invention can solve the problems of data fragmentation, diagnostic lag, unclear risk transmission relationships, and difficulty in quantifying the carrying capacity boundaries of new loads in distribution network health status diagnosis, which are inherent in traditional methods. By outputting grid health level, equipment risk level, carrying capacity boundary, governance priority, and a visualized diagnostic report, power supply companies can identify high-risk areas in advance before the addition of new charging piles, distributed power sources, or regular load growth, avoiding the blind connection of new loads in weak power grids, thereby improving the accuracy, timeliness, and feasibility of distribution network health status diagnosis and governance.

[0037] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for diagnosing the health status of a power distribution network based on big data analysis, characterized in that, Includes the following steps: S1. Collect multi-source health correlation data during the operation of the distribution network, and perform unified equipment encoding, spatial calibration and time stamping on the multi-source health correlation data to generate the original health dataset of the distribution network. S2. Perform data cleaning, anomaly verification, unit unification and spatiotemporal alignment on the original health dataset of the distribution network, and organize it hierarchically according to historical state, current state and planned state to generate a multi-temporal distribution network health feature set; S3. Based on the power supply area boundary, distribution network topology and equipment spatial location, map the multi-temporal distribution network health feature set to the power supply grid to generate a gridded distribution network digital model; S4. Extract distribution network equipment nodes and new load access nodes based on the gridded distribution network digital model, and construct a heterogeneous health map of the distribution network based on the power supply relationship, operation relationship and fault relationship between the distribution network equipment nodes and the new load access nodes; S5. Input the heterogeneous health map of the distribution network into the improved HGT network, perform map inference diagnosis on the heterogeneous health map of the distribution network, and generate map diagnosis results; S6. Based on the diagnostic results of the aforementioned map, a comprehensive evaluation is conducted on the reliability status, adaptability status, operational economy status, and security status of new load access in the distribution network, and grid health evaluation results are generated. S7. Based on the grid health assessment results, perform load-bearing boundary deduction and governance priority ranking to generate distribution network health status diagnosis decision information.

2. The method for diagnosing the health status of a distribution network based on big data analysis according to claim 1, characterized in that, Specifically, S1 is: Acquire power distribution network operation data, equipment ledger data, topology data, fault repair data, defect and hidden danger data, automation coverage data, new load access data, and planning and construction data to generate multi-source health correlation data; Read the device name, device type, voltage level, feeder, transformer area, power grid, and asset number from the multi-source health association data, and generate a unified device code according to the preset coding rules; Based on the equipment installation location, line path, power supply area boundary and grid division results in the multi-source health association data, spatial location matching is performed on substations, feeders, distribution transformers, switching equipment, distributed power sources and new load access points to generate spatial calibration results; The time base is unified based on the collection time, event occurrence time, fault handling time, defect registration time, and planned commissioning time in the multi-source health correlation data, and a time stamp result is generated. The unified encoding of the equipment, the spatial calibration result, and the time stamp result are written into the multi-source health association data to generate the original health dataset of the power distribution network.

3. The method for diagnosing the health status of a distribution network based on big data analysis according to claim 1, characterized in that, Specifically, S2 is: Read the equipment unified code, spatial calibration result, time stamp result and health association field from the original health dataset of the power distribution network, clean the data with missing fields, duplicate data, incorrect format data and invalid collected data, and generate a cleaned health dataset. Based on the distribution network topology, equipment rated parameters, operating thresholds, and the time sequence of business records, the cleaned health dataset is subjected to anomaly verification. Abnormal data that does not meet the requirements of topology connection relationship, equipment operating boundary, and time continuity are removed, and a verified health dataset is generated. The voltage, current, load, capacity, number of faults, defect level, automation coverage and scale of new load access in the verified health dataset are standardized in terms of dimensions to generate a standardized health dataset. Using the unified device encoding, the spatial calibration result, and the time stamp result as indexes, the unified health dataset is spatiotemporally aligned to generate a spatiotemporally aligned health dataset. Historical health features are generated based on historical operational data in the spatiotemporal aligned health dataset; current health features are generated based on current operational data in the spatiotemporal aligned health dataset; and planned health features are generated based on planning and construction data in the spatiotemporal aligned health dataset. The historical health features, the current health features, and the planned health features are organized hierarchically to generate a multi-temporal distribution network health feature set.

4. The method for diagnosing the health status of a distribution network based on big data analysis according to claim 1, characterized in that, Specifically, S3 is: Read the unified equipment codes, spatial calibration results, time stamp results, historical health features, current health features, and planned health features from the multi-temporal distribution network health feature set, and read the power supply area boundary, the distribution network topology, and the spatial location of the equipment to generate grid mapping basic data; The power supply range of the distribution network is divided into grids according to the power supply area boundary, generating a power supply grid, and the power supply grid is configured with grid number, administrative region attribute, power supply unit attribute and spatial boundary attribute; Based on the spatial location of the equipment, substations, feeders, distribution transformers, switchgear, distributed power sources, and new load access points are matched to the corresponding power supply grids to generate equipment grid affiliation relationships. Based on the power distribution network topology, the topology of the upper and lower level power supply paths, feeder connection paths, distribution transformer access paths and new load access paths in the equipment grid affiliation relationship is checked to generate a grid topology mapping relationship. Based on the time stamp results, the historical health features, the current health features, and the planned health features are respectively bound to the power supply grid, the device grid affiliation relationship, and the grid topology mapping relationship to generate multi-temporal grid health mapping results; Based on the multi-temporal grid health mapping results, a gridded model structure including a power supply grid layer, a power supply path layer, a power distribution equipment layer, and a load access layer is constructed to generate a gridded power distribution network digital model.

5. The method for diagnosing the health status of a distribution network based on big data analysis according to claim 1, characterized in that, Specifically, S4 is: Read the power grid layer, power path layer, power distribution equipment layer and load access layer in the gridded distribution network digital model, extract the distribution network equipment nodes and new load access nodes, and generate a heterogeneous node set; Based on the unified equipment code, equipment type, voltage level, power supply grid to which the distribution network equipment node belongs, historical health characteristics, current health characteristics, and planned health characteristics of the equipment node, a set of equipment node attributes is generated; A set of load access node attributes is generated based on the access location, access capacity, access type, power grid to which the new load access node belongs, and access time. Based on the upstream and downstream power supply paths and feeder connection paths in the gridded distribution network digital model, the power supply relationships in the heterogeneous node set are determined; The operational relationships within the heterogeneous node set are determined based on the operational measurement associations, capacity occupancy associations, and planned access associations. Based on the associations of fault records, defects and potential hazards, and risk propagation in the heterogeneous node set, the fault relationships in the heterogeneous node set are determined; The power supply relationship, the operation relationship, and the fault relationship are labeled with relationship types to generate a heterogeneous relationship set; The heterogeneous node set, the device node attribute set, the load access node attribute set, and the heterogeneous relationship set are organized into a graph to generate a heterogeneous health graph of the distribution network.

6. The method for diagnosing the health status of a distribution network based on big data analysis according to claim 1, characterized in that, Specifically, S5 is: Read the heterogeneous node set, heterogeneous relationship set, equipment node attribute set, and load access node attribute set from the heterogeneous health map of the power distribution network, and generate map inference input data; Node state data, edge relationship data, temporal feature data, and grid attribute data are extracted from the graph inference input data, and the node state data, edge relationship data, temporal feature data, and grid attribute data are input into the improved HGT network; The improved HGT network is used to perform heterogeneous relationship reasoning on the distribution network equipment nodes and new load access nodes in the heterogeneous health map of the distribution network, and to generate node health representations. Based on the node health representation, abnormal equipment operation, abnormal power supply path, abnormal load access, and abnormal fault association within the power supply grid are identified, and grid health diagnostic features are generated. The risk contribution of the distribution network equipment nodes is ranked according to the node health representation, and the equipment weakness identification results are generated. Based on the set of heterogeneous relationships and the identification results of the equipment weaknesses, the risk transmission direction in the power supply relationship, operation relationship and fault relationship is tracked to generate risk transmission paths; The grid health diagnosis features, the equipment weakness identification results, and the risk transmission path are combined to generate a map diagnosis result.

7. The method for diagnosing the health status of a distribution network based on big data analysis according to claim 6, characterized in that, The process involves improving the HGT network to perform heterogeneous relationship reasoning on distribution network equipment nodes and new load access nodes in the heterogeneous health map of the distribution network, generating node health representations. Specifically: The node status data, edge relationship data, temporal characteristic data, and grid attribute data in the heterogeneous health map of the distribution network are input into the improved HGT network to obtain the node health representation; The improved HGT network includes a heterogeneous node embedding layer, a heterogeneous relation encoding layer, a topology-gated inference layer, and a power flow consistency output layer; By reading the node status data, the temporal feature data and the grid attribute data through the heterogeneous node embedding layer, node feature embedding is performed on the distribution network equipment nodes and the new load access nodes to generate an initial representation of the heterogeneous nodes. The edge relationship data is read through the heterogeneous relationship encoding layer, and the relationship type encoding is performed on the power supply relationship, operation relationship and fault relationship. The heterogeneous relationship representation is generated by combining the initial representation of the heterogeneous nodes. Based on the topology gating mechanism, the topology-gated inference layer performs joint inference on the initial representation of heterogeneous nodes, the representation of heterogeneous relationships, the temporal feature data, and the grid attribute data to generate a topology-gated node representation. The topology gating mechanism determines the temporal correlation strength between distribution network equipment nodes and new load access nodes based on node operation changes under historical, current, and planned states. It also generates relationship gating coefficients by combining power supply paths, switch on / off states, feeder connection states, load access locations, voltage boundaries, current boundaries, and capacity boundaries. These relationship gating coefficients are used to adjust the propagation weights of different heterogeneous relationships during the health state reasoning process. The propagation weights are reduced for heterogeneous relationships that are inconsistent with the actual power supply topology, power flow direction, or capacity constraints, while the propagation weights are increased for heterogeneous relationships that have continuous anomalies, sudden load increases, or planned access effects, thus obtaining a topology-gated node representation that conforms to the operating rules of the distribution network. The power flow consistency output layer performs power flow consistency verification and health status mapping on the topology gated node representation, and outputs a node health representation that includes the node anomaly degree, risk propagation impact, and health status level.

8. The method for diagnosing the health status of a distribution network based on big data analysis according to claim 1, characterized in that, Specifically, S6 is: Read the grid health diagnosis features, equipment weakness identification results, and risk transmission paths from the atlas diagnosis results to generate health assessment input data; Based on the health assessment input data, power supply reliability characteristics, equipment anomaly characteristics, fault impact characteristics, and power supply recovery characteristics are extracted to generate reliability status assessment results. Based on the health assessment input data, extract the grid load-bearing characteristics, load growth characteristics, planned access characteristics, and equipment capacity margin characteristics to generate adaptive state assessment results; Based on the health assessment input data, extract line loss characteristics, equipment utilization characteristics, operation and maintenance cost characteristics, and governance input characteristics to generate an operational economic status evaluation result; Based on the health assessment input data, the scale of new load access, access location, access time period and risk transmission path are extracted to generate a new load access security status assessment result; The reliability status evaluation results, the adaptability status evaluation results, the operational economy status evaluation results, and the new load access security status evaluation results are comprehensively evaluated to generate a distribution network grid health score. The health level result is determined based on the distribution network grid health score, and the problem type label is determined based on the reliability status evaluation result, the adaptability status evaluation result, the operation economy status evaluation result, and the new load access security status evaluation result. The grid health score, health level result, problem type label, equipment weakness identification result, and risk transmission path are combined to generate a grid health evaluation result.

9. The method for diagnosing the health status of a distribution network based on big data analysis according to claim 1, characterized in that, Specifically, S7 is: Read the distribution network grid health score, health level result, problem type label, equipment weak link identification result and risk transmission path from the grid health assessment result, and generate load-bearing boundary simulation input data; Based on the load-bearing boundary extrapolation input data, the power supply grid to be extrapolated is determined, and the equipment capacity, operating load, voltage status and power supply path corresponding to the power supply grid to be extrapolated are read to generate grid load-bearing analysis data; Based on the grid load analysis data, the access scale of charging pile access load, distributed power supply access load and regular growth load are respectively simulated, access location is verified and operation risk is checked to generate the load capacity boundary. Based on the health level results, the problem type labels, the carrying capacity boundaries, and the risk transmission paths, the grid health level corresponding to the power supply grid to be simulated is determined; Based on the identification results of the weak links in the equipment, the health score of the distribution network grid, the problem type label, and the carrying capacity boundary, the equipment risk level corresponding to the distribution network equipment node is determined. Based on the grid health level, the equipment risk level, the carrying capacity boundary, and the problem type label, the power grid to be simulated is sorted by the urgency of governance, and a governance priority is generated. Based on the grid health level, the equipment risk level, the carrying capacity boundary, and the governance priority, a visual diagnostic report is generated that includes the location of the power grid, the location of abnormal equipment, the risk transmission path, and the governance order. The grid health level, the equipment risk level, the carrying capacity boundary, the governance priority, and the visual diagnostic report are combined to generate power distribution network health status diagnostic decision information.

10. A distribution network health status diagnosis system based on big data analysis, comprising executing the distribution network health status diagnosis method based on big data analysis as described in any one of claims 1 to 9, characterized in that, Includes the following modules: The multi-source data acquisition module is used to collect multi-source health correlation data during the operation of the distribution network, and to perform unified equipment encoding, spatial calibration and time stamping on the multi-source health correlation data to generate the original health dataset of the distribution network. The multi-temporal feature processing module is used to perform data cleaning, anomaly verification, unit unification and spatiotemporal alignment on the original health dataset of the distribution network, and to organize it hierarchically according to historical state, current state and planned state to generate a multi-temporal distribution network health feature set; The gridded modeling module is used to map multi-temporal distribution network health feature sets to the power supply grid based on the power supply area boundary, distribution network topology and equipment spatial location, and generate a gridded distribution network digital model. The heterogeneous health map construction module is used to extract distribution network equipment nodes and new load access nodes based on the gridded distribution network digital model, and to construct a heterogeneous health map of the distribution network based on the power supply relationship, operation relationship and fault relationship between the nodes. The graph reasoning and diagnosis module is used to input the heterogeneous health graph of the distribution network into the improved HGT network, perform graph reasoning and diagnosis on the heterogeneous health graph of the distribution network, and generate graph diagnosis results. The grid health assessment module is used to comprehensively evaluate the reliability status, adaptability status, operational economy status, and safety status of new load access of the distribution network based on the map diagnosis results, and generate grid health assessment results. The diagnostic decision output module is used to extrapolate the carrying capacity boundary and prioritize governance based on the grid health assessment results, and generate diagnostic decision information for the health status of the distribution network.

Images