Grounding grid corrosion prediction method fused with soil parameter knowledge graph

Abstract

The invention discloses a grounding grid corrosion prediction method fused with a soil parameter knowledge graph, and relates to the related field of power system grounding grid corrosion monitoring, and the method comprises the steps: laying a multi-parameter soil sensor based on a grounding grid region for real-time collection, and generating a soil parameter data set; performing knowledge association based on the soil parameter data set in combination with soil corrosion mechanism data, and constructing a soil corrosivity knowledge graph; acquiring an operation electrical data set of the grounding grid in real time, and performing fusion evaluation on the operation electrical data set and the soil corrosivity knowledge graph to generate a grounding grid corrosion grade evaluation result; and performing corrosion prediction on the grounding grid according to a grounding grid corrosion level evaluation result, obtaining a corrosion prediction trend graph, backtracking to the soil corrosivity knowledge graph for verification, performing operation analysis on the grounding grid according to a verification result, and generating a grounding grid residual life prediction report. The technical problem that existing grounding grid corrosion prediction is insufficient in accuracy is solved, and the technical effect of improving corrosion state prediction accuracy is achieved.

Description

TECHNICAL FIELD

[0001] The application relates to the field of corrosion monitoring of grounding grids of power systems, in particular to a grounding grid corrosion prediction method fusing a soil parameter knowledge graph. BACKGROUND

[0002] As a key safety facility of a power system, the corrosion state of a grounding grid directly affects the operation safety of a power grid and the service life of equipment, and accurate corrosion prediction is of great significance for preventing grounding grid failure accidents. At present, the state of a grounding grid is mainly analyzed by means of periodic excavation inspection, single parameter monitoring or corrosion evaluation models based on statistical experience. However, the existing methods cannot effectively fuse the complex correlation between soil multi-parameters and operating electrical data, and lack of knowledge mining of corrosion mechanism and historical cases, resulting in obvious lag in early corrosion warning and trend prediction.

[0003] At the present stage, the related art has the technical problem of insufficient accuracy in predicting the corrosion of a grounding grid. SUMMARY

[0004] The application provides a grounding grid corrosion prediction method fusing a soil parameter knowledge graph, which generates a soil parameter data set by laying multi-parameter soil sensors to collect data in real time, constructs a soil corrosivity knowledge graph based on soil corrosion mechanism data, obtains a grounding grid operating electrical data set, and evaluates the corrosion grade by fusing the knowledge graph, predicts the corrosion according to the evaluation results, performs operation analysis after trend graph backtracking verification, and generates a remaining life prediction report, thereby solving the technical problem of insufficient accuracy in predicting the corrosion of a grounding grid existing in the prior art, and achieving the technical effect of improving the accuracy of corrosion state prediction.

[0005] The application provides a grounding grid corrosion prediction method fusing a soil parameter knowledge graph, which includes: laying multi-parameter soil sensors in the grounding grid area to collect data in real time and generate a soil parameter data set; correlating knowledge based on the soil parameter data set and soil corrosion mechanism data to construct a soil corrosivity knowledge graph; obtaining a grounding grid operating electrical data set in real time, fusing the operating electrical data set with the soil corrosivity knowledge graph for evaluation, and generating a grounding grid corrosion grade evaluation result; predicting the corrosion of the grounding grid according to the grounding grid corrosion grade evaluation result, obtaining a corrosion prediction trend graph, backtracking to the soil corrosivity knowledge graph for verification, performing operation analysis on the grounding grid according to the verification result, and generating a grounding grid remaining life prediction report.

[0006] In one possible implementation, multi-parameter soil sensors are deployed in the grounding grid area for real-time data acquisition to generate a soil parameter dataset. The following processes are then performed: soil distribution analysis is conducted throughout the grounding grid area to generate soil distribution data; the soil distribution data is then gridded to divide the data into gridded points; multi-parameter soil sensors are deployed in the grounding grid area according to the gridded point data to obtain a distributed monitoring network; a preset period is set, and the multi-parameter soil sensors are activated automatically to acquire data according to the preset period in conjunction with the distributed monitoring network, generating raw soil acquisition data; the raw soil acquisition data is then normalized to generate a soil parameter dataset.

[0007] In one possible implementation, multi-parameter soil sensors are deployed across a grounding grid area according to gridded point data to obtain a distributed monitoring network. The following processes are then performed: topology analysis is conducted on the grounding grid area based on the gridded point data; the radiation area boundary is delineated according to the grounding grid topology; the grounding grid area is divided into several equal-area grid cells based on the radiation area boundary; the multi-parameter soil sensors are deployed across several equal-area grid cells according to the gridded point data to determine multiple sensor nodes; and the geographical locations of the multiple sensor nodes are calibrated to construct a distributed monitoring network.

[0008] In a possible implementation, the grounding grid area is divided into several equal-area grid units based on the radiation area boundary, and the following processes are performed: the grounding grid area is monitored and analyzed based on the radiation area boundary to determine the monitoring area range; the grounding grid area is divided into several grid units according to the monitoring area using a square grid division method; and several equal-area grid units are numbered based on the several grid units to obtain several equal-area grid units.

[0009] In one possible implementation, the raw soil data is normalized to generate a soil parameter dataset, and the following processing is performed: multiple soil value ranges are set based on the raw soil data; the raw soil data is filtered based on the multiple soil value ranges to identify abnormal data points; the abnormal data points are marked; data repair is performed based on the marked data to obtain repaired data; the repaired data is normalized to construct a soil parameter dataset.

[0010] In a possible implementation, a soil corrosion knowledge graph is constructed by combining the soil parameter dataset with soil corrosion mechanism data for knowledge association, and the following processes are performed: corrosion analysis is conducted based on the soil parameter dataset to define multiple core entity types; the multiple core entity types are traversed and attribute-divided to determine multiple core entity attributes; relationships are defined for the multiple core entity types based on the multiple core entity attributes to generate multiple core entity relationship types; the multiple core entity types and the multiple core entity attributes are integrated according to the multiple core entity relationship types to construct a knowledge graph schema layer; soil corrosion mechanism data is retrieved, and the soil parameter dataset is mapped to the knowledge graph schema layer according to the soil corrosion mechanism data to determine multiple entity instance parameters; corrosion analysis is conducted based on the multiple entity instance parameters, and corrosion rules are set to connect the multiple entity instance parameters to construct the soil corrosion knowledge graph.

[0011] In a possible implementation, corrosion analysis is performed based on the soil parameter dataset, defining multiple core entity types and performing the following processing: multiple soil point entity instances are created for the multi-parameter soil sensors in the distributed monitoring network based on gridded point data, and the multiple soil point entity instances correspond to the multi-parameter soil sensors; corrosion analysis is performed on the soil parameter dataset based on the multiple soil point entity instances, creating multiple corrosion factor entity instances, and the multiple corrosion factor entity instances correspond to the multiple soil point entity instances.

[0012] In a possible implementation, corrosion rules are set to connect the parameters of the multiple entity instances to construct the soil corrosivity knowledge graph. The following processes are then performed: connecting the data of multiple entity instances based on the corrosion rules to construct an initial knowledge graph; introducing and parsing historical grounding corrosion case data to obtain multiple corrosive soil data; aligning and merging the multiple corrosive soil data according to the historical grounding corrosion case data onto the initial knowledge graph to generate a merged knowledge graph; performing frequent mining based on the merged knowledge graph to extract corrosion factor combination patterns; updating the merged knowledge graph according to the corrosion factor combination patterns to construct the soil corrosivity knowledge graph.

[0013] In a possible implementation, the operating electrical dataset of the grounding grid is acquired in real time. This operating electrical dataset is then fused and evaluated with the soil corrosion knowledge graph to generate a grounding grid corrosion level assessment result. The following processes are performed: an electrical monitoring network is constructed to collect data from the grounding grid in real time, obtaining the operating electrical dataset; the operating electrical dataset is mapped to the soil corrosion knowledge graph for expansion, identifying multiple soil monitoring points; the parameters of the multiple entity instances are spatially co-located with the multiple soil monitoring points to obtain dynamic electrical attribute nodes; based on the association between the dynamic electrical attribute nodes and soil corrosion factors, a fusion evaluation graph is constructed for feature learning, extracting a comprehensive feature vector; and corrosion level classification is performed based on the comprehensive feature vector to generate the grounding grid corrosion level assessment result.

[0014] In a possible implementation, the grounding grid operation electrical dataset is acquired in real time, and the following processing is performed: current sensors and voltage sensors are deployed to the accessible nodes of the distributed monitoring network to construct an electrical parameter monitoring network; the distributed monitoring network is synchronously sampled within the same time window through the electrical parameter monitoring network to obtain the original electrical dataset; the original electrical dataset is baseline corrected to obtain the grounding grid operation electrical dataset.

[0015] In a possible implementation, based on the association between the dynamic electrical attribute nodes and soil corrosion factors, a fusion evaluation graph is constructed for feature learning, and a comprehensive feature vector is extracted. The following processing is performed: the soil corrosion factors are extracted based on the soil corrosivity knowledge graph; the dynamic electrical attribute nodes are associated with the soil corrosion factors to construct a fusion evaluation graph; a graph neural network is used to perform multi-layer stacked graph convolution on the fusion evaluation graph to obtain semantic information of multiple nodes; the semantic information of the multiple nodes is aggregated to generate multiple node features; the fusion evaluation graph is transformed into graph-level features according to the multiple node features to construct the comprehensive feature vector.

[0016] In a possible implementation, the dynamic electrical attribute nodes are associated with the soil corrosion factors to construct a fusion evaluation map, and the following processing is performed: the dynamic electrical attribute nodes and the soil corrosion factors are aligned and analyzed; based on the analysis results, causal analysis is performed on the dynamic electrical attribute nodes to obtain the grounding resistance value, branch current distribution non-uniformity, and node potential difference variation coefficient of the dynamic electrical attribute nodes; based on the analysis results, causal analysis is performed on the soil corrosion factors to obtain the soil strong corrosion factor, soil electrochemical corrosion factor, and soil redox factor; based on a preset anomaly threshold judgment rule, the anomaly threshold judgment rule includes a first threshold, a second threshold, and a third threshold; when the grounding resistance... When the resistance value exceeds the first threshold, the location coordinates of the grounding resistance value are correlated with the soil strong corrosion factor to establish a high resistance-strong corrosion correlation; when the branch current distribution non-uniformity exceeds the second threshold, the location coordinates of the branch current distribution node are correlated with the soil electrochemical corrosion factor to establish a current anomaly-electrochemical corrosion correlation; when the node potential difference variation coefficient exceeds the third threshold, the location coordinates of the node potential difference variation coefficient are correlated with the soil redox factor to establish a potential non-uniformity-redox correlation; based on the high resistance-strong corrosion correlation, the current anomaly-electrochemical corrosion correlation, and the potential non-uniformity-redox correlation, the fusion evaluation map is constructed.

[0017] In a possible implementation, corrosion prediction of the grounding grid is performed based on the grounding grid corrosion level assessment results. A corrosion prediction trend map is obtained and backtracked to the soil corrosivity knowledge graph for verification. Based on the verification results, operational analysis of the grounding grid is performed, generating a grounding grid remaining life prediction report. The following processes are then performed: a multivariate time series dataset is constructed; time series learning is performed based on the multivariate time series dataset to obtain corrosion process time series regularity parameters; corrosion prediction is performed according to the grounding grid corrosion level assessment results and the corrosion process time series regularity parameters, and a corrosion prediction trend map is constructed; corrosion prediction is then performed based on the soil corrosivity knowledge graph. A full-cycle corrosion analysis is performed to extract the corrosion development trajectory. The corrosion prediction trend map is then matched with the corrosion development trajectory to generate a trend similarity score. If the trend similarity score is less than a preset similarity threshold, the verification fails. Abnormal deviations in the corrosion prediction trend map are identified and backtested. The corrosion prediction trend map is updated based on the test results until the trend similarity score is greater than or equal to the preset similarity threshold. If the trend similarity score is greater than or equal to the preset similarity threshold, the verification passes. Based on the corrosion prediction trend map, the grounding grid is analyzed to generate a remaining lifespan prediction report for the grounding grid.

[0018] In a possible implementation, corrosion prediction is performed based on the corrosion level assessment results of the grounding grid according to the corrosion process time sequence parameters, and a corrosion prediction trend map is constructed. The following processing is then performed: a multivariate time series dataset is constructed, and a long short-term memory network model is used to learn the time sequence pattern of the corrosion process according to the multivariate time series dataset to determine the corrosion process time sequence parameters; the grounding grid corrosion level assessment results are predicted according to the corrosion process time sequence parameters to obtain corrosion level change data at multiple time points; the corrosion level change data at multiple time points are visualized and labeled according to the time dimension to construct a corrosion prediction trend map.

[0019] In a possible implementation, when the trend similarity is greater than or equal to the preset similarity threshold, the verification is successful. Based on the corrosion prediction trend map, the grounding grid is analyzed for operation, and a grounding grid remaining life prediction report is generated. The following processes are performed: Trend calculation is performed on the verified corrosion prediction trend map to obtain the trend slope; trend inflection points are identified based on the trend slope, and trend stability is assessed to obtain trend fluctuation characteristic parameters; corrosion development analysis is performed based on the trend fluctuation characteristic parameters and the operating environment to obtain the corrosion acceleration stage; the peak power grid operation period of the grounding grid is retrieved, and the corrosion acceleration stage is overlapped with the peak power grid operation period to obtain the time overlap; performance degradation analysis is performed on the grounding grid based on the time overlap to determine the grounding grid corrosion risk level; the service life is calculated based on the grounding grid corrosion risk level and the characteristics of the grounding grid conductor material, and the grounding grid remaining life prediction report is generated. The proposed grounding grid corrosion prediction method, which integrates a soil parameter knowledge graph, firstly involves deploying multi-parameter soil sensors in the grounding grid area for real-time data acquisition to generate a soil parameter dataset. Next, based on this soil parameter dataset and soil corrosion mechanism data, a soil corrosivity knowledge graph is constructed. Then, the operational electrical dataset of the grounding grid is acquired in real-time and fused with the soil corrosion knowledge graph to generate a grounding grid corrosion level assessment result. Finally, based on the corrosion level assessment result, corrosion prediction is performed on the grounding grid, and the obtained corrosion prediction trend map is back-tested against the soil corrosion knowledge graph for verification. Based on the verification results, operational analysis of the grounding grid is conducted, generating a grounding grid remaining life prediction report. This method achieves the technical effect of improving the accuracy of corrosion state prediction. Attached Figure Description

[0020] Figure 1 A schematic flowchart illustrating the grounding grid corrosion prediction method integrating soil parameter knowledge graphs provided in this application embodiment.

[0021] Figure 2This is a schematic diagram illustrating the process of constructing a soil corrosivity knowledge graph in the grounding grid corrosion prediction method that integrates soil parameter knowledge graphs, as provided in the embodiments of this application. Detailed Implementation

[0022] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0023] This application provides a grounding grid corrosion prediction method that integrates soil parameter knowledge graphs, such as... Figure 1 As shown, the method includes: Step S100: Deploy multi-parameter soil sensors in the grounding grid area to collect data in real time and generate a soil parameter dataset.

[0024] Specifically, a multi-parameter soil sensor network is deployed at several key locations within the grounding grid area, such as the edges, center, and high-humidity zones. These multi-parameter soil sensors are devices integrating multiple sensing elements, capable of simultaneously measuring various physicochemical properties of the soil, such as resistivity sensors, pH electrodes, and humidity probes. These sensors utilize Internet of Things (IoT) technology to transmit data in real-time to a central data platform via wireless communication protocols. The parameters collected by the sensors include soil resistivity, pH value, water content, temperature, salinity, redox potential, and chloride ion concentration. The central data platform preprocesses the received raw data, including removing outliers through data cleaning, standardizing parameters of different dimensions through normalization, and aligning time series data to ensure consistent timestamps for all parameters. This ultimately generates a structured soil parameter dataset, which can be stored in tabular or time-series format. For example, in the substation grounding grid area, a sensor node is deployed every 10 meters, and data is collected every 5 minutes. After preliminary filtering by an edge computing device, the data is uploaded to the cloud platform. The platform uses Apache Spark for streaming data processing to generate a JSON format dataset containing timestamps, location coordinates, and multiple soil parameters.

[0025] In one possible implementation, multi-parameter soil sensors are deployed in the grounding grid area for real-time data acquisition to generate a soil parameter dataset. Step S100 further includes step S110, which involves traversing the grounding grid area to perform soil distribution analysis and generate soil distribution data. Specifically, in the area where the grounding grid is located, a combination of a Geographic Information System (GIS) system and remote sensing technology is used to perform spatial analysis on key parameters such as soil type, moisture distribution, and salinity content. By collecting historical geological survey data and conducting on-site surveys, the distribution areas of different soil types, such as clay areas, sandy soil areas, and high-salt-alkali areas, are identified, and structured soil distribution data is generated. This data is stored in the form of vector layers or raster data, containing the soil type identifier, spatial boundary coordinates, and main physicochemical properties of each area.

[0026] Step S120: The soil distribution data is processed into a grid, dividing the data into gridded points. Specifically, based on the generated soil distribution data, the monitoring area is spatially discretized using a square grid method. Specifically, according to the grounding grid topology and monitoring accuracy requirements, the grid unit size is set, such as 10m × 10m, dividing the entire grounding grid area into several equal-area grid units. Each grid unit is assigned a unique number, and its center point coordinates, soil type, and boundary range are recorded to form a gridded point dataset, which provides a spatial reference for sensor deployment.

[0027] Step S130 involves deploying multi-parameter soil sensors across the grounding grid area according to the gridded point data to obtain a distributed monitoring network. Specifically, based on the generated gridded point data, multi-parameter soil sensors are deployed at the center or key locations of each grid cell to form a distributed monitoring network covering the entire area.

[0028] Step S140: Set a preset cycle. Based on this preset cycle, activate multi-parameter soil sensors in conjunction with the distributed monitoring network to automatically collect data, generating raw soil data. Specifically, set a uniform data collection cycle, such as once every 5 minutes. Send a collection command to the distributed monitoring network through the central control platform to activate all sensor nodes for synchronous operation. The sensors perform real-time measurements according to preset parameters and upload the raw data to the data platform via the communication network. The raw data is stored in time-series format, including timestamps, sensor numbers, measurement parameters, and values, forming the initial raw soil dataset.

[0029] Step S150 involves normalizing the raw soil data to generate a soil parameter dataset. Specifically, the acquired raw soil data undergoes data cleaning and standardization. Reasonable numerical ranges for each parameter are set based on historical data and expert experience; outlier data points are identified and marked; interpolation or mean-filling methods are used to repair the outlier data; the repaired data is then normalized to convert parameters of different dimensions to the [0,1] interval, eliminating the influence of dimensions. After processing, a structured soil parameter dataset is generated, stored in tabular or time-series database format, for use in knowledge graph construction and corrosion analysis.

[0030] In one possible implementation, multi-parameter soil sensors are deployed across the grounding grid area according to gridded point data to obtain a distributed monitoring network. Step S130 further includes step S131, which involves performing topological analysis on the grounding grid area based on the gridded point data, dividing the radiation area boundaries according to the grounding grid topology, and dividing the grounding grid area into several equal-area grid units based on the radiation area boundaries. Specifically, based on the gridded point data generated in step S120, electrical topological analysis is performed on the grounding grid area to identify the main radiation paths and branch structures of the grounding grid. According to the electrical connection relationship and spatial layout of the grounding grid, multiple radiation areas are divided, each radiation area radiating outward from the center of the grounding electrode or main conductor to form a closed boundary. Based on these radiation area boundaries, a spatial geometric partitioning method is used to further subdivide the entire grounding grid area into several equal-area grid units, each unit being evenly distributed in space to provide a more refined spatial reference for sensor deployment.

[0031] Step S132 involves traversing several equal-area grid cells and deploying multi-parameter soil sensors according to the gridded point data, thus determining multiple sensor nodes. Specifically, based on the divided equal-area grid cells, each grid cell is traversed, and a multi-parameter soil sensor is deployed at the geometric center of each cell according to the gridded point data determined in step S120. During deployment, the communication distance and data coverage range between sensors are considered to ensure that each sensor node can effectively collect soil parameters from its cell. Finally, the spatial location and number of all sensor nodes are determined, forming a complete sensor node set.

[0032] Step S133 involves calibrating the geographical locations of multiple sensor nodes to construct a distributed monitoring network. Specifically, each sensor node identified in step S132 is calibrated using the GPS or BeiDou system to obtain its latitude and longitude coordinates and record its elevation information. All calibrated sensor nodes are then networked using a wireless communication protocol to construct a distributed monitoring network covering the entire grounding network area.

[0033] In one possible implementation, the grounding grid area is divided into several equal-area grid units based on the radiation area boundary. Step S131 further includes step S1311, which involves monitoring and analyzing the grounding grid area based on the radiation area boundary to determine the monitoring area range. Specifically, using the radiation area boundary as a reference boundary, a preset safety distance is uniformly extended outward to form the monitoring area range. This safety distance is determined according to on-site safety regulations and monitoring requirements, for example, it can be set to 5-10 meters to ensure that the monitoring range completely covers the grounding grid body and its surrounding potentially affected soil environment.

[0034] Step S1312 involves dividing the grounding grid area into several grid units according to the monitoring area using a square grid division method. Specifically, based on the monitoring area range determined in step S1311, each radiation area is spatially discretized using a square grid division method. A uniform grid unit size is set, dividing each monitoring area into several square grid units, where the grid side length is set according to the monitoring accuracy requirements. Each grid unit has a unique spatial code, recording the coordinate information of its four corner points to ensure seamless connection between grid units and full coverage of the monitoring area.

[0035] Step S1313 involves numbering several grid cells to obtain several equal-area grid cells. Specifically, all grid cells divided in step S1312 are systematically numbered, and a unique identifier is assigned to each cell using the rule of "region code-row number-column number". For example, the grid cell in the 2nd row and 3rd column within the radiation area R1 is numbered "R1-2-3". After numbering, a structured equal-area grid cell dataset is formed. This dataset contains information such as the cell number, spatial extent, and center point coordinates of each cell, which is used to provide a standardized spatial basis for sensor deployment.

[0036] In one possible implementation, the raw soil data is normalized to generate a soil parameter dataset. Step S150 further includes step S151, which involves setting multiple soil value ranges based on the raw soil data, and filtering the raw soil data based on these ranges to identify outlier data points. Specifically, based on historical statistical values ​​and industry standards for soil parameters, reasonable value ranges are set for each soil parameter, such as resistivity, pH, and salinity. The raw soil data is then iterated through, and the measured values ​​of each parameter are compared with the preset ranges to identify outlier data points. Simultaneously, time series analysis is used to identify abrupt anomalies, such as drastic fluctuations in a parameter within a short period.

[0037] Step S152 involves marking abnormal data points and performing data repair based on the marked data to obtain repaired data. Specifically, all abnormal data points identified in step S151 are marked, and their sensor number, parameter type, anomaly type, and occurrence time are recorded. Appropriate data repair strategies are adopted for different types of anomalies. For example, for isolated anomalies, linear interpolation of data from before and after the anomaly is used for repair; for continuous anomaly segments, sliding window mean or median is used for filling; for systemic anomalies caused by sensor failures, spatial interpolation repair is performed by combining data from neighboring sensors. After repair, a repaired dataset is formed to ensure data continuity and reliability.

[0038] Step S153 involves normalizing the remediation data to construct a soil parameter dataset. Specifically, the remediation data obtained in step S152 is normalized, converting soil parameters of different dimensions and ranges to the [0,1] interval. A minimum-maximum normalization method is used, calculating the minimum and maximum values ​​for each parameter column and converting them according to the minimum-maximum normalization formula. This normalization eliminates the influence of dimensions, ultimately generating a structured soil parameter dataset, stored in tabular or time-series format, containing information such as timestamps, sensor numbers, and normalized multi-parameter values.

[0039] Step S200: Based on the soil parameter dataset and soil corrosion mechanism data, knowledge association is performed to construct a soil corrosivity knowledge graph.

[0040] Specifically, soil corrosion mechanism data refers to the scientific knowledge describing the soil corrosion process, including electrochemical, chemical, and biological corrosion principles. By linking soil parameters with corrosion mechanisms using knowledge graph technology, the process begins with: First, natural language processing tools, such as Stanford CoreNLP, are used to extract corrosion mechanism data from soil corrosion literature and standards, forming structured rules. Then, a schema layer of the knowledge graph is constructed based on a graph database, defining entity types, attributes, and relation types. Finally, the soil parameter dataset is mapped to the schema layer, generating entity instances, and these instances are connected through corrosion rules to construct a complete knowledge graph.

[0041] like Figure 2As shown, in one possible implementation, a soil corrosivity knowledge graph is constructed by combining the soil parameter dataset with soil corrosion mechanism data for knowledge association. Step S200 further includes step S210, which involves performing corrosion analysis based on the soil parameter dataset and defining multiple core entity types. Specifically, core entity types are object categories in the knowledge graph and are the core nodes for constructing the knowledge network. By analyzing the data structure and corrosion mechanism of the soil parameter dataset, multiple core entity types are defined in the knowledge graph, including soil location, corrosion factor, corrosion type, and corrosion level. For example, the node labeling function of the graph database is used to define these types: a SoilPoint label is created for "soil location", a CorrosionFactor label is created for "corrosion factor", a CorrosionType label is created for "corrosion type", and a CorrosionLevel label is created for "corrosion level".

[0042] Step S220 involves traversing the multiple core entity types to classify attributes and determine multiple core entity attributes. Specifically, core entity attributes are key-value pairs describing the specific characteristics of an entity, used to characterize the entity's state and features in detail. Specific attribute fields are defined for each core entity type. For example, for a soil location entity, geographic location identifier and sampling time attributes are defined; for a corrosion factor entity, factor name, measured value, and unit attributes are defined; for a corrosion type entity, corrosion type name attribute is defined, with values ​​including "chemical corrosion," "electrochemical corrosion," etc.; for a corrosion level entity, level identifier and descriptive information attributes are defined. These attributes are defined using the attribute key mechanism of the graph database.

[0043] Step S230: Define relationships between multiple core entity types based on the attributes of the multiple core entities, generating multiple core entity relationship types. Specifically, a core entity relationship type refers to a directed edge type connecting different entity types, expressing a specific semantic association between entities. Semantic relationships between entities are defined based on corrosion mechanisms and domain knowledge. These are created using the relationship type functionality of a graph database, such as a "located" relationship from a soil location to a geographical location, a "has" relationship from a soil location to a corrosion factor, a "promotes" relationship from a corrosion factor to a corrosion type, and a "leads to" relationship from a corrosion type or corrosion factor to a corrosion level.

[0044] Step S240 involves integrating the multiple core entity types and attributes according to their relational types to construct a knowledge graph schema layer. Specifically, the knowledge graph schema layer is the metadata layer of the knowledge graph, serving as a framework for defining entity types, attribute structures, and relational types. By integrating the aforementioned defined entity types, attributes, and relational types into a unified knowledge graph schema layer, and using the schema definition capabilities of a graph database, a complete metadata framework is constructed. For example, attributes, vertex labels, and edge labels can be defined using the Gremlin language of Apache TinkerPop to define an attribute graph schema.

[0045] Step S250: Retrieve soil corrosion mechanism data and map the soil parameter dataset to the knowledge graph pattern layer according to the soil corrosion mechanism data to determine multiple entity instance parameters. Specifically, each parameter in the soil parameter dataset is converted into an entity instance parameter, which is a set of attribute values ​​of a specific entity instance, filled based on actual measurement data. For example, the record "chloride ion concentration: 0.15%, sampling time: 2025-06-01 10:00:00, location: soil point A" in the soil parameter dataset will be mapped and created as a corrosion factor entity instance, where the factor name is set to "chloride ion concentration", the measured value is set to 0.15, and the unit is set to "percentage". At the same time, a corresponding soil point entity instance is created, containing the geographic location identifier "soil point A" and the sampling time "2025-06-01 10:00:00" attributes, and the two entity instances are connected by a "have" relationship.

[0046] Step S260: Corrosion analysis is performed based on the parameters of the multiple entity instances. Corrosion rules are set to connect the parameters of the multiple entity instances, constructing the soil corrosivity knowledge graph. Specifically, threshold rules are set based on soil electrochemical principles to automatically create corrosion type entity instances and establish relationships. For example, a rule-based inference engine is used: when the chloride ion concentration is greater than 0.1%, a "pitting corrosion" corrosion type entity instance is automatically created, and a "promoting" relationship is used to point from the chloride ion corrosion factor instance to this corrosion type instance; when the pH value is less than 6.5, an "acidic corrosion" corrosion type entity instance is created; when the redox potential is < -200mV, an "anaerobic corrosion" corrosion type entity instance is created. These rules can be executed through the Drools rule engine.

[0047] In one possible implementation, corrosion analysis is performed based on the soil parameter dataset, defining multiple core entity types. Step S210 further includes step S211, creating multiple soil point entity instances for the multi-parameter soil sensors in the distributed monitoring network based on gridded point data. These multiple soil point entity instances correspond to the multi-parameter soil sensors. Specifically, based on the generated gridded point data, a corresponding soil point entity instance is created for each multi-parameter soil sensor in the distributed monitoring network. Each soil point entity instance contains attributes such as a unique identifier, geographic coordinates, and deployment time, and establishes a one-to-one correspondence with the actually deployed sensor. For example, in the knowledge graph, an entity instance named "Soil Point_S001" is created for the sensor numbered "S001," setting its "longitude" attribute to 119.5°E, its "latitude" attribute to 32.2°N, and its "deployment time" attribute to "2025-06-01 08:00:00." This correspondence is used to map physical sensors to entity instances in the knowledge graph.

[0048] Step S212 involves performing corrosion analysis on the soil parameter dataset based on multiple soil point entity instances, creating multiple corrosion factor entity instances. These corrosion factor entity instances correspond to the multiple soil point entity instances. Specifically, based on the created soil point entity instances and combined with measured data from the soil parameter dataset, corrosion mechanism analysis is performed to identify and create multiple corrosion factor entity instances. Each corrosion factor entity instance includes attributes such as factor type, measured value, and unit of measurement, and is associated with its corresponding soil point entity instance through a "have" relationship. This correspondence constructs a semantic association between soil environmental parameters and corrosion factors, providing data support for corrosion inference in the knowledge graph.

[0049] In one possible implementation, corrosion rules are set to connect the parameters of the multiple entity instances, constructing the soil corrosivity knowledge graph. Step S260 further includes step S261, connecting the data of multiple entity instances based on the corrosion rules to construct an initial knowledge graph. Specifically, the initial knowledge graph is a preliminary knowledge network constructed based on basic corrosion rules, containing basic entities, attributes, and relational structures. Based on the defined corrosion rules, the created entity instances are connected through specific relations. Using the batch operation function of the graph database, the CREATE or MERGE command of the Cypher statement is used to establish relations between entities. At the same time, more complex relational networks are established based on multi-parameter combination rules, such as creating a "cause" relation pointing to the "high corrosion level" entity when "acid corrosion" and "pitting corrosion" exist simultaneously.

[0050] Step S262 involves analyzing historical grounding corrosion case data to obtain multiple sets of corroded soil data. Specifically, historical grounding corrosion case data consists of records of past grounding grid corrosion events, including environmental parameters and corrosion outcome information. Corroded soil data is a set of soil parameters directly related to corrosion extracted from historical cases. Structured data is extracted from the grounding grid historical maintenance database or corrosion case library, including the location of the grounding grid where corrosion faults occurred, the corresponding soil parameters, corrosion morphology characteristics, and corrosion severity rating. ETL tools are used to analyze the data and extract key fields such as geographical location, soil parameter values, corrosion type descriptions, and actual corrosion levels.

[0051] Step S263 involves aligning and merging the multiple corroded soil data according to the historical grounding corrosion case data onto the initial knowledge graph, generating a merged knowledge graph. Specifically, alignment and merging refers to the process of identifying entities from different sources as the same object and integrating their attributes. The merged knowledge graph is an enhanced knowledge network that integrates real-time monitoring data and historical case knowledge. Entity alignment technology is used to merge historical case data with the initial knowledge graph, performing entity matching based on geographic location identifiers and timestamps. For existing soil point entities, attributes are updated using the MERGE statement; for new corrosion types and corrosion levels, new nodes are created and relationships are established. For example, when the historical data contains a "microbial corrosion" type not included in the initial graph, a new corrosion type node is created and connected to the corresponding corrosion level through a "cause" relationship. Simultaneously, weight scores are added to relationships validated in historical cases.

[0052] Step S264: Based on the merged knowledge graph, frequent mining is performed to extract corrosion factor combination patterns. The merged knowledge graph is then updated according to these patterns to construct the soil corrosivity knowledge graph. Specifically, frequent mining refers to a data mining technique that discovers frequently occurring item sets from data. Corrosivity factor combination patterns refer to regular patterns where multiple corrosion factors co-occur and lead to specific corrosion results. The soil corrosivity knowledge graph is the final knowledge graph after pattern mining and rule optimization, possessing the ability to automatically infer corrosion levels from multi-factor combinations. Frequent pattern mining algorithms, such as FP-Growth or Apriori, are applied to analyze the co-occurrence relationships of corrosion factors in the merged knowledge graph. All corrosion factors and their associated corrosion levels are extracted from the graph database, and frequent itemsets are mined. For example, it is found that when the three factors "chloride ion concentration > 0.1%", "pH < 6.5", and "redox potential < -200mV" appear simultaneously, there is an 85% probability of leading to a "severe corrosion level". Based on these patterns, new inference rules are created in the graph: when the factor combination appears in new soil data, a "cause" relationship for "severe corrosion level" is automatically created, and a confidence attribute is set. Simultaneously, summary nodes are created for frequent combinations, such as a "coastal acidic anaerobic environment" composite factor node, simplifying the graph structure and improving query efficiency.

[0053] Step S300: Obtain the operating electrical dataset of the grounding grid in real time, and fuse the operating electrical dataset with the soil corrosivity knowledge graph to generate a grounding grid corrosion level assessment result.

[0054] Specifically, electrical sensors, such as current transformers and voltage probes, are installed at key nodes of the grounding grid. Operational data, such as grounding resistance, branch current, and node potential, are collected in real time through a SCADA system. The data is transmitted to a central data platform via the MQTT protocol. The platform uses GPS coordinate matching to spatially correlate the electrical data with soil nodes in a knowledge graph, and employs graph-based reasoning algorithms to analyze corrosion status and assess corrosion levels. For example, electrical data is mapped to dynamic nodes in the graph, and a corrosion risk index is calculated through graph queries.

[0055] In one possible implementation, the operational electrical dataset of the grounding grid is acquired in real time, and then fused and evaluated with the soil corrosivity knowledge graph to generate a grounding grid corrosion level assessment result. Step S300 further includes step S310, constructing an electrical monitoring network to collect data from the grounding grid in real time and obtain the operational electrical dataset. Specifically, high-precision sensors are installed at the branch points and grounding electrodes of the grounding grid to form a monitoring network and collect electrical data. For example, Hall effect sensors are used to measure branch currents, and the data is aggregated to a gateway via a ZigBee network. The gateway then sends the data to a central data platform using the Modbus protocol.

[0056] Step S320 involves mapping the operational electrical dataset to the soil corrosivity knowledge graph for expansion, identifying multiple soil monitoring points. Specifically, based on geographic location information, the coordinates of electrical sensors are matched with the coordinates of soil sensors. For example, a GIS tool is used for spatial querying to find the soil monitoring point closest to the electrical sensor, and a "co-location" relationship is created in the knowledge graph.

[0057] Step S330: Spatially co-locate the parameters of the multiple entity instances with the multiple soil monitoring points to obtain dynamic electrical attribute nodes. Specifically, in the graph, a dynamic node is created for each electrical data point and connected to the corresponding soil entity. For example, if electrical sensor S1 is co-located with soil point A, then an "Electrical node S1" is created in the graph and an "Located at" edge is added pointing to "Soil point A".

[0058] Step S340: Based on the association between the dynamic electrical attribute nodes and soil corrosion factors, a fusion evaluation map is constructed for feature learning, and a comprehensive feature vector is extracted. Specifically, the dynamic nodes are associated with soil corrosion factors to construct a fusion evaluation map, which is a graph structure combining soil and electrical data. A graph neural network (GNN) model is used for feature learning. Multiple layers of graph convolution aggregate neighbor node information to generate an embedding vector for each node. Pooling layers are used to convert the node vectors into a graph-level comprehensive feature vector to represent the corrosion state.

[0059] Step S350: Based on the comprehensive feature vector, corrosion level classification is performed to generate the grounding grid corrosion level evaluation result. Specifically, a machine learning classifier is used to classify the corrosion level of the extracted comprehensive feature vector. First, a labeled training dataset is constructed, associating the comprehensive feature vector corresponding to historical grounding grid corrosion with the actual corrosion level, such as no corrosion, light corrosion, moderate corrosion, heavy corrosion, and severe corrosion. Then, a multi-classification model, such as Support Vector Machine (SVM), Random Forest, or Deep Neural Network, is trained. In practical applications, the real-time extracted comprehensive feature vector is input into the trained classification model, and the corresponding corrosion level probability distribution is output. The level with the highest probability is taken as the evaluation result. For example, a three-layer fully connected neural network can be constructed using PyTorch, with the input layer dimension consistent with the comprehensive feature vector dimension, and the output layer having 5 dimensions corresponding to 5 corrosion levels. The probability of each level is output through the Softmax function.

[0060] In one possible implementation, real-time acquisition of the grounding grid's operational electrical dataset is achieved. Step S310 further includes step S311, which involves deploying current and voltage sensors across accessible nodes in the distributed monitoring network to construct an electrical parameter monitoring network. Specifically, in the distributed monitoring network of the grounding grid, key electrically accessible nodes, such as grounding electrode leads and branch connections, are selected to deploy high-precision current and voltage sensors. The current sensors utilize the Hall effect principle to measure branch currents in real time; the voltage sensors employ differential probes to monitor the node's potential to ground. The sensor nodes are networked using ZigBee or LoRa wireless communication protocols to form an electrical parameter monitoring network covering the entire grounding grid.

[0061] Step S312 involves synchronously sampling the distributed monitoring network within the same time window using the electrical parameter monitoring network to obtain the original electrical dataset. Specifically, a unified time window is set, such as one second as a sampling cycle, and a central clock synchronization mechanism ensures that all electrical sensors collect data at the same time. The collected parameters include branch current values, node voltage values, instantaneous grounding resistance values, etc. The original electrical dataset is stored in time-series format, containing timestamps, sensor numbers, measurement parameters, and values, forming the initial electrical monitoring data set.

[0062] Step S313 involves baseline correction of the original electrical dataset to obtain the grounding grid operation electrical dataset. Specifically, the original electrical dataset undergoes data preprocessing, including baseline correction and noise filtering. Specifically, a sliding window averaging method is used to perform baseline correction on the current and voltage data to eliminate systematic errors caused by temperature drift or sensor zero-point offset; wavelet transform is used to remove high-frequency noise interference. The corrected data forms a structured grounding grid operation electrical dataset, stored in tabular or time-series database format.

[0063] In one possible implementation, based on the association between the dynamic electrical attribute nodes and soil corrosion factors, a fusion evaluation map is constructed for feature learning, and a comprehensive feature vector is extracted. Step S340 further includes step S341, which involves extracting the soil corrosion factors based on the soil corrosion knowledge graph, associating the dynamic electrical attribute nodes with the soil corrosion factors, and constructing the fusion evaluation map. Specifically, soil corrosion factors are soil environmental factors that cause or accelerate grounding grid corrosion, such as chloride ion content, pH value, and resistivity. All soil corrosion factor nodes and their attributes, including factor type, influence weight, and spatial location, are extracted from the soil corrosion knowledge graph. Using a graph query language, such as Cypher, pattern matching queries are performed to associate the dynamic electrical attribute nodes with spatially similar soil corrosion factor nodes. The distance between nodes is calculated based on GIS spatial analysis, and when the distance is less than a set threshold, a "co-location association" is established between them. Finally, a fusion evaluation map containing soil corrosion factor nodes, dynamic electrical attribute nodes, and their association relationships is formed.

[0064] Step S342 involves using a graph neural network to perform multi-layer stacked graph convolutions on the fused evaluation graph to obtain semantic information for multiple nodes. Specifically, a multi-layer graph convolutional network is constructed using a graph neural network (GNN) framework. Each graph convolutional operation updates the current node representation by aggregating neighbor node information. The implementation includes defining the number of graph convolutional layers, setting the hidden layer dimensions, and selecting an aggregation function, such as the GraphSAGE algorithm. Node semantic information is a semantically rich feature representation of a node obtained after learning through the GNN.

[0065] Step S343 involves aggregating the semantic information of the multiple nodes to generate multiple node features. Specifically, feature aggregation is performed on the semantic information of the nodes obtained after multi-layer graph convolution. A multi-head attention mechanism is used to calculate the importance weights between nodes, and then a weighted sum is performed. The specific implementation includes: setting the number of attention heads, calculating attention scores, applying a softmax function for normalization to obtain attention weights, and finally aggregating neighbor information based on the weights. The final numerical representation of a node after feature learning and aggregation is the node feature.

[0066] Step S344: Perform graph-level feature transformation on the fused evaluation graph according to the multiple node features to construct the comprehensive feature vector. Specifically, graph-level feature transformation refers to the process of compressing the features of the entire graph structure into a fixed-length vector. The comprehensive feature vector is a numerical vector representing the global information of the entire fused evaluation graph. A hierarchical pooling strategy is adopted to convert the node features of the entire graph into a fixed-length graph-level feature vector. First, SortPooling is used to sort the node features by importance, retaining the top k most important node features. Then, a 1D convolutional layer is applied to extract features from the sorted feature sequence. Finally, global max pooling is used to generate a fixed-dimensional comprehensive feature vector. Alternatively, graph pooling layers in graph isomorphic networks can be used to calculate node importance scores through a learnable projection matrix and select representative node features for concatenation. The final output comprehensive feature vector is set to 128-256 dimensions to ensure that it can fully express the global information of the graph while avoiding the curse of dimensionality.

[0067] In one possible implementation, the dynamic electrical attribute nodes are associated with the soil corrosion factors to construct a fusion evaluation map. Step S341 further includes step S3411, which involves aligning and resolving the dynamic electrical attribute nodes with the soil corrosion factors. Based on the resolving results, causal analysis is performed on the dynamic electrical attribute nodes to obtain the grounding resistance value, branch current distribution non-uniformity, and node potential difference variation coefficient of the dynamic electrical attribute nodes. Specifically, the dynamic electrical attribute nodes are associated and resolved with the soil corrosion factors based on spatial location matching and temporal alignment. Spatial connectivity algorithms, such as KD-tree nearest neighbor search, are used to find all soil corrosion factors within a radius smaller than a set threshold for each electrical monitoring point. Then, causal analysis is performed on the dynamic electrical attribute nodes, and three key indicators are calculated in real time using electrical parameters: the grounding resistance value is obtained directly from the measurement data; the branch current distribution non-uniformity is obtained by calculating the ratio of the standard deviation to the average value of each branch current value; and the node potential difference variation coefficient is obtained by analyzing the ratio of the standard deviation to the average value of the potential difference between different nodes. These calculations are performed in real time using a Python numerical computing library deployed on an edge computing device. Among them, the grounding resistance value is the resistance between the grounding device and the distant earth, reflecting the conductivity of the grounding grid. Branch current distribution non-uniformity is an indicator characterizing the degree of balance in the current distribution across the branches of the grounding grid. The coefficient of variation of node potential difference is a statistical measure reflecting the uniformity of potential distribution at each node of the grounding grid.

[0068] Step S3412: Based on the analysis results, a causal analysis is performed on the soil corrosion factors to obtain soil strong corrosion factors, soil electrochemical corrosion factors, and soil redox factors. Specifically, based on the soil corrosion mechanism and historical data statistics, a causal classification analysis is performed on the soil corrosion factors. The influence degree of each corrosion factor is evaluated using a random forest feature importance analysis algorithm, and the factors with the top 20% importance scores are marked as soil strong corrosion factors. According to the principle of electrochemical corrosion, factors affecting electrode potential, such as chloride ion content, salinity, and humidity, are classified as soil electrochemical corrosion factors. Based on the principle of redox reaction, factors affecting electron transfer, such as redox potential, sulfide content, and microbial activity, are classified as soil redox factors.

[0069] Step S3413: Based on preset anomaly threshold judgment rules, which include a first threshold, a second threshold, and a third threshold. Specifically, anomaly threshold judgment rules based on industry standards and historical statistics are established to identify abnormal electrical parameter states. The first threshold is the anomaly judgment threshold for grounding resistance value, determined based on grounding device characteristic parameter measurement guidelines, for example, set to 5Ω. The second threshold is the anomaly judgment threshold for branch current distribution non-uniformity, determined based on the 95th percentile of historical normal operation data, for example, set to 0.35. The third threshold is the anomaly judgment threshold for node potential difference variation coefficient, determined based on Monte Carlo simulation analysis, for example, set to 0.25. These thresholds are stored in configuration files and can be dynamically adjusted according to actual field conditions.

[0070] Step S3414: When the grounding resistance value exceeds the first threshold, the location coordinates of the grounding resistance value are correlated with the soil corrosion factor to establish a high resistance-strong corrosion correlation. Specifically, when real-time monitoring shows that the grounding resistance value exceeds the first threshold, the geographic information system positioning function is activated. The geographic coordinates of the abnormal monitoring point are obtained using the WGS84 global coordinate system. Using this coordinate point as the center, all soil factors marked as highly corrosive are automatically searched within a circular area of ​​a preset radius. A high resistance-strong corrosion correlation is established between the qualified electrical nodes and soil factor nodes.

[0071] Step S3415: When the unevenness of the branch current distribution exceeds the second threshold, the location coordinates of the branch current distribution nodes are correlated with the soil electrochemical corrosion factor to establish a current anomaly-electrochemical corrosion correlation. Specifically, when the unevenness of the branch current distribution exceeds the second threshold, all nodes of the abnormal branch are located, and the soil electrochemical corrosion factor is matched within a preset range around the node coordinates to establish a current anomaly-electrochemical corrosion correlation connection, and metadata such as the correlation timestamp and correlation strength are recorded.

[0072] Step S3416: When the coefficient of variation of the node potential difference exceeds the third threshold, the position coordinates of the coefficient of variation of the node potential difference are correlated with the soil redox factor to establish a potential unevenness-redox correlation. Specifically, when the coefficient of variation of the node potential difference exceeds the third threshold, all nodes in the area of ​​abnormal potential distribution are identified, the influence range is determined based on Voronoi diagram analysis, and soil redox factors are searched within the range. The relationship connection of potential unevenness-redox correlation is established, and the persistence index of the correlation is calculated through time series analysis to evaluate the stability of the correlation.

[0073] Step S3417: Based on the high resistance-strong corrosion correlation, the current anomaly-electrochemical corrosion correlation, and the potential non-uniformity-redox correlation, the fusion evaluation graph is constructed. Specifically, the three types of correlations established in the preceding steps are integrated into a unified graph structure to construct a complete fusion evaluation graph. Transaction processing in the graph database is used to ensure data consistency, and query performance is optimized by creating composite indexes. The graph contains a multi-layered structure: the bottom layer consists of original monitoring data nodes, the middle layer contains various correlations, and the top layer contains the fusion evaluation results. Graph structure optimization is performed regularly, including updating relation weights, cleaning up redundant relations, and community detection analysis, to ensure the accuracy and timeliness of the graph. The final constructed fusion evaluation graph supports graph traversal queries and real-time inference analysis.

[0074] Step S400: Based on the corrosion level assessment results of the grounding grid, corrosion prediction is performed on the grounding grid. The corrosion prediction trend map is then backtracked to the soil corrosivity knowledge graph for verification. Based on the verification results, the operation of the grounding grid is analyzed, and a grounding grid remaining life prediction report is generated.

[0075] Specifically, time series prediction models, such as LSTM, are used to generate trend charts based on historical corrosion level data, showing how the corrosion level changes over time. These trends are compared to corrosion development trajectories in a knowledge graph, and the Dynamic Time Warping (DTW) algorithm is used to calculate similarity. If the similarity is low, data quality is checked and the model is retrained; otherwise, the remaining lifetime is analyzed based on the trend. For example, linear regression is used with Python's scikit-learn to calculate the lifetime, generating a remaining lifetime prediction report for the grounding grid. This report estimates the safe service life of the grounding grid.

[0076] In one possible implementation, corrosion prediction of the grounding grid is performed based on the corrosion level assessment results. A corrosion prediction trend map is obtained and backtracked to the soil corrosivity knowledge graph for verification. Based on the verification results, operational analysis of the grounding grid is performed, and a grounding grid remaining life prediction report is generated. Step S400 further includes step S410: constructing a multivariate time series dataset; performing time series learning based on the multivariate time series dataset to obtain time series regularity parameters of the corrosion process. Specifically, soil parameters, electrical parameters, and corrosion level assessment results are integrated to construct a time series dataset containing multiple variables. A time series database is used to store monitoring data aligned to timestamps. Each time point includes variables such as soil resistivity, pH value, salinity, grounding resistance, branch current, and corrosion level. An LSTM or Transformer time series model is used for deep learning, and the time series regularity of the corrosion process is extracted through an encoder-decoder structure. During training, a sliding window technique is used to generate training samples, and network parameters are optimized through backpropagation. The final output includes time series regularity parameters such as corrosion rate change trends, periodic fluctuation characteristics, and abrupt change probability.

[0077] Step S420: Based on the grounding grid corrosion level assessment results, corrosion prediction is performed according to the corrosion process time-series parameters, and a corrosion prediction trend chart is constructed. Specifically, based on the time-series parameters obtained in step S410, a time series prediction algorithm is used to predict the future corrosion level of the grounding grid. An autoregressive integrated moving average (ARIMA) model or the Prophet algorithm is used, combined with historical data from the corrosion level assessment results, to predict the corrosion level change trends for the next 30 days, 90 days, and 180 days. The prediction results are used to generate a corrosion prediction trend chart through a data visualization component. The chart includes historical data curves, prediction curves, confidence intervals, etc., with the horizontal axis representing time and the vertical axis representing corrosion level.

[0078] Step S430: Based on the soil corrosivity knowledge graph, perform a full-cycle corrosion analysis to extract the corrosion development trajectory. Specifically, query the complete development path of historical corrosion cases from the soil corrosivity knowledge graph to extract the full-cycle corrosion trajectory. Use graph traversal algorithms, such as depth-first search, to find all possible paths from the initial corrosion state to the final corrosion result in the knowledge graph, and analyze the key characteristics and transformation conditions of different corrosion stages. The extracted corrosion development trajectory includes the corrosion level change sequence in the time dimension and the corrosion area expansion pattern in the spatial dimension, forming a standardized corrosion development reference template.

[0079] Step S440: Perform similarity matching between the corrosion prediction trend map and the corrosion development trajectory to generate trend similarity. Specifically, the Dynamic Time Warping (DTW) algorithm is used to calculate the similarity between the corrosion prediction trend and the historical corrosion development trajectory. The prediction trend map and corrosion development trajectory are converted into time series data in a unified format. The DTW algorithm is used to calculate the minimum alignment distance between the two sequences, and the distance value is converted into a similarity score in the range of 0-1 to obtain the trend similarity.

[0080] Step S450: When the trend similarity is less than a preset similarity threshold, the verification fails. Abnormal deviation segments in the corrosion prediction trend map are identified for data backtracking verification. The corrosion prediction trend map is updated based on the verification results until the trend similarity is greater than or equal to the preset similarity threshold. Specifically, the preset similarity threshold is the minimum similarity standard for judging whether the prediction result is acceptable. When the trend similarity is lower than this threshold, the time segment in the corrosion prediction trend map with the greatest difference from the historical corrosion development trajectory is identified. Data quality backtracking verification is performed on the original monitoring data corresponding to this segment to check whether the sensor data is abnormal, whether the data transmission is complete, and whether the data preprocessing is correct. After discovering data problems, data cleaning and correction are performed, and time-series learning and prediction calculations are re-executed to generate an updated corrosion prediction trend map. The verification process is repeated until the trend similarity reaches the threshold requirement.

[0081] Step S460: When the trend similarity is greater than or equal to the preset similarity threshold, the verification is successful. Based on the corrosion prediction trend map, the grounding grid is analyzed operationally, and a remaining life prediction report for the grounding grid is generated. Specifically, when the trend similarity reaches or exceeds the preset similarity threshold, the prediction result is verified. Based on the verified corrosion prediction trend map, combined with factors such as grounding grid material characteristics, operating load, and environmental conditions, the corrosion development rate is analyzed. Using reliability theory, the remaining life distribution at different confidence levels is calculated, and a complete prediction report containing remaining life estimates, reliability indicators, and maintenance recommendations is finally generated.

[0082] In one possible implementation, corrosion prediction is performed based on the corrosion level assessment results of the grounding grid according to the time-series parameters of the corrosion process, and a corrosion prediction trend map is constructed. Step S420 further includes step S421, constructing a multivariate time series dataset, and using a long short-term memory network model to learn the time-series pattern of the corrosion process according to the multivariate time series dataset to determine the time-series parameters of the corrosion process. Specifically, historical corrosion level data, soil parameter change sequences, and electrical characteristic data are integrated to construct a multivariate time series dataset. The historical corrosion level data originates from the corrosion level assessment result sequence generated in step S350; the soil parameter change sequence includes the time-series changes of parameters such as soil resistivity, pH value, and salinity; the electrical characteristic data includes the time-series records of electrical parameters such as grounding resistance, branch current distribution unevenness, and node potential difference variation coefficient. A time-series database is used to align and store the multi-source data to ensure timestamp consistency. A long short-term memory network model is used to train a multivariate time series dataset. By inputting historical time window data, the corrosion level of the next time step is predicted. The model learns the temporal regularity parameters of the corrosion process, including key parameters such as corrosion rate change patterns, periodic fluctuation characteristics, and abrupt change triggering conditions.

[0083] Step S422 involves predicting the corrosion level of the grounding grid based on the corrosion process time-series parameters, obtaining corrosion level change data at multiple time points. Specifically, based on the corrosion process time-series parameters learned in step S421, a trained LSTM model is used to perform multi-step predictions of the corrosion level of the grounding grid in future time periods. Inputting the current and historical corrosion level assessment results, soil parameters, and electrical characteristic data, the model outputs predicted corrosion level values ​​for multiple time points, such as the next 30 days, 90 days, and 180 days, forming a corrosion level change data sequence. An autoregressive strategy is employed during the prediction process, using the prediction result from the previous moment as the input for the next moment to ensure the continuity of the prediction sequence. Simultaneously, Monte Carlo sampling is used to calculate the confidence interval of the prediction results to assess the prediction uncertainty.

[0084] Step S423 involves visually labeling the corrosion level changes at multiple time points along the time dimension to construct a corrosion prediction trend chart. Specifically, the corrosion level change data sequence obtained in step S422 is visualized on a time axis to construct a corrosion prediction trend chart. The horizontal axis represents time, and the vertical axis represents corrosion level. Historical corrosion level curves and future prediction curves are plotted, and confidence intervals are marked to show the reliability of the predictions. Key turning points are identified and marked in the trend chart, including inflection points where the corrosion rate significantly accelerates and abrupt changes in corrosion level. Simultaneously, the expected time points for reaching the next corrosion level are calculated and marked based on the prediction curves, such as the expected time for upgrading from "moderate corrosion" to "severe corrosion." The trend chart uses color coding to distinguish different corrosion level stages and adds legends to form a corrosion prediction trend chart that intuitively reflects the corrosion development process.

[0085] In one possible implementation, when the trend similarity is greater than or equal to the preset similarity threshold, the verification is successful. Based on the corrosion prediction trend map, the grounding grid is analyzed to generate a remaining life prediction report for the grounding grid. Step S460 further includes step S461: performing trend calculation on the verified corrosion prediction trend map to obtain the trend slope; identifying trend inflection points based on the trend slope; and evaluating trend stability to obtain trend fluctuation characteristic parameters. Specifically, linear or nonlinear regression analysis is performed on the predicted data sequence in the corrosion prediction trend map to calculate its trend slope, which reflects the average rate of corrosion development. The first or second derivative of the trend sequence is analyzed to find trend inflection points. An inflection point is the moment when the trend direction or rate changes significantly, for example, the turning point where the corrosion rate changes from gradual to a sharp increase, marking the beginning of a new stage of corrosion. The statistical characteristics of the trend sequence are calculated to obtain trend fluctuation characteristic parameters, including standard deviation, coefficient of variation, autocorrelation, and parameters based on inflection point analysis. Standard deviation measures the magnitude of fluctuation in the predicted corrosion level around the trend line; the greater the fluctuation, the worse the stability. The coefficient of variation is the ratio of the standard deviation to the mean, used to compare the degree of fluctuation at different levels. Autocorrelation refers to the correlation between a series at different time lags, used to determine the persistence or periodicity of a trend. Parameters based on inflection point analysis include the frequency of inflection points and the change in slope at the inflection point.

[0086] Step S462: Based on the trend fluctuation characteristic parameters and the operating environment, corrosion development analysis is performed to obtain the corrosion acceleration stage. Specifically, the trend fluctuation characteristic parameters calculated in step S461 are combined with the operating environment data of the grounding grid. The operating environment data includes soil parameters and electrical operating parameters obtained from the knowledge graph and real-time monitoring. Based on the above comprehensive analysis, the corrosion acceleration stage is determined and classified. For example, when the trend slope is consistently higher than a certain threshold, and the environmental data indicates that it is a rainy season with high humidity and high salinity, the rainy season period can be determined as the corrosion acceleration stage; when an upward trend inflection point is detected, and the soil redox potential decreases abnormally during the same period, it indicates the aggravation of microbial corrosion, thus identifying a corrosion acceleration stage.

[0087] Step S463: Retrieve the peak operating phase of the grounding grid and evaluate the overlap between the corrosion acceleration phase and the peak operating phase to obtain the time overlap. Specifically, obtain planned or historically typical peak operating phases within the prediction period from grid dispatch or operation records. These phases correspond to periods of high load and high fault current probability, placing the highest demands on the current conduction and thermal stability of the grounding grid. Compare the corrosion acceleration phase identified in step S462 with the peak operating phase on the time axis to calculate the time overlap. The time overlap is an indicator that quantifies the degree of overlap between two time periods. It can be calculated as the proportion of the intersection length of the corrosion acceleration phase and the peak operating phase to the total length of the corrosion acceleration phase, or the proportion of the intersection length to the total analysis time. The higher the overlap, the more significant the risk superposition effect.

[0088] Step S464: Based on the time overlap, perform a performance degradation analysis on the grounding grid to determine the corrosion risk level of the grounding grid. Specifically, based on the corrosion prediction trend, the severity of the corrosion acceleration phase, and the time overlap, comprehensively analyze the potential performance degradation of the grounding grid in the future. For example, a corrosion trend with a gentle slope and no overlap with the peak operating period indicates a low risk of performance degradation; while a corrosion trend with a steep slope and complete overlap with the peak operating period indicates that the grounding grid performance may decline sharply during the system's most vulnerable period, posing an extremely high risk. According to a preset risk matrix or scoring rules, the above analysis results are mapped to a comprehensive grounding grid corrosion risk level. For example, it can be defined as "low risk," "medium risk," "high risk," or "extremely high risk."

[0089] Step S465: Based on the corrosion risk level of the grounding grid and the characteristics of the grounding grid conductor material, the service life is calculated, and a remaining service life prediction report of the grounding grid is generated. Specifically, the corrosion risk level of the grounding grid determined in step S464 is used as a correction factor for the service life calculation. A high risk level indicates that a more conservative service life estimation model or a shorter safety margin is required. The characteristics of the grounding grid conductor material are the physical basis for calculating the remaining service life, such as empirical data on corrosion rates, cross-sectional areas, and tensile strength degradation laws of steel and copper. Different materials have their inherent degradation models under different corrosion environments. Within the safety margin, the time interval from the current point in time to the predicted corrosion level reaching a critical dangerous state is calculated as the basic remaining service life. The basic remaining service life is corrected according to the risk level. For example, for the "extremely high risk" case, the basic remaining service life is multiplied by a discount factor less than 1. Combining all the above analysis, calculation, and evaluation results, a remaining service life prediction report of the grounding grid is compiled. The report includes: current corrosion status assessment, corrosion prediction trend analysis, identified risk factors, determined corrosion risk level, calculated remaining service life, and corresponding maintenance or replacement recommendations. This application employs various technical means, including deploying multi-parameter soil sensors to collect and generate soil parameter datasets in real time, constructing a soil corrosion knowledge graph by combining soil corrosion mechanism data, obtaining an electrical dataset of the grounding grid operation, fusing it with the knowledge graph to assess the corrosion level, predicting corrosion based on the assessment results, performing operational analysis after backtracking verification using trend graphs, and generating a remaining life prediction report. These methods solve the technical problem of insufficient accuracy in existing grounding grid corrosion prediction and achieve the technical effect of improving the accuracy of corrosion state prediction.

Claims

1. A grounding grid corrosion prediction method fused with soil parameter knowledge graph, characterized in that, The method comprises: Based on the grounding net area layout multi-parameter soil sensor for real-time collection, generate soil parameter data set; Based on the soil parameter data set combined with soil corrosion mechanism data for knowledge association, build soil corrosion knowledge graph; Real-time acquisition of the running electrical data set of the grounding net, fusion evaluation of the running electrical data set and the soil corrosion knowledge graph, generation of the grounding net corrosion grade evaluation result; According to the grounding net corrosion grade evaluation result, corrosion prediction of the grounding net is carried out, and the corrosion prediction trend graph is obtained. Backtracking to the soil corrosion knowledge graph for verification, according to the verification result, the running analysis of the grounding net is carried out, and the remaining life prediction report of the grounding net is generated.

2. The ground grid corrosion prediction method of fusing soil parameter knowledge graph according to claim 1, wherein, Based on the grounding net area layout multi-parameter soil sensor for real-time collection, generate soil parameter data set, the method comprising: Traverse the soil distribution analysis of the grounding net area, generate soil distribution data; According to the soil distribution data, carry out grid processing, and divide the grid point data; Deploy the multi-parameter soil sensor according to the grid point data to the grounding net area to obtain a distributed monitoring network; Set a preset period, activate the multi-parameter soil sensor according to the preset period combined with the distributed monitoring network for automatic collection, and generate soil original collection data; The soil original collection data is normalized to generate soil parameter data set. 3.The method of claim 2, wherein, Deploy the multi-parameter soil sensor according to the grid point data to the grounding net area to obtain a distributed monitoring network, the method comprising: Based on the grid point data, topological analysis is carried out on the grounding net area, the radiation area boundary is divided according to the grounding net topological structure, and the grounding net area is divided into several equal-area grid units based on the radiation area boundary; Traverse several equal-area grid units to deploy multi-parameter soil sensors according to grid point data to determine multiple sensor nodes; Based on the multiple sensor nodes, the geographical position is calibrated, and the distributed monitoring network is constructed.

4. The ground grid corrosion prediction method of fusing soil parameter knowledge graph according to claim 3, wherein, Based on the radiation area boundary, the grounding net area is divided into several equal-area grid units, the method comprising: Based on the radiation area boundary, the monitoring area of the grounding net area is determined; The square grid division method is adopted to divide the grounding net area into several grid units according to the monitoring area; Based on the several grid units, numbering is carried out to obtain several equal-area grid units.

5. The method of claim 2, wherein, The soil original collection data is normalized to generate soil parameter data set, including: Based on the soil original collection data, set multiple soil value ranges, filter the soil original collection data based on the multiple soil value ranges, and identify abnormal data points; Label the abnormal data points, repair the data according to the labeled data, and obtain the repaired data; The repaired data is normalized to construct the soil parameter data set.

6. The ground grid corrosion prediction method of fusing soil parameter knowledge graph according to claim 2, wherein, Based on the soil parameter data set combined with soil corrosion mechanism data for knowledge association, build soil corrosion knowledge graph, the method comprising: Based on the soil parameter data set, corrosion analysis is carried out, and multiple core entity types are defined; Traverse the multiple core entity types to divide the attributes, and determine the multiple core entity attributes; Defining relationships of a plurality of core entity types based on the plurality of core entity attributes, to generate a plurality of core entity relationship types; Integrating the plurality of core entity types and the plurality of core entity attributes according to the plurality of core entity relationship types to construct a knowledge graph mode layer; Retrieving soil corrosion mechanism data, and mapping the soil parameter data set to the knowledge graph mode layer according to the soil corrosion mechanism data to determine a plurality of entity instance parameters; Based on the plurality of entity instance parameters, corrosion analysis is performed, and the plurality of entity instance parameters are connected by setting corrosion rules to construct the soil corrosivity knowledge graph.

7. The ground grid corrosion prediction method of fusing soil parameter knowledge graph according to claim 6, wherein, Based on the soil parameter data set, corrosion analysis is performed, and a plurality of core entity types are defined, the method comprising: Based on the grid point data, a plurality of soil point entity instances are created for the multi-parameter soil sensor of the distributed monitoring network, and the plurality of soil point entity instances have a corresponding relationship with the multi-parameter soil sensor; Based on the plurality of soil point entity instances, corrosion analysis is performed on the soil parameter data set, and a plurality of corrosion factor entity instances are created, and the plurality of corrosion factor entity instances have a corresponding relationship with the plurality of soil point entity instances.

8. The ground grid corrosion prediction method of fusing soil parameter knowledge graph according to claim 6, wherein, Setting corrosion rules to connect the plurality of entity instance parameters to construct the soil corrosivity knowledge graph, the method comprising: Based on the corrosion rules, a plurality of entity instance data are connected to construct an initial knowledge graph; Introducing historical grounding corrosion case data for analysis to obtain a plurality of corrosion soil data; Mapping the plurality of corrosion soil data to the initial knowledge graph according to the historical grounding corrosion case data for alignment and merging to generate a merged knowledge graph; Based on the merged knowledge graph, frequent mining is performed to extract corrosion factor combination patterns, and the merged knowledge graph is updated according to the corrosion factor combination patterns to construct the soil corrosivity knowledge graph. 9.The grounding grid corrosion prediction method of fusing soil parameter knowledge graph according to claim 6, wherein, Real-time acquisition of the running electrical data set of the grounding grid, fusion evaluation of the running electrical data set and the soil corrosivity knowledge graph, and generation of a grounding grid corrosion grade evaluation result, the method comprising: Constructing an electrical monitoring network to collect the grounding grid in real time to obtain the running electrical data set; Mapping the running electrical data set to the soil corrosivity knowledge graph for expansion to determine a plurality of soil monitoring points; Spatially co-locating the plurality of entity instance parameters and the plurality of soil monitoring points to obtain dynamic electrical property nodes; Based on the dynamic electrical property nodes and soil corrosion factors, a fusion evaluation graph is constructed for feature learning to extract a comprehensive feature vector; Based on the comprehensive feature vector, corrosion grade classification is performed to generate the grounding grid corrosion grade evaluation result.

10. The method of claim 9, wherein, Real-time acquisition of the running electrical data set of the grounding grid, comprising: Deploying current sensors and voltage sensors on accessible nodes of the distributed monitoring network to construct an electrical parameter monitoring network; Synchronously sampling the distributed monitoring network within the same time window through the electrical parameter monitoring network to obtain an original electrical data set; Performing baseline correction processing on the original electrical data set to obtain the running electrical data set of the grounding grid.

11. The ground grid corrosion prediction method of fusing soil parameter knowledge graph according to claim 9, wherein, Based on the dynamic electrical property node and the soil corrosion factor, a fusion evaluation graph is constructed for feature learning, and a comprehensive feature vector is extracted, the method comprising: Based on the soil corrosion knowledge graph, the soil corrosion factors are extracted, and the dynamic electrical property node is associated with the soil corrosion factors to construct a fusion evaluation graph; A graph neural network is used to perform multi-layer stack graph convolution on the fusion evaluation graph to obtain multiple node semantic information; The multiple node semantic information is aggregated to generate multiple node features; According to the multiple node features, the fusion evaluation graph is converted into a graph-level feature to construct the comprehensive feature vector.

12. The ground grid corrosion prediction method of fusing soil parameter knowledge graph according to claim 11, wherein, Based on the dynamic electrical property node and the soil corrosion factor, a fusion evaluation graph is constructed, the method comprising: Align and analyze the dynamic electrical property node and the soil corrosion factor, and perform causal analysis on the dynamic electrical property node based on the analysis results to obtain the grounding resistance value, branch current distribution unevenness, and node potential difference variation coefficient of the dynamic electrical property node; According to the analysis results, the soil corrosion factors are analyzed to obtain strong corrosion factors, soil electrochemical corrosion factors, and soil oxidation-reduction factors; Based on a preset abnormal threshold judgment rule, the abnormal threshold judgment rule includes a first threshold, a second threshold, and a third threshold; When the grounding resistance value exceeds the first threshold, the position coordinates of the grounding resistance value are associated with the strong corrosion factors to establish a high resistance-strong corrosion association; When the branch current distribution unevenness exceeds the second threshold, the position coordinates of the branch current distribution node are associated with the soil electrochemical corrosion factors to establish a current anomaly-electrochemical corrosion association; When the node potential difference variation coefficient exceeds the third threshold, the position coordinates of the node potential difference variation coefficient are associated with the soil oxidation-reduction factors to establish a potential unevenness-oxidation-reduction association; Based on the high resistance-strong corrosion association, the current anomaly-electrochemical corrosion association, and the potential unevenness-oxidation-reduction association, the fusion evaluation graph is constructed.

13. The ground grid corrosion prediction method of fusing soil parameter knowledge graph according to claim 1, wherein, According to the grounding grid corrosion grade evaluation result, the corrosion of the grounding grid is predicted to obtain a corrosion prediction trend graph, which is verified back to the soil corrosivity knowledge graph, and according to the verification result, the operation of the grounding grid is analyzed to generate a grounding grid remaining life prediction report, the method comprising: A multivariate time series data set is constructed, and time series learning is performed based on the multivariate time series data set to obtain corrosion process time series rule parameters; According to the grounding grid corrosion grade evaluation result, the corrosion prediction is performed according to the corrosion process time series rule parameters to construct a corrosion prediction trend graph; Based on the soil corrosivity knowledge graph, the corrosion development trajectory is extracted; The corrosion prediction trend graph and the corrosion development trajectory are similar matched to generate a trend similarity. When the trend similarity is less than the preset similarity threshold, the verification fails, an abnormal section deviating from the corrosion prediction trend graph is identified for data backtracking verification, the corrosion prediction trend graph is updated according to the verification result, and the verification stops until the trend similarity is greater than or equal to the preset similarity threshold; When the trend similarity is greater than or equal to the preset similarity threshold, the verification passes, the grounding grid is analyzed according to the corrosion prediction trend graph, and the grounding grid remaining life prediction report is generated.

14. The method of claim 13, wherein, According to the corrosion grade evaluation result of the grounding grid, corrosion prediction is performed according to the corrosion process time sequence rule parameter, and a corrosion prediction trend graph is constructed, including: A multivariate time series data set is constructed, a long short-term memory network model is used to learn the time sequence rule of the corrosion process according to the multivariate time series data set, and the corrosion process time sequence rule parameter is determined; The corrosion grade evaluation result of the grounding grid is predicted according to the corrosion process time sequence rule parameter, and corrosion grade change data at multiple time points are obtained; The corrosion grade change data at multiple time points are visually labeled according to the time dimension, and a corrosion prediction trend graph is constructed.

15. The ground grid corrosion prediction method of fusing soil parameter knowledge graph according to claim 13, wherein, When the trend similarity is greater than or equal to the preset similarity threshold, the verification passes, the grounding grid is analyzed according to the corrosion prediction trend graph, and the grounding grid remaining life prediction report is generated, and the method comprises: Trend calculation is performed on the corrosion prediction trend graph that passes the verification, a trend slope is obtained, trend stability evaluation is performed according to the trend slope, and a trend fluctuation characteristic parameter is obtained; Based on the trend fluctuation characteristic parameter, corrosion development analysis is performed in combination with the operating environment, and a corrosion acceleration stage is obtained; The power grid operating peak stage of the grounding grid is called, the corrosion acceleration stage is overlapped with the power grid operating peak stage, and a time overlap degree is obtained; Based on the time overlap degree, performance degradation analysis is performed on the grounding grid, a grounding grid corrosion risk level is determined, life calculation is performed based on the grounding grid corrosion risk level in combination with the grounding grid conductor material characteristics, and the grounding grid remaining life prediction report is generated.

Images