A substation intelligent maintenance method and system based on digital twinning

Abstract

A substation intelligent maintenance method and system based on digital twinning belong to the technical field of data processing, comprising obtaining operation parameters of multiple substations, establishing digital twinning models of each substation and serving as nodes to constitute a digital twinning network; when the operation parameters exceed a threshold value, the corresponding node is determined as a target node; based on the digital twinning network, an associated node and its associated digital twinning model are obtained; abnormal operation characteristics of the target node within an alarm period and associated operation characteristics of the associated node are extracted; the characteristics are input into a fault propagation analysis model based on a graph neural network, the model nodes correspond to the digital twinning network nodes, and the edge weights of the model are determined based on historical fault propagation probabilities; the model is used to determine a fault source position and a propagation path; based on the fault source position, the propagation path and the target digital twinning model, a maintenance strategy is generated and sent to a corresponding execution terminal. The present application realizes accurate tracing and intelligent maintenance of cross-substation faults.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, and in particular relates to a method and system for intelligent maintenance of substations based on digital twins. Background Technology

[0002] As a critical node in the power system, the operational reliability of substations directly affects the safety and stability of the entire power grid. With the continuous expansion of the power grid and its increasing structural complexity, faults in a single substation often propagate through electrical coupling to adjacent or even more distant substations, triggering cascading failures and causing widespread power outages. Therefore, how to quickly and accurately identify the fault source, trace the fault propagation path, and formulate targeted maintenance strategies based on this information has become a pressing technical problem to be solved in the field of power system operation and maintenance.

[0003] Currently, substation maintenance mainly relies on single-station fault diagnosis based on local monitoring data. This involves deploying an independent monitoring system in each substation, triggering alarms when equipment parameters exceed preset thresholds, and then having maintenance personnel conduct on-site troubleshooting and repairs based on the alarm information. Because the monitoring systems of each substation are independent and lack cross-station data fusion and collaborative analysis capabilities, when a fault propagates from one substation to others, multiple stations may issue alarms simultaneously. Maintenance personnel often struggle to accurately determine the initial fault source from multiple alarms, requiring significant time for manual troubleshooting and fault tracing, leading to maintenance delays and an expansion of the fault's impact. Summary of the Invention

[0004] This application provides a digital twin-based intelligent substation maintenance method and system to improve the above-mentioned problems.

[0005] To achieve the above objectives, this application adopts the following technical solution: The first aspect of this invention proposes a substation intelligent maintenance method based on digital twins, comprising: Based on the execution terminal, the operating parameters of multiple substations are obtained, and digital twin models of multiple substations are established based on the operating parameters. Each digital twin model is used as a node to form a digital twin network. When the operating parameters of any substation exceed the preset alarm threshold, the node corresponding to the substation will be identified as the target node. Based on the digital twin network, nodes whose electrical distance from the target node is less than a preset distance threshold are obtained as associated nodes, or nodes that have the same fault root cause as the target node in the historical fault record are obtained as associated nodes, and the digital twin model corresponding to the associated node is obtained as the associated digital twin model. Obtain the abnormal operation characteristics of the target digital twin model corresponding to the target node during the alarm period, as well as the associated operation characteristics of the related digital twin model during the alarm period; Abnormal operation characteristics and associated operation characteristics are input into a preset fault propagation analysis model, which is constructed based on a graph neural network model, and the network nodes of the graph neural network model correspond to the nodes in the digital twin network. The edge weights of the graph neural network are determined based on the historical fault propagation probability between substations corresponding to each node. Based on the fault propagation analysis model, the location of the fault source corresponding to the abnormal operation characteristics and the propagation path of the fault source location to other nodes affected by the digital twin network are determined. Based on the fault source location, propagation path, and target digital twin model, a maintenance strategy is determined and sent to the execution terminal corresponding to the fault source location.

[0006] In conjunction with the first aspect, in some implementations, operating parameters corresponding to multiple substations are obtained based on the execution terminal, digital twin models corresponding to multiple substations are established based on the operating parameters, and each digital twin model is used as a node to form a digital twin network, including: Based on the execution terminal, electrical parameters, environmental parameters and equipment status parameters corresponding to multiple substations are obtained, and the electrical parameters, environmental parameters and equipment status parameters are aligned according to timestamps to form a time-series operation dataset corresponding to each substation. Building device models based on time-series runtime datasets; Based on the physical connection relationships between the equipment in each substation, the equipment models are merged to form a digital twin model of the substation.

[0007] In conjunction with the first aspect, in some implementations, obtaining the abnormal operational characteristics of the target digital twin model corresponding to the target node during the alarm period, and the associated operational characteristics of the associated digital twin model during the alarm period, includes: Based on the target digital twin model, the time-series data of the device model's operating parameters during the alarm period are obtained to form the target runtime sequence set; Abnormal operating characteristics are determined based on the target runtime sequence set. These abnormal operating characteristics include electrical quantity mutation characteristics, non-electrical quantity limit exceeding characteristics, and timing correlation deviation characteristics between devices. Based on the associated digital twin model, the time series data of the device model's operating parameters within the alarm period are determined, forming an associated runtime sequence set; The associated operational characteristics are determined based on the associated runtime sequence set. These associated operational characteristics include electrical quantity fluctuation characteristics, non-electrical quantity change rate characteristics, and inter-device timing synchronization characteristics.

[0008] In conjunction with the first aspect, in some implementations, abnormal operating characteristics and associated operating characteristics are input into a preset fault propagation analysis model, wherein the fault propagation analysis model is based on a graph neural network model, and the network nodes of the graph neural network model correspond to the nodes in the digital twin network, including: Align the target runtime sequence set with the associated runtime sequence set, obtain the runtime parameter difference values ​​between the target digital twin model and each associated digital twin model at the same time, and generate a difference feature matrix based on the runtime parameter difference values; Abnormal operation characteristics, associated operation characteristics, and difference characteristic matrices are used together as input features for the fault propagation analysis model.

[0009] In conjunction with the first aspect, in some implementations, the difference feature matrix participates in message passing as the edge features of the corresponding edges of the graph neural network within the fault propagation analysis model.

[0010] In conjunction with the first aspect, in some implementations, during message transmission, for a directed edge from the source node to the target node, the fault propagation analysis model reads the difference feature vector in the difference feature matrix corresponding to the edge, concatenates or multiplies the difference feature vector with the feature vector of the source node, and then multiplies it by the edge weight of the edge to generate a message to be transmitted to the target node.

[0011] In conjunction with the first aspect, in some implementations, the edge weights of the graph neural network are determined based on the historical fault propagation probability between substations corresponding to each node, including: Obtain historical fault records between substations within a preset historical time period. The historical fault records include the fault source substation, the substations through which the fault propagates, and the substations ultimately affected by the fault. Obtain the first and second substations corresponding to any two nodes in the digital twin network, the number of times the fault propagated from the first substation to the second substation in the historical fault records, and the total number of faults in the historical fault records where the first substation is the fault source. Based on the ratio of the number of fault propagation attempts to the total number of faults, the historical fault propagation probability from the first substation to the second substation is obtained. The edge weights of the graph neural network are determined based on the historical fault propagation probability between each node pair.

[0012] In conjunction with the first aspect, in some implementations, the location of the fault source corresponding to the abnormal operating characteristics and the propagation path of the fault source location through the digital twin network are determined based on the fault propagation analysis model, including: Based on edge weights, message passing and feature aggregation are performed between each node to output a corresponding fault probability vector. The target node and each associated node are sorted from high to low according to the fault probability vector, and the node at the top of the sort is selected as the fault source location. Starting from the location of the fault source, the transmission path to other nodes is determined by tracing back along the edge weights in the graph neural network model. A propagation path is defined as a path in which the edge weights are continuous and greater than a preset propagation threshold. The propagation path includes a fault source node, at least one intermediate propagation node, and an end node.

[0013] In conjunction with the first aspect, in some implementations, a maintenance strategy is determined based on the fault source location, propagation path, and target digital twin model, and the maintenance strategy is sent to the execution terminal corresponding to the fault source location, including: Based on the location of the fault source, determine the equipment model corresponding to the location of the fault source from the target digital twin model, and obtain the equipment type, equipment identifier and current operating status of the equipment model; Based on the propagation path, determine the set of nodes affected by the location of the fault source, and obtain the affected devices and the degree of impact for each node in the set of affected nodes; The fault type and severity of the fault source location are obtained. Based on the degree of impact and the severity of the fault, at least one candidate maintenance plan is generated through a preset maintenance strategy generation model. The candidate maintenance plan includes the target equipment to be maintained, the maintenance operation sequence, the required spare parts and components, and the estimated maintenance time.

[0014] The second aspect of this invention proposes a digital twin-based intelligent substation maintenance system, comprising a control terminal and execution terminals corresponding one-to-one with multiple substations, configured as follows: Based on the execution terminal, the operating parameters of multiple substations are obtained, and digital twin models of multiple substations are established based on the operating parameters. Each digital twin model is used as a node to form a digital twin network. When the operating parameters of any substation exceed the preset alarm threshold, the node corresponding to the substation will be identified as the target node. Based on the digital twin network, nodes whose electrical distance from the target node is less than a preset distance threshold are obtained as associated nodes, or nodes that have the same fault root cause as the target node in the historical fault record are obtained as associated nodes, and the digital twin model corresponding to the associated node is obtained as the associated digital twin model. Obtain the abnormal operation characteristics of the target digital twin model corresponding to the target node during the alarm period, as well as the associated operation characteristics of the related digital twin model during the alarm period; Abnormal operation characteristics and associated operation characteristics are input into a preset fault propagation analysis model, wherein the fault propagation analysis model is based on a graph neural network model, and the network nodes of the graph neural network model correspond to the nodes in the digital twin network. The edge weights of the graph neural network are determined based on the historical fault propagation probability between substations corresponding to each node. Based on the fault propagation analysis model, the location of the fault source corresponding to the abnormal operation characteristics and the propagation path of the fault source location to other nodes affected by the digital twin network are determined. Based on the fault source location, propagation path, and target digital twin model, a maintenance strategy is determined and sent to the execution terminal corresponding to the fault source location.

[0015] A third aspect of the present invention provides an electronic device comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method proposed in the first aspect of the present invention.

[0016] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in the first aspect of the present invention.

[0017] In summary, the above method and apparatus have the following technical effects: This invention discloses a substation intelligent maintenance method and system based on digital twins. The method includes: acquiring operating parameters of multiple substations; establishing digital twin models of each substation and using them as nodes to form a digital twin network; when operating parameters exceed a threshold, identifying the corresponding node as the target node; acquiring associated nodes and their associated digital twin models based on the digital twin network; extracting abnormal operating characteristics of the target node during the alarm period and associated operating characteristics of the associated nodes; inputting the features into a fault propagation analysis model based on a graph neural network, where the model nodes correspond to the digital twin network nodes, and their edge weights are determined based on historical fault propagation probabilities; using the model to determine the fault source location and propagation path; generating a maintenance strategy based on the fault source location, propagation path, and target digital twin model and sending it to the corresponding execution terminal. This invention achieves accurate fault tracing and intelligent maintenance across substations. Attached Figure Description

[0018] Figure 1 This is a flowchart of a substation intelligent maintenance method based on digital twins according to the present invention. Figure 2 This is a flowchart of a method according to an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this invention are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0020] The present invention proposes the following technical solutions, such as... Figure 1 As shown, this invention proposes an intelligent substation maintenance method based on digital twins, comprising: Obtain the operating parameters corresponding to multiple substations, establish digital twin models corresponding to multiple substations based on the operating parameters, and use each digital twin model as a node to form a digital twin network; Obtain electrical parameters, environmental parameters, and equipment status parameters corresponding to multiple substations, and align the electrical parameters, environmental parameters, and equipment status parameters according to timestamps to form a time-series operation dataset corresponding to each substation; A device model is constructed based on the aforementioned time-series runtime dataset; Based on the physical connection relationships between the devices in each substation, the device models are merged to form the digital twin model corresponding to the substation.

[0021] When the operating parameters of any of the substations exceed the preset alarm threshold, the node corresponding to the substation is identified as the target node. Based on the digital twin network, nodes whose electrical distance to the target node is less than a preset distance threshold are obtained as associated nodes; or, nodes that have the same fault root cause as the target node in historical fault records are obtained as associated nodes, and the digital twin model corresponding to the associated node is obtained as the associated digital twin model. Obtain the abnormal operation characteristics of the target digital twin model corresponding to the target node during the alarm period, and the associated operation characteristics of the associated digital twin model during the alarm period; The abnormal operation characteristics and the associated operation characteristics are input into a preset fault propagation analysis model, wherein the fault propagation analysis model is constructed based on a graph neural network model, and the network nodes of the graph neural network model correspond to the nodes in the digital twin network; Align the target runtime sequence set with the associated runtime sequence set, obtain the runtime parameter difference values ​​between the target digital twin model and each associated digital twin model at the same time, and generate a difference feature matrix based on the runtime parameter difference values; The abnormal operation characteristics, the associated operation characteristics, and the difference feature matrix are used together as input features of the fault propagation analysis model.

[0022] During message transmission, for a directed edge from the source node to the target node, the fault propagation analysis model reads the difference feature vector in the difference feature matrix corresponding to the edge, concatenates the difference feature vector with the feature vector of the source node or multiplies it element by element, and then multiplies it by the edge weight of the edge to generate a message to be transmitted to the target node.

[0023] The edge weights of the graph neural network are determined based on the historical fault propagation probability between the substations corresponding to each node. Obtain historical fault records between the substations within a preset historical time period. The historical fault records include the fault source substation, the substations through which the fault propagates, and the substations ultimately affected by the fault. Obtain the first substation and the second substation corresponding to any two nodes in the digital twin network, the number of times the fault propagated from the first substation to the second substation in the historical fault record, and the total number of faults in the historical fault record where the first substation is the fault source. Based on the ratio of the number of fault propagation attempts to the total number of faults, the historical fault propagation probability from the first substation to the second substation is obtained; The edge weights of the graph neural network are determined based on the historical fault propagation probability between each pair of nodes.

[0024] Based on the fault propagation analysis model, the location of the fault source corresponding to the abnormal operation characteristics and the propagation path of the fault source location through the digital twin network affecting other nodes are determined. Based on the edge weights, message passing and feature aggregation are performed between each node to output the corresponding fault probability vector for each node. The target node and each of the associated nodes are sorted from high to low according to the fault probability vector, and the node at the top of the sort is selected as the fault source location. Starting from the location of the fault source, the transmission path to other nodes is determined by tracing back along the edge weight direction in the graph neural network model. The path in the propagation path in which the edge weights are continuous and greater than a preset propagation threshold is determined as the propagation path, wherein the propagation path includes a fault source node, at least one intermediate propagation node and an end node.

[0025] Based on the location of the fault source, the propagation path, and the target digital twin model, a maintenance strategy is determined, and the maintenance strategy is sent to the execution terminal corresponding to the location of the fault source.

[0026] Based on the location of the fault source, determine the device model corresponding to the location of the fault source from the target digital twin model, and obtain the device type, device identifier and current operating status of the device model; Based on the propagation path, determine the set of affected nodes influenced by the location of the fault source, and obtain the affected devices and the degree of impact corresponding to each node in the set of affected nodes; The fault type and severity of the fault source location are obtained. Based on the degree of impact and the severity of the fault, at least one candidate maintenance plan is generated through a preset maintenance strategy generation model. The candidate maintenance plan includes the target equipment to be maintained, the maintenance operation sequence, the required spare parts and components, and the estimated maintenance time.

[0027] The second aspect of this invention proposes a digital twin-based intelligent substation maintenance system, comprising a control terminal and execution terminals corresponding one-to-one with multiple substations, configured as follows: Based on the execution terminal, the operating parameters of multiple substations are obtained, and digital twin models of multiple substations are established based on the operating parameters. Each digital twin model is used as a node to form a digital twin network. When the operating parameters of any substation exceed the preset alarm threshold, the node corresponding to the substation will be identified as the target node. Based on the digital twin network, nodes whose electrical distance from the target node is less than a preset distance threshold are obtained as associated nodes, or nodes that have the same fault root cause as the target node in the historical fault record are obtained as associated nodes, and the digital twin model corresponding to the associated node is obtained as the associated digital twin model. Obtain the abnormal operation characteristics of the target digital twin model corresponding to the target node during the alarm period, as well as the associated operation characteristics of the related digital twin model during the alarm period; Abnormal operation characteristics and associated operation characteristics are input into a preset fault propagation analysis model, wherein the fault propagation analysis model is based on a graph neural network model, and the network nodes of the graph neural network model correspond to the nodes in the digital twin network. The edge weights of the graph neural network are determined based on the historical fault propagation probability between substations corresponding to each node. Based on the fault propagation analysis model, the location of the fault source corresponding to the abnormal operation characteristics and the propagation path of the fault source location to other nodes affected by the digital twin network are determined. Based on the fault source location, propagation path, and target digital twin model, a maintenance strategy is determined and sent to the execution terminal corresponding to the fault source location.

[0028] A third aspect of this invention provides an electronic device, which includes: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method proposed in the first aspect of the present invention.

[0029] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in the first aspect of the present invention.

[0030] Example 1 This embodiment proposes a substation intelligent maintenance method based on digital twins. Please refer to [link / reference]. Figure 2 The method includes the following steps: S101: Obtain the operating parameters corresponding to multiple substations based on the execution terminal, establish digital twin models corresponding to multiple substations based on the operating parameters, and use each digital twin model as a node to form a digital twin network.

[0031] Understandably, the first step is to collect various parameters generated during the operation of each substation by deploying execution terminals within the substations. These operational parameters include at least electrical parameters characterizing the operating status of the power system, non-electrical parameters reflecting the operating environment of the equipment, and equipment status parameters characterizing the health status of the equipment itself. Of course, these parameters are aligned using a unified timestamp to form a time-series operational dataset for each substation, ensuring the time synchronization of multi-source data during subsequent model building.

[0032] Based on the acquired data, the control terminal constructs a corresponding digital twin model for each substation.

[0033] Specifically, as one implementation method, step S101 may include the following steps: S1011: Based on the execution terminal, obtain the electrical parameters, environmental parameters and equipment status parameters corresponding to multiple substations, align the electrical parameters, environmental parameters and equipment status parameters according to the timestamp to form a time-series operation dataset corresponding to each substation.

[0034] Understandably, execution terminals deployed in each substation can be used to characterize the operating status of the substation's primary system, including but not limited to three-phase voltage, three-phase current, active power, reactive power, power factor, frequency, and harmonic content. For example, for a 220kV transmission line, the execution terminal can collect electrical quantity data such as the effective value of its A-phase voltage, the effective value of its A-phase current, and the active power transmitted by the line. The second category is environmental parameters, used to reflect the external conditions of substation operation, including temperature, humidity, and wind speed. For example, temperature sensors deployed in the transformer area can collect ambient temperature and transformer body temperature in real time to assess the impact of heat load on equipment operation. The third category is equipment status parameters, used to characterize the health status and mechanical condition of various equipment in the substation, including the opening and closing positions of circuit breakers, energy storage status, opening and closing coil current waveforms, transformer oil temperature, oil level, and oil chromatography analysis data, disconnector contact temperature and mechanical vibration signals, and the self-test status and number of operations of relay protection devices.

[0035] After obtaining the above parameters, they are aligned according to their timestamps. For example, the voltage value, ambient temperature value, and circuit breaker opening / closing status at a certain moment are grouped into a complete set of data records for that moment. For parameters with different sampling frequencies, interpolation or hold mechanisms are used during the alignment process to match high-frequency and low-frequency data on a unified time axis, forming a complete data profile for each moment. For each substation, a corresponding time-series operation dataset is generated. This dataset is indexed by time, with each row corresponding to a complete snapshot of the operating status at a timestamp, containing the specific values ​​of all electrical parameters, environmental parameters, and equipment status parameters at that moment.

[0036] S1012: Construct a device model based on a time-series runtime dataset.

[0037] Specifically, for each type of equipment within a substation, the equipment model can consist of two parts: a static model and a dynamic model. The static model describes the inherent properties of the equipment. The dynamic model, based on real-time parameters from a time-series operational dataset, describes the operating state and behavioral characteristics of the equipment.

[0038] In this invention, taking a transformer as an example, a static model of the transformer is first constructed, recording fixed information such as its rated capacity, rated voltage, connection group, cooling method, winding material, and factory test data. Subsequently, based on the time-series parameters of the transformer, such as voltage, current, active power, and reactive power, collected from the time-series operational data, a dynamic model of the transformer is constructed. This dynamic model includes multiple dimensions, such as an electrical characteristic model, a thermal characteristic model, and an insulation characteristic model. Specific construction methods have been disclosed in relevant technical documents and are not limited in this invention.

[0039] It should be noted that during the construction of the equipment model, a parameter identification and model calibration mechanism can be employed. Historical data from the time-series dataset is used to continuously correct the model parameters, bringing the error between the model output and the measured values ​​within an acceptable range. For example, the heat capacity and thermal resistance parameters in the transformer thermal characteristic model can be identified using optimization algorithms based on historical oil temperature and load data, enabling the model to accurately reflect the actual heat dissipation characteristics of the transformer. This parameter identification and calibration process ensures that the equipment model can adapt to individual equipment differences and changes in the operating environment.

[0040] Understandably, a separate equipment model is built for each major piece of equipment within the substation. Each equipment model has independent input / output interfaces and can receive real-time operating parameters.

[0041] S1013: Based on the physical connection relationship between the equipment in each substation, the equipment models are merged to form a digital twin model of the substation.

[0042] Based on the completed equipment model, the individual equipment models are integrated according to the actual physical connection relationships within the substation to form a station-level digital twin model that reflects the overall structure and operating characteristics of the substation.

[0043] Specifically, the first step is to obtain the physical connection relationships between the various devices within each substation. The physical connection relationships mainly refer to the electrical wiring topology, including the main wiring configuration of the substation, the electrical connection paths between various devices, the busbar segmentation method, and the incoming and outgoing line circuits.

[0044] Based on the aforementioned physical connections, the various device models can be merged. The fusion process involves establishing interface mapping relationships between the device models. Taking a transformer and a circuit breaker as an example, the transformer's output side and the circuit breaker's input side are physically connected directly via conductors. In the digital twin model, the control terminal associates the output port of the transformer model with the input port of the circuit breaker model, defining the data types that interact between the two device models, including electrical quantity information such as voltage, current, power flow direction, and fault current flowing through the connection point. When the transformer model calculates the output side voltage and current based on operating parameters, this electrical quantity data is transmitted to the connected circuit breaker model through the interface, serving as the boundary conditions for the circuit breaker model's calculations.

[0045] After the above fusion process, a complete digital twin model of the substation is formed. This digital twin model can reflect the overall operating status of the substation, such as the balance between total incoming and outgoing power, the distribution of bus voltage, and the load rate level of the main transformer. Furthermore, when an anomaly occurs in a piece of equipment within the substation, the model can, based on the electrical coupling relationships between equipment models, extrapolate the scope and extent of the impact of the anomaly on adjacent equipment and even the entire substation operation.

[0046] S102: When the operating parameters of any substation exceed the preset alarm threshold, the node corresponding to the substation will be identified as the target node.

[0047] Specifically, for each substation, alarm thresholds adapted to its equipment characteristics and operating conditions can be pre-configured. These preset alarm thresholds cover various parameter types, including upper and lower limits for electrical parameters, safe ranges for non-electrical parameters, and normal ranges for equipment status parameters. The control terminal can compare the received operating parameters with the corresponding preset alarm thresholds item by item. When one or more operating parameters of any substation exceed the preset alarm threshold, the substation is determined to be in an abnormal state. After identifying the abnormal substation, the control terminal determines the node corresponding to that substation in the digital twin network as the target node.

[0048] S103: Based on the digital twin network, obtain nodes whose electrical distance from the target node is less than a preset distance threshold as associated nodes, or obtain nodes that have the same fault root cause as the target node in the historical fault record as associated nodes, and obtain the digital twin model corresponding to the associated node as the associated digital twin model.

[0049] For example, the control terminal acquires the electrical distances between the target node and other nodes in the digital twin network. Electrical distance is a quantitative indicator measuring the tightness of electrical coupling between two substations, typically calculated based on the power grid topology and line parameters. For instance, the electrical distance can be taken as the equivalent impedance modulus of the transmission line between the two substations; a smaller impedance value indicates a tighter electrical connection, making it easier for faults or disturbances to propagate from one substation to the other. The control terminal identifies nodes with electrical distances less than a preset distance threshold as associated nodes. This preset distance threshold can be flexibly set according to factors such as the power grid voltage level and the regional power grid structure. For example, for a substation, the electrical distance threshold can be set to the corresponding line impedance value of 0.05; all substation nodes with equivalent impedances less than this threshold to the target substation are selected as associated nodes. Associated nodes selected in this way typically include adjacent substations directly connected to the target node via short lines, as well as other substations located in the same electrical hub area within the power grid topology.

[0050] In other embodiments, filtering can also be based on the same root cause of faults in historical fault records. For example, past fault records can be retrieved from a historical fault database to analyze whether multiple substations in each fault event share a common root cause. A common root cause refers to multiple substations experiencing anomalies simultaneously due to the same initial fault event, such as a system oscillation affecting multiple substations. The control terminal identifies nodes that share the same root cause of faults with the target node in historical fault events related to the target node as associated nodes.

[0051] After identifying the associated nodes, the digital twin model corresponding to each associated node can be obtained from the digital twin network, and these models can be used as associated digital twin models.

[0052] S104: Obtain the abnormal operation characteristics of the target digital twin model corresponding to the target node during the alarm period, and the associated operation characteristics of the associated digital twin model during the alarm period.

[0053] After identifying the target node and associated nodes, the next step is to extract key feature information from the time dimension. Specifically, as one implementation method, step S104 may include the following steps: S1041: Based on the target digital twin model, obtain the time-series data of the device model's operating parameters during the alarm period to form the target runtime sequence set.

[0054] Specifically, the alarm period can be determined first. The alarm period is typically based on the moment the target node triggers the alarm, extending backward by a first preset duration and forward by a second preset duration, forming a continuous observation time window. After determining the alarm period, all device models included in the target digital twin model can be traversed. For each device model, the control terminal extracts its operational parameter time-series data within the alarm period. Operational parameter time-series data refers to the data sequence of parameter values ​​recorded by the device model during the alarm period, showing how these values ​​change over time.

[0055] Different equipment models contain different types of parameters. In this invention, taking a transformer equipment model as an example, the extracted time series data of operating parameters may include the high-voltage side A-phase voltage value sequence (unit kV, time interval 1 second), the high-voltage side A-phase current value sequence (unit A), the active power sequence (unit MW), the reactive power sequence (unit Mvar), etc.

[0056] For each parameter, the extracted parameter values ​​are arranged in chronological order to form a time-series data sequence. Parameters with different sampling frequencies retain their original temporal resolution after extraction. The control terminal retains the timestamp information of each data point during the extraction process to ensure accurate alignment of data from different sources during subsequent processing. These data are then aggregated by timestamp to form the target runtime sequence set. The target runtime sequence set is a structured dataset indexed by time, with each timestamp containing all operational parameter values ​​for all device models in the target digital twin model at that specific moment.

[0057] S1042: Determine abnormal operation characteristics based on the target runtime sequence set, wherein the abnormal operation characteristics include electrical quantity mutation characteristics, non-electrical quantity limit exceeding characteristics, and timing correlation deviation characteristics between devices.

[0058] Understandably, by using the target runtime sequence set as input, three types of abnormal runtime features can be extracted.

[0059] The first category is electrical quantity abrupt change characteristics, used to describe changes in substation electrical parameters during the alarm period. Electrical quantities mainly include parameters reflecting the electromagnetic transient processes of the power system, such as voltage, current, power, and frequency. The control terminal can identify abrupt change points and quantify the abrupt change characteristics. Extraction of abrupt change characteristics typically employs methods such as the finite difference method, wavelet transform, or sliding window detection.

[0060] For example, the control terminal analyzes the timing data of the A-phase current on the high-voltage side of the transformer and finds that between 14:23:05 and 14:23:06, the current value jumps instantaneously from the normal operating 200A to 1500A, and then drops back within 0.1 seconds. To address this phenomenon, the control terminal extracts the following electrical quantity abrupt change characteristics: abrupt change start time (14:23:05), abrupt change amplitude (from 200A to 1500A, a change of 1300A), a change rate (13000A / second), abrupt change duration (0.1 seconds), and abrupt change waveform characteristics (such as impulse waveform and attenuation characteristics). For voltage parameters, features such as voltage sag depth, sag duration, and voltage recovery time can be extracted; for frequency parameters, features such as frequency offset and frequency change rate can be extracted. Specific abrupt change characteristics are not limited in this invention.

[0061] The second category is non-electrical quantity limit-breaking characteristics, used to describe situations where non-electrical parameters of equipment in a substation exceed their normal operating range. Non-electrical quantities mainly include parameters reflecting the physical state of the equipment and environmental conditions, such as temperature, pressure, displacement, vibration, gas concentration, and oil level. The non-electrical quantity parameters in the target operating sequence set can be compared with their corresponding normal ranges or thresholds to identify limit-breaking events and quantify the limit-breaking characteristics.

[0062] The third category is the inter-equipment timing correlation deviation characteristic, which describes the abnormal deviation of the coordination relationship between different devices in the time series. Under normal operating conditions, the state changes of each device in the substation follow specific timing logic relationships. For example, there should be a definite sequence and time interval between the circuit breaker opening operation and the protection device action signal, and there should be a lag relationship between the transformer oil temperature change and the load current change that conforms to thermal inertia.

[0063] By analyzing the parameter sequences of multiple device models in the target runtime sequence set, the degree of deviation of these time-series correlations is extracted. For example, there is a thermal inertia hysteresis relationship between changes in transformer oil temperature and changes in load current. The time delay between the rise in load current and the rise in oil temperature can be calculated. If this delay is significantly less than the historical normal value, it may reflect an abnormality in the transformer heat dissipation system. This deviation is also extracted as a time-series correlation deviation feature.

[0064] S1043: Based on the associated digital twin model, determine the timing data of the device model's operating parameters during the alarm period to form an associated runtime timing set.

[0065] The structure of the associated digital twin model is the same as that of the target digital twin model. The control terminal can extract the time series data of the operating parameters for that period from each associated digital twin model.

[0066] For each associated digital twin model, it is possible to iterate through all the device models contained within it.

[0067] Taking a 110kV substation corresponding to a certain associated node as an example, the associated digital twin model of this substation includes incoming transformer equipment models, circuit breaker equipment models, busbar equipment models, etc. The control terminal extracts the time-series data of its operating parameters during the alarm period from each equipment model. Taking the incoming transformer equipment model as an example, the extracted time-series data of operating parameters may include high-voltage side voltage time-series, high-voltage side current time-series, active power time-series, reactive power time-series, etc.

[0068] After extracting the timing data of the operating parameters of all device models in a certain associated digital twin model, all the extracted data under that associated node can be aggregated according to timestamps to form the associated runtime sequence subset corresponding to that associated node. Finally, the associated runtime sequence subsets of all associated nodes are integrated to form a complete associated runtime sequence set.

[0069] For example, suppose the target node is a 220kV substation A, and the associated nodes include the adjacent 110kV substation B and the 220kV substation C in the same area. The control terminal extracts the time-series data of the operating parameters of substation B from 14:00 to 14:50, forming an associated operating sequence subset for substation B. This subset includes data such as the voltage time-series of the high-voltage side of the incoming transformer, the incoming current time-series, and the protection device start-up signal time-series. Simultaneously, the time-series data of the operating parameters of substation C during the same time period is extracted, forming an associated operating sequence subset for substation C. These two subsets together constitute the associated operating sequence set, which corresponds to the target operating sequence set of the target node within the same time window.

[0070] S1044: Determine the associated operation characteristics based on the associated runtime sequence set, wherein the associated operation characteristics include electrical quantity fluctuation characteristics, non-electrical quantity change rate characteristics, and inter-device timing synchronization characteristics.

[0071] Based on the acquisition of the associated runtime sequence set, the process that reflects the changes in the operating status of the associated substation during the alarm period can be extracted by analyzing and processing the time sequence data of each associated node.

[0072] Specifically, the associated runtime sequence set is used as input. This dataset contains the time-series data of the operating parameters of each device model during the alarm period for all associated nodes. The control terminal uses methods such as signal analysis, statistical calculation, and correlation analysis to extract three types of associated operational features. The specific methods have been disclosed in relevant technical documents and are not limited here.

[0073] The first category is electrical quantity fluctuation characteristics, used to describe the magnitude and fluctuation pattern of electrical parameters in associated substations during the alarm period. Electrical quantities mainly include parameters such as voltage, current, power, and frequency. Unlike the abrupt changes in electrical quantities at the target node, the fluctuation characteristics of electrical quantities at associated nodes focus on reflecting the degree and pattern of change in electrical quantities during disturbance propagation, rather than the abrupt change itself. Specifically, the time-series data of electrical quantities at each associated node in the associated runtime sequence set can be analyzed to extract quantitative indicators such as fluctuation amplitude, fluctuation frequency, and fluctuation duration. When the change in electrical quantity time-series data relative to its historical steady-state baseline value exceeds a preset threshold during the alarm period, it is extracted as an electrical quantity fluctuation characteristic. For example, for current parameters, the change in current harmonic content can be extracted. If the current harmonic distortion rate of an associated node increases from the normal 1.2% to 4.5% during the alarm period, this fluctuation characteristic is recorded. For frequency parameters, the oscillation amplitude and oscillation period can be extracted.

[0074] The second category is non-electrical quantity change rate characteristics, used to describe the rate and trend of change of non-electrical parameters of equipment within the associated substation. Non-electrical quantities mainly include parameters reflecting the physical state of equipment, such as temperature, pressure, vibration, and gas concentration. Unlike the non-electrical quantity limit-breaking characteristics of the target node, the non-electrical quantity change rate characteristics of the associated node focus on identifying the rate and direction of parameter change, rather than whether it exceeds the threshold. Specifically, the non-electrical quantity parameters in the associated operating sequence set can be differentially or differentially calculated to extract the change rate value and its change law over time. If the change rate exceeds the preset rate threshold during the alarm period, it is extracted as a non-electrical quantity change rate characteristic. For example, for vibration parameters, if the mechanical vibration amplitude of the circuit breaker in the associated node increases at a rate of 0.05g every 10 minutes during the alarm period, the control terminal records this vibration change rate characteristic; for gas concentration parameters, the change rate of characteristic gas concentration can be extracted to reflect whether there is a slowly developing abnormal state in the equipment.

[0075] The third category is the timing synchronization characteristic between devices, which describes the degree of consistency in the timing of state changes between different devices within an associated node, and between different associated nodes. Timing synchronization characteristics reflect the time relationship of fault or disturbance propagation between different devices and different substations. Specifically, timing alignment analysis can be performed on the state change events of multiple device models and multiple associated nodes in the associated runtime timing set to calculate the time delay and synchronization degree between each event. For example, within the same associated node, the control terminal analyzes the time interval between the protection device's action signal and the circuit breaker's tripping signal. If this interval meets the normal operating requirements, it is recorded as good timing synchronization; if there is an abnormal delay, the delay duration is recorded. Furthermore, by comparing the action timing of similar devices in multiple associated nodes, if it is found that the protection devices of multiple nodes act sequentially within a similar time frame (e.g., within 1 second), the control terminal extracts the time sequence characteristics of these action events to reflect the propagation timing of the fault in the regional power grid.

[0076] S105: Input the abnormal operation characteristics and related operation characteristics into the preset fault propagation analysis model. The fault propagation analysis model is constructed based on the graph neural network model, and the network nodes of the graph neural network model correspond to the nodes in the digital twin network.

[0077] In this invention, the operational characteristics of the target node and associated nodes can be combined with the topology of the digital twin network to form feature data that meets the input requirements of a graph neural network. Specifically, the target runtime sequence set can be aligned with the associated runtime sequence set, and the difference values ​​of the operating parameters between the target digital twin model and each associated digital twin model at the same time can be obtained. A difference feature matrix is ​​generated based on the difference values ​​of the operating parameters. Then, the abnormal operating features, associated operating features, and difference feature matrix are used together as input features of the fault propagation analysis model. The difference feature matrix participates in message passing as the edge features of the corresponding edges in the graph neural network within the model. The specific interaction method is described in detail in S1071.

[0078] Understandably, during the input process, these features are assigned according to nodes. That is, the abnormal operation features of the target node are assigned to the corresponding target node in the graph neural network, and the associated operation features of each associated node are assigned to the corresponding associated nodes in the graph neural network. For other nodes in the digital twin network that are neither target nodes nor associated nodes, the control terminal sets their input features to default values ​​or to zero, indicating that these nodes do not participate in active computation in this analysis.

[0079] After feature assignment, each node in the graph neural network model carries a corresponding operational feature vector. For example, taking the target node, its abnormal operational feature vector includes quantified values ​​across multiple dimensions, such as electrical quantity mutation characteristics, non-electrical quantity limit exceedance characteristics, and inter-device timing correlation deviation characteristics. Taking a related node as an example, its associated operational feature vector includes quantified values ​​such as electrical quantity fluctuation characteristics, non-electrical quantity change rate characteristics, and inter-device timing synchronization characteristics.

[0080] The network nodes of the graph neural network model correspond to the nodes in the digital twin network. That is, the model can use the electrical connection relationships and topology between substations reflected by the digital twin network to pass messages. Each node not only performs state calculations based on its own input features, but can also receive feature information from neighboring nodes through the edges in the graph structure. Through a multi-layer iterative message passing mechanism, feature information from nodes over a larger range is aggregated, thereby realizing the modeling of the fault propagation process.

[0081] S106: Determine the edge weights of the graph neural network based on the historical fault propagation probability between substations corresponding to each node.

[0082] Specifically, step S106 may include the following steps: S1061: Obtain historical fault records between substations within a preset historical time period. The historical fault records include the fault source substation, the substations through which the fault propagates, and the substations ultimately affected by the fault.

[0083] Understandably, for each fault event, the control terminal records three types of key information according to the timing and spatial range of the fault propagation.

[0084] The first category is the fault source substation, which is the location where the initial fault occurred. The fault source substation is the starting point that triggers the entire fault event. It usually refers to the substation that first exhibits an anomaly and whose abnormal state directly leads to a chain reaction in other substations.

[0085] The second category is fault propagation path substations, which are intermediate nodes along the path of a fault as it spreads from the source substation to other substations. Fault propagation path substations are critical nodes in the fault's propagation path within the power grid. While they may not be the source of the fault themselves, they are significantly affected during the fault propagation process, such as by abnormal states like protection device activation, switchgear tripping, voltage dips, and current surges. In a single fault event, there may be one or more fault propagation path substations, forming one or more propagation chains.

[0086] The third category is substations that are ultimately affected by a fault. These are substations that are affected by the fault event but are not the source of the fault and do not act as a major propagation node. Substations that are ultimately affected by a fault are usually at the end of the fault propagation chain. Their abnormal state may be due to fault propagation, but it does not further transmit the fault to other substations.

[0087] Understandably, each record corresponds to an independent fault event, including the clearly identified source substation, zero or more substations along the fault propagation path, and zero or more substations ultimately affected by the fault. For simple, isolated fault events, only the source substation may be included, without any propagation path substations or ultimately affected substations. For complex cascading fault events, multiple propagation path substations and multiple ultimately affected substations may be included.

[0088] S1062: Obtain the first and second substations corresponding to any two nodes in the digital twin network, the number of times the fault propagated from the first substation to the second substation in the historical fault records, and the total number of faults in the historical fault records where the first substation is the fault source.

[0089] Understandably, the substations corresponding to any two nodes in the digital twin network can be identified, and denoted as Substation 1 and Substation 2, respectively. Substation 1 and Substation 2 can be any two different substation nodes in the digital twin network, and there may or may not be a direct electrical connection between them.

[0090] For the selected first and second substations, the number of fault propagations from the first substation to the second substation can be counted. Specifically, in the historical record of a fault event, if the first substation is marked as the fault source substation of the event, and the second substation appears in the fault propagation path substation or the substation ultimately affected by the fault, then it is determined as a fault propagation event from the first substation to the second substation. In a single fault event, regardless of the degree of impact on the second substation, only one propagation count is recorded, and the counts are not cumulative. For example, in a fault event, if the fault source is substation A, and substation B is both recorded as a propagation path substation and is ultimately affected, then when counting the propagation count from A to B, this event is counted as only one.

[0091] Furthermore, the total number of faults originating from the first substation in historical fault records can be counted. Each historical fault record originating from the first substation is counted as one fault, regardless of its propagation range or the number of affected nodes.

[0092] It should be noted that if another pair of nodes is considered, namely the first substation B and the second substation A, the statistical logic is the same. The control terminal needs to count the total number of faults with substation B as the fault source, and the number of times substation A was affected in these events. Since fault propagation is directional, the statistical results from A to B are usually different from those from B to A, and the two are calculated independently.

[0093] S1063: Based on the ratio of the number of fault propagation times to the total number of faults, obtain the historical fault propagation probability from the first substation to the second substation.

[0094] For each ordered pair of nodes in a digital twin network, i.e., in the direction from the first substation to the second substation, we can obtain the number of times a fault propagates from the first substation to the second substation, and the total number of faults where the first substation is the source of the fault. Dividing the former by the latter, the resulting ratio is the historical fault propagation probability from the first substation to the second substation.

[0095] Understandably, the calculated result ranges from 0 to 1. A ratio of 0 indicates that, historically, although the first substation experienced a fault as a source, it never affected the second substation. A ratio of 1 indicates that whenever the first substation experiences a fault as a source, the second substation is inevitably affected. A ratio between 0 and 1 indicates that the fault in the first substation propagates to the second substation with a certain probability. Let's illustrate this with a specific example. Assume that within a preset historical period, substation A experiences a total of 10 faults as a source. Of these 10 fault events, 7 propagated to substation B, 4 to substation C, and 0 to substation D. The control terminal calculates the following: the historical fault propagation probability from substation A to substation B is 7 ÷ 10 = 0.7, or 70%; the historical fault propagation probability from substation A to substation C is 4 ÷ 10 = 0.4, or 40%; and the historical fault propagation probability from substation A to substation D is 0 ÷ 10 = 0, or 0%.

[0096] It's important to note that the historical fault propagation probability has a clear directionality. Using the data above as an example, the probability from A to B is 0.7, but the probability from B to A needs to be calculated separately. Assuming substation B is the source of a total of 8 faults, and 3 of these faults propagate to substation A, then the historical fault propagation probability from B to A is 3 ÷ 8 = 0.375, or 37.5%. Since the numerator and denominator are different, the propagation probabilities in the two directions are usually not equal.

[0097] The control terminal repeats the above calculation process for all ordered node pairs in the digital twin network. For any ordered node pair, as long as the total number of faults caused by the first substation as a fault source is greater than zero, the corresponding historical fault propagation probability can be calculated. If the first substation has never caused a fault as a fault source, the propagation probability to other substations cannot be calculated. In this case, the propagation probability in this direction can be set to a default value, such as 0.

[0098] S1064: Determine the edge weights of the graph neural network based on the historical fault propagation probability between each node pair.

[0099] Understandably, specifically, after calculating the historical fault propagation probability for all ordered node pairs, a probability matrix containing directional weights is obtained. This matrix records the fault propagation probability value between any two nodes in the digital twin network, from one node to the other.

[0100] S107: Based on the fault propagation analysis model, determine the location of the fault source corresponding to the abnormal operation characteristics and the propagation path of the fault source location to other nodes affected by the digital twin network.

[0101] Specifically, step S107 may include the following steps: S1071: Based on edge weights, message passing and feature aggregation are performed between each node, and the corresponding fault probability vector is output.

[0102] Specifically, after feature allocation and edge weight assignment, the graph neural network model has complete input conditions: each node carries a corresponding operational feature vector, and each directed edge carries a corresponding edge weight. Therefore, the graph neural network model can begin multi-layer iterative message passing and feature aggregation based on this. Furthermore, each element in the difference feature matrix represents the difference pattern of operating parameters between the target node and its corresponding associated node during the alarm period. This matrix is ​​used as the edge feature vector of the corresponding edge in the graph neural network. During message passing, for a directed edge from node u to node v, the model not only uses the historical fault propagation probability of that edge as a weight, but also reads the corresponding difference feature vector of that edge, interacts with it and the feature vector of the source node u (e.g., through concatenation or element-wise multiplication), and then multiplies it by the edge weight to generate a more accurate message representation. In this way, when the actual operating differences between the two substations are significant, the edge features will correspondingly change the information content being transmitted, helping the model identify the dynamic correlation of fault propagation and avoiding the shortcomings of relying solely on historical statistical probabilities while ignoring the current actual operating conditions.

[0103] The message passing process follows the basic computational paradigm of graph neural networks, which will not be elaborated upon in this invention. Understandably, in each iteration of the digital twin network, for each node, the model collects information passed from all neighboring nodes pointing to that node. The strength of information passing is determined by the edge weights; that is, the larger the edge weight, the higher the proportion of information passed from the source node to the target node during aggregation. The information passed from a neighboring node to the current node is the product of the current feature vector of that neighboring node and the edge weight. The model performs a weighted summation of the information passed from all neighboring nodes pointing to the current node, forming the aggregated information received by the current node at this layer.

[0104] The feature aggregation process fuses the original feature vector of the current node with the aggregated information to generate the updated feature vector for the node in the next layer. The fusion method typically employs a non-linear transformation, such as combining the node's own features and aggregated information through a fully connected layer with an activation function, enabling the model to learn complex interactions between features. In this way, each node's feature vector gradually incorporates information from its more distant neighbors in each iteration. After multiple iterations, each node's feature vector will aggregate the feature information from all nodes within its multi-hop range, while the edge weights continuously guide the direction and intensity of information flow in each layer.

[0105] After completing message passing and feature aggregation across a predetermined number of layers, the model adds an output layer above the last layer. This output layer maps the final feature vector of each node to a scalar value, representing the node's failure probability. This mapping process typically uses a fully connected layer, with the output activated by a sigmoid or softmax function to ensure the probability value falls within the range of 0 to 1. Ultimately, the model outputs a failure probability value for each node in the digital twin network, and these values ​​collectively form a failure probability vector.

[0106] After the control terminal obtains the fault probability vector, the fault probability value of each node reflects the likelihood that the node is the source of the fault in the current abnormal event. Typically, the fault probability value of the source node is significantly higher than that of other nodes. This is because the source node's own abnormal operation characteristics are the most significant, and its features propagate to other nodes through high-weighted edges. This allows the model to effectively capture this propagation relationship during message passing and feature aggregation, and assign the highest probability value to the source node in the output layer.

[0107] S1072: Sort the target node and each associated node in descending order of fault probability vector, and select the node at the top of the sort as the fault source location.

[0108] Specifically, the fault probability values ​​of the target node and its associated nodes can be extracted from the fault probability vector, forming a set of nodes and probabilities to be sorted. This set is then sorted in descending order of fault probability values. The sorting process follows numerical comparison rules: the node with the highest fault probability value is placed first, the next highest second, and so on. After sorting, the node at the top of the sorted list is selected as the fault source location. This top-ranked node, with the highest fault probability value, is considered, according to the computational logic of the graph neural network model, the most likely initial fault location to trigger the current abnormal event.

[0109] S1073: Starting from the location of the fault source, trace back along the edge weight direction in the graph neural network model to determine the propagation path to other nodes.

[0110] Specifically, the location of the fault source is first identified and designated as the starting node. Starting from this node, a reverse tracing is performed along the edge weights in the graph neural network model. Reverse tracing refers to starting from the fault source node and expanding outwards along the direction of the edges, identifying the nodes to which the fault characteristics have propagated layer by layer. Since the edge weights in the graph neural network model are determined based on historical fault propagation probabilities, the magnitude of the weight reflects the strength of the probability of the fault propagating along that direction. During the reverse tracing process, the control terminal prioritizes propagation directions with higher edge weights, as these directions are more likely to propagate faults historically.

[0111] Specifically, the tracing process follows either a breadth-first or depth-first search strategy. Starting from the fault source node, the control terminal examines all directed edges originating from that node and pointing to other nodes, obtaining the target node connected to each edge and its corresponding edge weight. For target nodes with edge weights greater than a preset threshold, they are included in the first-layer node set of the propagation path. Subsequently, using each node in the first-layer node set as a new starting point, the process continues to expand outward along the edge weight direction to identify the second-layer node set, and so on, until no more nodes satisfying the edge weight threshold condition can be included, or the preset maximum tracing depth is reached.

[0112] During this process, the weight values ​​of each node and its predecessor node can be recorded, thus forming multiple propagation paths starting from the fault source node, passing through several intermediate nodes, and finally reaching the terminal node. Each propagation path is an ordered sequence of nodes, and there is a directed edge with a weight greater than a threshold between adjacent nodes in the sequence.

[0113] The above process is illustrated with a specific example. Assume there are five substation nodes A, B, C, D, and E in the digital twin network, with the fault source located at node A. The edge weights in the graph neural network model are configured as follows: the weight of the edge from A to B is 0.8, the weight of the edge from A to C is 0.6, the weight of the edge from B to D is 0.7, the weight of the edge from C to D is 0.4, the weight of the edge from D to E is 0.9, and the weights of other edges are all below a preset threshold of 0.2. The preset tracing threshold is 0.3.

[0114] The control terminal traces backwards from node A. First, it checks the edges originating from A: the edge pointing to B has a weight of 0.8, which is greater than the threshold, so node B is included in the propagation path, and A→B is recorded; the edge pointing to C has a weight of 0.6, which is greater than the threshold, so node C is included in the propagation path, and A→C is recorded. At this point, the first-level nodes are B and C.

[0115] Next, starting from node B, check the edges it points to: the edge pointing to D has a weight of 0.7, which is greater than the threshold, so node D is included in the propagation path, and the path is recorded as B→D. Starting from node C, check the edges it points to: the edge pointing to D has a weight of 0.4, which is greater than the threshold, so node D is already in the propagation path, and the path is recorded as C→D. At this point, the second-level node is D.

[0116] Starting from node D, check the edges it points to: the edge pointing to E has a weight of 0.9, which is greater than the threshold, so add node E to the propagation path and record D→E. At this point, the third-level node is E. Starting from node E, check the edges it points to; if there are no edges that meet the threshold condition, the tracing terminates.

[0117] Through the above tracing process, the control terminal identified two main transmission paths: Path 1 is A→B→D→E, and Path 2 is A→C→D→E. These two paths together constitute the complete set of propagation paths from the fault source location A to other nodes (B, C, D, E) via the digital twin network.

[0118] In determining the propagation path, the control terminal not only records the sequence of nodes along the path but also retains the weight information of each edge and the propagation direction between nodes. This information reflects the intensity and directionality of fault propagation, providing a concrete basis for the formulation of subsequent maintenance strategies regarding the scope of fault impact. This allows maintenance personnel to clearly understand the entire process of the fault gradually spreading from the source node to other nodes.

[0119] S1074: The path in the propagation path with continuous edge weights and greater than a preset propagation threshold is determined as the propagation path, wherein the propagation path includes the fault source node, at least one intermediate propagation node and the terminal node.

[0120] Specifically, some of these propagation paths may contain connections with low edge weights. These low-weight connections are statistically less likely to propagate faults, and including them in the propagation path could lead to misjudgments of the fault's impact. Therefore, the control terminal needs to filter the propagation paths, retaining only those paths with continuous edge weights that are all greater than a preset propagation threshold.

[0121] As is understandable, continuous edge weights mean that the weight of each directed edge between adjacent nodes in the path is greater than the preset propagation threshold, and there is no connection in the path where the edge weight is lower than or equal to the threshold.

[0122] The preset propagation threshold is a configurable numerical parameter, and its specific value can be adjusted according to the sparsity of historical fault data, the characteristics of the power grid structure, and the required analysis accuracy. This threshold is used to distinguish between significant and accidental propagation relationships. Connections with edge weights greater than the threshold are considered statistically significant and can be used as effective path segments for fault propagation; connections with edge weights less than or equal to the threshold are considered to have weak propagation relationships and are excluded when determining propagation paths.

[0123] The control terminal traverses each propagation path, checking the weight of each directed edge in the path to see if it is greater than a preset propagation threshold. If all edges in a path satisfy this condition, the path is determined to be a valid propagation path; if the weight of any edge in the path is less than or equal to the preset propagation threshold, the entire path is excluded and not used as the final propagation path.

[0124] After determining the propagation path, the control terminal further classifies the nodes in each propagation path by role. The propagation path is an ordered sequence of nodes, where the first node is the fault source node, i.e., the starting point of the fault. The last node in the sequence is the terminal node, representing the end point of the fault propagation; this node is affected by the fault but does not further propagate the fault to other nodes. Nodes located between the fault source node and the terminal node in the sequence are intermediate propagation nodes.

[0125] S108: Based on the fault source location, propagation path, and target digital twin model, determine the maintenance strategy and send the maintenance strategy to the execution terminal corresponding to the fault source location.

[0126] Understandably, the fault source location indicates the substation node where the initial fault point triggering the current abnormal event is located, while the propagation path reveals the complete chain of impact from the source node to other nodes. Therefore, based on the fault source location, the equipment model corresponding to the fault source location can be determined from the target digital twin model, and the equipment type, equipment identifier, and current operating status of the equipment model can be obtained. Based on the propagation path, the set of affected nodes influenced by the fault source location can be determined, and the affected equipment and the degree of impact corresponding to each node in the set of affected nodes can be obtained.

[0127] In this invention, the fault type and severity at the fault source location can be obtained. Based on the degree of impact and the severity, at least one candidate maintenance plan is generated through a preset maintenance strategy generation model. The candidate maintenance plan includes the target equipment to be maintained, the maintenance operation sequence, the required spare parts, and the estimated maintenance time. It is understood that the maintenance strategy can be generated by combining the fault type, fault severity, and degree of impact. The fault type and severity can be extracted from abnormal operating characteristics. For example, an inter-turn short circuit fault in the winding can be identified based on an excessive acetylene concentration in the transformer oil chromatogram, and a severe fault can be identified based on the magnitude of the fault current. This information is input into the preset maintenance strategy generation model, which generates one or more candidate maintenance plans based on the equipment type, fault mode, historical maintenance cases, and the current state of the equipment. Specific maintenance strategy generation models have been disclosed in relevant technical documents and are not limited in this invention.

[0128] Example 2 Based on the same inventive concept as Embodiment 1, this embodiment also proposes a digital twin-based intelligent substation maintenance system. The system includes a control terminal and execution terminals corresponding one-to-one with multiple substations. The system is configured as follows: Based on the execution terminal, the operating parameters of multiple substations are obtained, and digital twin models of multiple substations are established based on the operating parameters. Each digital twin model is used as a node to form a digital twin network. When the operating parameters of any substation exceed the preset alarm threshold, the node corresponding to the substation will be identified as the target node. Based on the digital twin network, nodes whose electrical distance from the target node is less than a preset distance threshold are obtained as associated nodes, or nodes that have the same fault root cause as the target node in the historical fault record are obtained as associated nodes, and the digital twin model corresponding to the associated node is obtained as the associated digital twin model. Obtain the abnormal operation characteristics of the target digital twin model corresponding to the target node during the alarm period, as well as the associated operation characteristics of the related digital twin model during the alarm period; Abnormal operation characteristics and associated operation characteristics are input into a preset fault propagation analysis model, wherein the fault propagation analysis model is based on a graph neural network model, and the network nodes of the graph neural network model correspond to the nodes in the digital twin network. The edge weights of the graph neural network are determined based on the historical fault propagation probability between substations corresponding to each node. Based on the fault propagation analysis model, the location of the fault source corresponding to the abnormal operation characteristics and the propagation path of the fault source location to other nodes affected by the digital twin network are determined. Based on the fault source location, propagation path, and target digital twin model, a maintenance strategy is determined and sent to the execution terminal corresponding to the fault source location.

[0129] Example 3 This invention discloses a digital twin-based intelligent substation maintenance method, applicable to a system including control terminals and execution terminals corresponding to multiple substations. The method includes: acquiring operating parameters of multiple substations; establishing digital twin models of each substation and using them as nodes to form a digital twin network; identifying the corresponding node as the target node when operating parameters exceed a threshold; acquiring associated nodes and their associated digital twin models based on the digital twin network; extracting abnormal operating characteristics of the target node during the alarm period and associated operating characteristics of the associated nodes; inputting the features into a fault propagation analysis model based on a graph neural network, where the model nodes correspond to the digital twin network nodes, and their edge weights are determined based on historical fault propagation probabilities; using the model to determine the fault source location and propagation path; generating a maintenance strategy based on the fault source location, propagation path, and target digital twin model, and sending it to the corresponding execution terminal. This invention achieves accurate fault tracing and intelligent maintenance across substations.

[0130] Example 4 Based on the same inventive concept as Embodiment 1, this embodiment also proposes an electronic device, which includes: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the digital twin-based intelligent substation maintenance method of the present invention.

[0131] Furthermore, to achieve the above objectives, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the digital twin-based intelligent substation maintenance method of the present invention.

[0132] The following is a detailed introduction to the various components of the electronic device: In this context, the processor is the control center of the electronic device. It can be a single processor or a collective term for multiple processing elements. For example, a processor can be one or more central processing units (CPUs), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0133] Alternatively, the processor can perform various functions of the electronic device by running or executing software programs stored in memory and by calling data stored in memory.

[0134] The memory is used to store the software program that executes the solution of the present invention, and the execution is controlled by the processor. The specific implementation method can be referred to the above method embodiment, which will not be repeated here.

[0135] Optionally, the memory can be read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions, random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory can be integrated with the processor or exist independently and coupled to the processor through the interface circuit of the electronic device; the embodiments of the present invention do not specifically limit this.

[0136] A transceiver is used to communicate with network devices or with terminal devices.

[0137] Optionally, the transceiver may include a receiver and a transmitter. The receiver is used to implement the receiving function, and the transmitter is used to implement the sending function.

[0138] Optionally, the transceiver can be integrated with the processor or exist independently and coupled to the processor through the router's interface circuit. This embodiment of the invention does not specifically limit this.

[0139] Furthermore, the technical effects of the electronic device can be referred to the technical effects of the data transmission method in the above method embodiments, and will not be repeated here.

[0140] It should be understood that the processor in the embodiments of the present invention can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0141] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDRSDRAM), enhanced synchronous DRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct memory bus RAM (DRRAM).

[0142] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0143] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0144] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0145] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0146] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

Claims

1. A substation intelligent maintenance method based on digital twins, characterized in that, include: Obtain the operating parameters corresponding to multiple substations, establish digital twin models corresponding to multiple substations based on the operating parameters, and use each digital twin model as a node to form a digital twin network; When the operating parameters of any of the substations exceed the preset alarm threshold, the node corresponding to the substation is identified as the target node. Based on the digital twin network, nodes whose electrical distance to the target node is less than a preset distance threshold are obtained as associated nodes; or, nodes that have the same fault root cause as the target node in historical fault records are obtained as associated nodes, and the digital twin model corresponding to the associated node is obtained as the associated digital twin model. Obtain the abnormal operation characteristics of the target digital twin model corresponding to the target node during the alarm period, and the associated operation characteristics of the associated digital twin model during the alarm period; The abnormal operation characteristics and the associated operation characteristics are input into a preset fault propagation analysis model, wherein the fault propagation analysis model is constructed based on a graph neural network model, and the network nodes of the graph neural network model correspond to the nodes in the digital twin network; The edge weights of the graph neural network are determined based on the historical fault propagation probability between the substations corresponding to each node. Based on the fault propagation analysis model, the location of the fault source corresponding to the abnormal operation characteristics and the propagation path of the fault source location through the digital twin network affecting other nodes are determined. Based on the location of the fault source, the propagation path, and the target digital twin model, a maintenance strategy is determined, and the maintenance strategy is sent to the execution terminal corresponding to the location of the fault source.

2. The intelligent substation maintenance method based on digital twins according to claim 1, characterized in that: Specific methods for constructing a digital twin network include: Obtain electrical parameters, environmental parameters, and equipment status parameters corresponding to multiple substations, and align the electrical parameters, environmental parameters, and equipment status parameters according to timestamps to form a time-series operation dataset corresponding to each substation; A device model is constructed based on the aforementioned time-series runtime dataset; Based on the physical connection relationships between the devices in each substation, the device models are merged to form the digital twin model corresponding to the substation.

3. The intelligent substation maintenance method based on digital twins according to claim 1, characterized in that: The abnormal operation characteristics include electrical quantity abrupt change characteristics, non-electrical quantity exceeding limit characteristics, and timing correlation deviation characteristics between devices; The associated operational characteristics include electrical quantity fluctuation characteristics, non-electrical quantity change rate characteristics, and inter-device timing synchronization characteristics.

4. The intelligent substation maintenance method based on digital twins according to claim 1, characterized in that: The method for constructing the input features of the fault propagation analysis model includes: Align the target runtime sequence set with the associated runtime sequence set, obtain the runtime parameter difference values ​​between the target digital twin model and each associated digital twin model at the same time, and generate a difference feature matrix based on the runtime parameter difference values; The difference feature matrix participates in message passing as the edge features of the corresponding edge in the graph neural network within the fault propagation analysis model; The abnormal operation characteristics, the associated operation characteristics, and the difference feature matrix are used together as input features of the fault propagation analysis model.

5. The intelligent substation maintenance method based on digital twins according to claim 4, characterized in that: During message transmission, for a directed edge from the source node to the target node, the fault propagation analysis model reads the difference feature vector in the difference feature matrix corresponding to the edge, concatenates the difference feature vector with the feature vector of the source node or multiplies it element by element, and then multiplies it by the edge weight of the edge to generate a message to be transmitted to the target node.

6. The intelligent substation maintenance method based on digital twins according to claim 1, characterized in that: The method for determining the edge weights of the graph neural network includes: Obtain historical fault records between the substations within a preset historical time period. The historical fault records include the fault source substation, the substations through which the fault propagates, and the substations ultimately affected by the fault. Obtain the first substation and the second substation corresponding to any two nodes in the digital twin network. Based on the total number of times the first substation is the source of the fault in the historical fault records and the number of times it propagates to the second substation, use the ratio of the two as the historical fault propagation probability, and determine the edge weights of the graph neural network accordingly.

7. The intelligent substation maintenance method based on digital twin according to claim 1, characterized in that: Based on the fault propagation analysis model, the location of the fault source corresponding to the abnormal operation characteristics and the propagation path of the fault source location through the digital twin network affecting other nodes are determined, including: Output each corresponding fault probability vector based on the edge weights; The target node and each of the associated nodes are sorted from high to low according to the fault probability vector, and the node at the top of the sort is selected as the fault source location. Starting from the location of the fault source, trace back along the edge weight direction to determine the propagation path to other nodes; The path in the propagation path whose edge weights are continuous and greater than a preset propagation threshold is determined as the propagation path, including the fault source node, at least one intermediate propagation node, and the terminal node.

8. The intelligent substation maintenance method based on digital twin according to claim 1, characterized in that: Based on the fault source location, the propagation path, and the target digital twin model, a maintenance strategy is determined, and the maintenance strategy is sent to the execution terminal corresponding to the fault source location, including: Based on the location of the fault source, the corresponding device model is determined from the target digital twin model, and the device type, device identifier, and current operating status of the device model are obtained. Based on the propagation path, determine the set of nodes affected by the location of the fault source, and obtain the affected devices and the degree of impact corresponding to each node in the set; The fault type and severity at the location of the fault source are obtained, and at least one candidate maintenance plan is generated through a preset maintenance strategy generation model. The candidate maintenance plan includes the target equipment to be maintained, the maintenance operation sequence, the required spare parts, and the estimated maintenance time.

9. A digital twin-based intelligent substation maintenance system utilizing the method described in any one of claims 1-8, characterized in that, The system includes a control terminal and execution terminals corresponding to multiple substations, and the system is configured as follows: Based on the execution terminal, the operating parameters corresponding to multiple substations are obtained, and digital twin models corresponding to multiple substations are established based on the operating parameters. Each digital twin model is used as a node to form a digital twin network. When the operating parameters of any of the substations exceed the preset alarm threshold, the node corresponding to the substation is identified as the target node. Based on the digital twin network, nodes whose electrical distance to the target node is less than a preset distance threshold are obtained as associated nodes, or nodes that have the same fault root cause as the target node in historical fault records are obtained as associated nodes, and the digital twin model corresponding to the associated node is obtained as the associated digital twin model. Obtain the abnormal operation characteristics of the target digital twin model corresponding to the target node during the alarm period, and the associated operation characteristics of the associated digital twin model during the alarm period; The abnormal operation characteristics and the associated operation characteristics are input into a preset fault propagation analysis model, wherein the fault propagation analysis model is based on a graph neural network model, and the network nodes of the graph neural network model correspond to the nodes in the digital twin network; The edge weights of the graph neural network are determined based on the historical fault propagation probability between the substations corresponding to each node. Based on the fault propagation analysis model, the location of the fault source corresponding to the abnormal operation characteristics and the propagation path of the fault source location through the digital twin network affecting other nodes are determined. Based on the location of the fault source, the propagation path, and the target digital twin model, a maintenance strategy is determined, and the maintenance strategy is sent to the execution terminal corresponding to the location of the fault source.

10. An electronic device, characterized in that, The electronic device includes: At least one processor; And, a memory communicatively connected to at least one of the processors; The memory stores instructions that can be executed by at least one of the processors, which are executed by at least one of the processors to enable the at least one of the processors to perform a digital twin-based intelligent substation maintenance method as described in any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements a digital twin-based intelligent substation maintenance method as described in any one of claims 1-8.

Images