Digital twinning-based transformer substation equipment full life cycle health management method and device

Abstract

The invention provides a transformer substation equipment full life cycle health management method and device based on digital twinning. The invention belongs to the technical field of digital twinning. The method comprises the following steps: driving a digital twinborn model to update a simulation state by using multi-source real-time operation data, and outputting a twinborn simulation value of substation equipment; performing deviation calculation and anomaly judgment on the twinborn simulation value and the physical observation value, performing graph reasoning on the fault knowledge graph when the twinborn simulation value and the physical observation value are abnormal, and outputting a fault type and a fault positioning result; a physical constraint deep learning prediction model obtained through pre-training is adopted to predict the health degree evolution trend of the substation equipment and output the residual life of the substation equipment; adopting a reinforcement learning strategy model to output an operation and maintenance strategy; and after the operation and maintenance strategy is executed, feedback data formed by an operation and maintenance execution result is obtained, and parameter correction and model evolution are performed on the digital twin model based on the feedback data. According to the method, the operation and maintenance cost can be reduced, and the operation safety level and the whole-process traceability of the transformer substation are improved.

Description

Technical Field

[0001] This invention belongs to the field of digital twin technology, and in particular relates to a method and device for full life-cycle health management of substation equipment based on digital twins. Background Technology

[0002] With the continuous advancement of new power systems and digital grids, substations, as key nodes in the power system for the collection, transformation, and distribution of electrical energy, have their core equipment, such as main transformers, GIS (Gas Insulated Switchgear), and circuit breakers, widely used in power transmission and distribution networks. Their operational status directly affects the safety and reliability of the power grid. In practical engineering, substation equipment is typically equipped with monitoring systems (SCADA, Supervisory Control and Data Acquisition), equipment management systems (PMS, Power Management System), and online monitoring devices to support equipment operation management and maintenance decisions.

[0003] In the context of the aforementioned applications, the prominent challenges faced by substation equipment operation and maintenance lie in the difficulty of data integration and timely identification of anomalies. On the one hand, different business systems lack unified standards in data format, time base, and spatial description, making it difficult to correlate and analyze operational data, ledger data, and sensing data, thus hindering the formation of a holistic understanding of equipment status. On the other hand, equipment anomalies and faults often exhibit subtle changes over long periods, and single monitoring indicators or threshold alarms are insufficient to reflect gradual degradation in a timely manner. Diagnostic conclusions still rely heavily on experience-based judgments, limiting their stability and interpretability.

[0004] To address the aforementioned issues, existing engineering practices typically improve the situation by adding online monitoring points, introducing data analysis models, or expert rule systems. For example, trend analysis can be conducted based on multi-source state variables, or statistical analysis and machine learning methods can be used to model operational data to assist in fault diagnosis and life assessment. These methods have improved the state awareness capability to some extent and mitigated the problems of low inspection efficiency and delayed response.

[0005] However, from the perspective of overall application effectiveness, existing improvements still have shortcomings in closed-loop management throughout the entire lifecycle: status monitoring is easily affected by data quality and operating condition fluctuations, resulting in insufficient consistency and reliability of results; fault identification and location are not traceable in complex coupled scenarios; and the linkage between health and remaining life assessment and operation and maintenance decisions is not close, making it difficult to form a collaborative optimization link from diagnosis, prediction, decision-making, execution to feedback correction, which in turn affects the pertinence and feasibility of the strategy. Summary of the Invention

[0006] The purpose of this invention is to provide a method and device for full lifecycle health management of substation equipment based on digital twins, which can improve the accuracy and timeliness of substation equipment status monitoring, enhance fault identification and location capabilities, improve the reliability of equipment health and remaining life prediction, and ensure the pertinence and feasibility of operation and maintenance strategy formulation and execution, thereby reducing operation and maintenance costs and improving the safety level and full-process traceability of substation operation.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, this invention provides a method for full lifecycle health management of substation equipment based on digital twins, comprising: The system acquires multi-source real-time operating data of substation equipment, and based on a pre-built digital twin model, uses the multi-source real-time operating data to drive the digital twin model to update the simulation state, and outputs the twin simulation value of at least one target state parameter of the substation equipment. The deviation between the twin simulation value of the target state parameter and the physical observation value of the target state parameter is calculated and anomaly is determined. When an anomaly is determined, an anomaly feature vector is extracted and input into the fault knowledge graph for graph reasoning, and the fault type and fault location result are output. Based on the fault type, fault location results, and historical operation and maintenance data, a pre-trained physical constraint deep learning prediction model is used to predict the health evolution trend of the substation equipment and output the remaining life of the substation equipment. The physical constraint deep learning prediction model includes a deep learning network for time-series prediction of the health evolution trend and physical constraint terms for constraining the prediction results to conform to the aging mechanism. Based on the remaining lifetime and power grid operation safety constraints, a pre-trained reinforcement learning strategy model is used to output an operation and maintenance strategy, which includes at least one of maintenance, load reduction operation, enhanced monitoring, or maintaining the status quo. After executing the operation and maintenance strategy, the feedback data formed by the operation and maintenance execution results is obtained, and the parameters of the digital twin model are corrected and the model is evolved based on the feedback data.

[0008] Furthermore, the construction steps of the digital twin model include: Geometric data of the substation equipment is obtained based on laser point cloud scanning, BIM modeling, or design drawings to construct a geometric twin; Based on the finite element analysis theory, a multi-physics simulation kernel integrating electromagnetics, thermodynamics, and fluid mechanics is used to construct a physical twin. Based on an expert knowledge base, equipment operation procedures, or finite state machine, the state transition rules of the substation equipment under different operating conditions are defined to construct a rule twin. The geometric twin, physical twin, and rule twin are integrated through a twin engine and a real-time data-driven interface is provided to use the multi-source real-time running data to drive the digital twin model to update the simulation state and output the twin simulation value of at least one target state parameter.

[0009] Furthermore, the step of acquiring multi-source real-time operating data of substation equipment includes: The system receives communication messages from the sensor array and the SCADA system respectively, and performs protocol parsing and field extraction on the communication messages to obtain the measurement point values ​​and timestamps corresponding to the sensor array and the SCADA system. It also performs ledger record reading and field extraction on the information from the PMS ledger. The obtained measurement point values ​​and timestamps are synchronized and aligned in time and space. The time synchronization includes synchronizing the timestamps of heterogeneous data using the IEEE 1588 PTP protocol, and the space-time alignment includes achieving spatial alignment using GIS and BIM coordinate mapping. Data cleaning and reduction are performed on the synchronized and aligned data. The data cleaning includes using Kalman filtering and wavelet transform to remove noise and outliers, and filling in missing data. The reduction includes converting multi-source data into a unified JSON or Protobuf format.

[0010] Furthermore, the steps for calculating deviation and determining anomalies include: Calculate the physical observation values ​​using the following formula. With twin simulation values The residuals between :

[0011] Based on the following formula, the sliding window statistical method is used to calculate the statistics of the residuals within the window and construct a dynamic threshold. :

[0012] in, , These are weighting coefficients. The historical mean of the residuals within the window. The historical standard deviation of the residuals within the window; When the residual is greater than the dynamic threshold, it is determined to be abnormal, and the extraction of the abnormal feature vector and the graph inference are triggered.

[0013] Furthermore, the construction steps of the physical constraint deep learning prediction model include: Based on the historical operation and maintenance data, historical health index HI, cumulative aging rate and environmental stress factor are extracted as input features to construct the deep learning network for time-series prediction of the health evolution trend. The historical operation and maintenance data includes historical operation records from the SCADA system and factory parameters and maintenance records from the PMS ledger. The deep learning network includes a Transformer encoder and an LSTM network. The physical constraint term is introduced during model training, and the following loss function is used:

[0014] in, This is a physical constraint term for thermal aging constructed based on Arrhenius's law. For data-driven loss terms, These are the weighting coefficients; The step of outputting the remaining lifespan of the substation equipment includes: The historical feature sequence is input into the physical constraint deep learning prediction model, and the physical constraint deep learning prediction model outputs the health curve and its confidence interval for future time. The time point at which the health curve reaches the failure threshold is determined based on a preset failure threshold, and the time difference between the predicted start time and the time point is used as the remaining lifespan of the substation equipment.

[0015] Furthermore, the steps for constructing the reinforcement learning policy model include: Under the condition of satisfying the power grid operation safety constraints, a near-end policy optimization (PPO) algorithm is used to train a reinforcement learning agent. The state space is defined to include at least: equipment health, power grid load level, and spare parts inventory, serving as the input to the reinforcement learning agent. The action space is defined to include at least: maintenance, load reduction operation, enhanced monitoring, and maintaining the status quo, serving as the output to the reinforcement learning agent. During training, policy parameters are updated based on experience samples formed by the interaction between the reinforcement learning agent and the environment to maximize long-term cumulative rewards. The following reward function is used during training:

[0016] in, For reliability benefits, For maintenance costs, , The weighting coefficients are used to determine the operation and maintenance strategy. When outputting the operation and maintenance strategy, the strategy is output by the reinforcement learning strategy model based on the remaining lifetime and the power grid operation safety constraints.

[0017] Furthermore, the parameter correction and model evolution of the digital twin model based on the feedback data includes: After executing the aforementioned operation and maintenance strategy, maintenance records, handling process records, and post-maintenance measured data are acquired as feedback data. This feedback data is then used as observation data to perform Bayesian updates on the model parameters in the digital twin model used to calculate the target state parameters, thereby obtaining the posterior parameter distribution of the model parameters. The Bayesian update satisfies the following formula:

[0018] in, Corresponding to the observation data, Corresponding to the model parameters, The prior parameter distribution of the model parameters before the update. Let be the likelihood function of the observed data given the model parameters; The feedback data, the fault type and fault location results, and the execution records of the operation and maintenance strategy are hashed and stored on the blockchain.

[0019] In a second aspect, the present invention provides a digital twin-based substation equipment lifecycle health management device, comprising: The twin simulation module is used to acquire multi-source real-time operating data of substation equipment, and based on the pre-built digital twin model, use the multi-source real-time operating data to drive the digital twin model to update the simulation state, and output the twin simulation value of at least one target state parameter of the substation equipment. The fault determination module is used to calculate the deviation and determine the anomaly between the twin simulation value of the target state parameter and the physical observation value of the target state parameter. When an anomaly is determined, the module extracts the anomaly feature vector and inputs the anomaly feature vector into the fault knowledge graph for graph reasoning, and outputs the fault type and fault location result. The lifespan prediction module is used to predict the health evolution trend of the substation equipment and output the remaining lifespan of the substation equipment based on the fault type, fault location results and historical operation and maintenance data, using a pre-trained physical constraint deep learning prediction model. The physical constraint deep learning prediction model includes a deep learning network for time-series prediction of the health evolution trend and physical constraint terms for constraining the prediction results to conform to the aging mechanism. The strategy output module is used to output an operation and maintenance strategy based on the remaining lifetime and power grid operation safety constraints, using a pre-trained reinforcement learning strategy model. The operation and maintenance strategy includes at least one of maintenance, load reduction operation, enhanced monitoring, or maintaining the status quo. The model evolution module is used to obtain feedback data formed by the operation and maintenance execution results after the operation and maintenance strategy is executed, and to perform parameter correction and model evolution on the digital twin model based on the feedback data.

[0020] In a third aspect, the present invention provides an electronic device including a processor and a memory, the processor being configured to execute a computer program stored in the memory to implement a digital twin-based method for full lifecycle health management of substation equipment.

[0021] In a fourth aspect, the present invention provides a computer-readable storage medium storing at least one instruction, which, when executed by a processor, implements a digital twin-based method for full lifecycle health management of substation equipment.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention provides a digital twin-based substation equipment lifecycle health management device, electronic device, and computer-readable storage medium, which also solves the problems mentioned in the background section.

[0023] 1. This invention constructs a digital twin-based method for the full lifecycle health management of substation equipment, achieving continuous perception, unified modeling, and dynamic evaluation of equipment status. This significantly improves the accuracy and consistency of equipment status monitoring under complex operating conditions. By unifying and associating multi-source operating data, historical ledger information, and the digital twin model, the health status, degradation trends, and potential risks of the equipment are continuously presented within the same framework, thereby enhancing the stability and traceability of fault identification and location results. Simultaneously, based on virtual-real interactive status assessment and evolutionary analysis, health assessment and remaining life prediction are more closely aligned with the actual operating status of the equipment, effectively reducing the risk of misjudgment caused by single indicators or experience-based judgments. Furthermore, the method supports the formation, execution, and feedback correction of operation and maintenance strategies, improving the pertinence and operability of strategy formulation and implementation, reducing overall operation and maintenance costs, minimizing the risk of unplanned power outages, and enhancing the safety and overall management level of substation equipment operation.

[0024] 2. By introducing unified acquisition of multi-source data, time synchronization, spatial alignment, and a digital twin model construction and virtual-real interaction mechanism, this invention can effectively improve the precision of complex equipment condition assessment. Multi-source operational data are fused under a unified spatiotemporal benchmark, making different monitoring dimensions comparable and correlated, avoiding state perception biases caused by data fragmentation. Simultaneously, virtual-real comparison and residual analysis based on the digital twin model enable the early identification and quantification of subtle changes in equipment operating status. This approach helps enhance the sensitivity and stability of abnormal state detection, improves the reliability of fault location results under complex operating conditions, and provides a consistent data and model foundation for subsequent analysis. This improves the problems of information fragmentation and discrete results in traditional condition assessment, enhancing maintenance personnel's understanding and judgment of the true state of the equipment.

[0025] 3. By integrating lifespan prediction, maintenance strategy generation, and feedback correction into a unified management process, this invention achieves closed-loop optimization across the entire equipment lifecycle. Health status and remaining lifespan prediction results can directly serve as crucial bases for maintenance decisions, shifting maintenance strategies from periodic, experience-based arrangements to differentiated decisions based on condition and risk, thereby improving resource allocation efficiency and reducing unnecessary maintenance interventions. Simultaneously, maintenance execution results and actual operational feedback are continuously fed back into the management system to correct assessment results and update subsequent decision-making basis, forming a continuously evolving management mechanism. This closed-loop process helps improve the feasibility and long-term stability of maintenance strategy execution, avoids the accumulation of risks caused by over-maintenance or neglect, and ultimately achieves a comprehensive improvement in equipment reliability, economy, and operational safety. Attached Figure Description

[0026] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart illustrating the digital twin-based substation equipment lifecycle health management method according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the substation equipment full lifecycle health management method according to an embodiment of the present invention. Figure 3 This is a flowchart illustrating the data processing and alignment process in an embodiment of the present invention. Figure 4 This is a flowchart illustrating the specific process of constructing a digital twin model according to an embodiment of the present invention. Figure 5 This is a flowchart illustrating the virtual-real interaction and fault diagnosis in an embodiment of the present invention. Figure 6 This is a diagram illustrating the lifetime prediction algorithm architecture of an embodiment of the present invention. Figure 7 This is a flowchart illustrating the strategy optimization and closed-loop feedback process in an embodiment of the present invention. Figure 8 This is a system architecture and data flow diagram of the substation equipment full life cycle health management method according to an embodiment of the present invention; Figure 9 This is a structural block diagram of the substation equipment full life cycle health management device according to an embodiment of the present invention; Figure 10 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0027] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0028] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0029] Example 1 This embodiment provides a method and device for full lifecycle health management of substation equipment based on digital twins, forming a closed-loop process of perception, simulation, diagnosis, prediction, optimization, and evolution. Figure 1 and Figure 2 As shown, the method generally includes steps S1 to S6.

[0030] In step S1, multi-source real-time operating data of substation equipment is acquired, forming a data foundation for subsequent twin simulation, diagnosis, prediction, and strategy optimization. Specifically, a multi-dimensional sensing network is constructed to collect real-time operating status data, environmental data, and multimedia data of substation equipment. Combined with historical operating data from the SCADA system and PMS ledger information, high-quality, unified-format panoramic equipment data is obtained through cleaning, reduction, and spatiotemporal alignment of multi-source heterogeneous data. The multi-dimensional sensing network may include: UHF partial discharge sensors, infrared thermal imaging probes, vibration sensors, online oil chromatography monitoring devices, and distributed fiber optic temperature measurement systems, etc., to form a multi-dimensional observation of equipment operating status. In addition to the above-mentioned online monitoring data, the data sources also include traditional SCADA operating quantities such as voltage, current, power, and load, as well as full lifecycle information such as factory parameters and maintenance records in the PMS ledger.

[0031] The steps for acquiring multi-source real-time operating data of substation equipment include: receiving communication messages from sensor arrays and SCADA systems, and performing protocol parsing and field extraction on the communication messages to obtain the measurement point values ​​and timestamps of each data source. The protocol parsing includes parsing IEC61850, MQTT, and Modbus communication protocols. The information from the PMS ledger is read from ledger records and its fields are extracted to obtain ledger field data. Time synchronization and spatiotemporal alignment are performed on the obtained measurement point values ​​and timestamps. Time synchronization includes synchronizing the timestamps of heterogeneous data using the IEEE 1588 PTP protocol, and spatiotemporal alignment includes achieving spatial alignment using GIS and BIM coordinate mapping. Data cleaning and reduction are performed on the synchronized and aligned data. Data cleaning includes using Kalman filtering and wavelet transform to remove noise and outliers and fill in missing data. Reduction includes converting the multi-source data into a unified JSON or Protobuf format.

[0032] To achieve unified access and processing of multi-source data, such as Figure 3 As shown, the data processing chain can be divided into data source, edge computing layer, and data preprocessing stages: At the edge computing layer, data from sensor arrays and SCADA systems is aggregated via an edge gateway, and information from the PMS ledger is read and fields extracted to complete the ledger information access. In the protocol parsing stage, the edge computing gateway parses various communication protocols such as IEC61850, MQTT, and Modbus, and can convert proprietary protocols from different manufacturers into a unified JSON or Protobuf format by deploying a protocol parsing container on the edge gateway. In the time synchronization stage, the IEEE 1588 PTP protocol is used for microsecond-level time synchronization of heterogeneous sensors. The PTP protocol is used to connect the timestamps of each sensor, with errors controlled at the microsecond level to ensure time consistency in subsequent multiphysics coupling analysis. In the data preprocessing stage, Kalman filtering and wavelet transform algorithms are used to remove data noise and outliers and fill in missing data. In the spatiotemporal alignment stage, the distributed sensor data is mapped to the unified three-dimensional coordinate system of the equipment using the GIS spatial coordinate system (spatial alignment can be achieved by combining GIS and BIM coordinate mapping). Then, multi-source fusion is performed, and finally a high-quality panoramic data stream in a unified format is output to support subsequent twin simulation, residual diagnosis, life prediction and strategy optimization.

[0033] In step S2, based on the pre-constructed digital twin model, multi-source real-time operational data is used to drive the digital twin model to update its simulation state, outputting the twin simulation value of at least one target state parameter of the substation equipment. Specifically, based on the equipment's multi-source geometric data, material physical properties, and operating mechanisms, a high-fidelity three-dimensional digital twin model of the substation equipment is constructed, establishing a multi-dimensional fusion mapping relationship of geometry, physics, and rules. Real-time panoramic data is used to drive the twin model for dynamic simulation, achieving millisecond-level synchronization and virtual-real interaction between the physical entity and the virtual model.

[0034] The steps for constructing a digital twin model include: acquiring the external geometric data of substation equipment using laser point cloud scanning and BIM modeling technology, or fusing laser point cloud data with design drawings to form a geometric model to construct a geometric twin; based on finite element analysis theory, integrating a multi-physics simulation kernel of electromagnetics, thermodynamics, and fluid mechanics, constructing a multi-physics coupling model of the internal electric field, thermal field, fluid field, and force field of the equipment to construct a physical twin; defining the state transition rules of the substation equipment under different operating conditions based on an expert knowledge base, equipment operating procedures, or a finite state machine to construct a rule twin; using a data-driven engine to map real-time data streams to model parameters, driving the virtual model to evolve synchronously, and integrating the geometric twin, physical twin, and rule twin through a Unity3D or UE5 engine; realizing visualization rendering and real-time interaction through a real-time data-driven interface, and outputting a high-fidelity dynamic digital twin.

[0035] To enable the twin model to possess a comprehensive capability of visual appearance, computational mechanism, and constrained logic, it can be implemented through the following specific process: S201, Data layer modeling, uses laser point cloud scanning and BIM modeling technology to obtain high-precision geometric data of the equipment's appearance, or integrates laser point cloud data with design drawings to form a high-precision geometric model and construct a geometric twin; S202, physical layer modeling, is based on finite element analysis (FEA) theory and integrates multi-physics simulation kernels such as electromagnetics, thermodynamics, and fluid mechanics to construct a multi-physics coupling model of the electric field, thermal field, fluid field and force field inside the equipment, and build a physical twin. S203, Behavior and Rule Layer Modeling, defines the equipment operation logic and associated constraints based on expert knowledge base and equipment operation procedures, or defines the state transition rules of the equipment under different operating conditions based on finite state machine (FSM), and constructs a rule twin to express the state machine and logical constraints; S204 utilizes a data-driven engine to map real-time data streams to model parameters, driving the synchronous evolution of virtual models. It integrates geometric twins, physical twins, and rule twins through the Unity3D or UE5 engine, and achieves visualization rendering and real-time interaction through a real-time data-driven interface.

[0036] Figure 4 The diagram illustrates the multi-source basic data of the equipment, the multi-dimensional modeling process, and the fusion relationship of the twin engine, consistent with the steps described above: The multi-source basic data of the equipment corresponds to geometric data (point cloud, CAD), physical properties (material, electromagnetic, thermal), and operating procedures (expert library, logic), respectively. Through the multi-dimensional modeling process, geometric twins (appearance, structure), physical twins (multiphysics coupling kernel), and rule twins (state machine, constraints) are formed. The real-time running data stream is accessed through the real-time data-driven interface of the twin engine, and the visualization rendering and interaction are completed by the Unity3D or UE5 integrated engine, outputting a high-fidelity dynamic digital twin. The digital twin is not only a three-dimensional model of the appearance, but also a simulation containing a physical kernel. For example, for a transformer, an electro-thermal-fluid coupled field model can be constructed using COMSOL Multiphysics, and after its order reduction (ROM), it can be embedded into the real-time running digital twin engine. When real-time current data is input, the twin can calculate the hot spot temperature distribution inside the winding in real time, rather than relying solely on the top oil temperature for estimation.

[0037] In step S3, the deviation between the twin simulation value and the physical observation value of the target state parameter is calculated and anomaly is determined. When an anomaly is determined, an anomaly feature vector is extracted and input into the fault knowledge graph for graph inference, outputting the fault type and fault location result. In step S3, the state of the digital twin model is scanned according to a set period, the residual vector between the physical observation value and the twin simulation value is calculated, and the deviation of the equipment operating state is identified using statistical process control and graph inference algorithms to locate the fault source and analyze the fault mechanism. This process is the core link of fault detection. The system compares the measured values ​​(such as physical observations such as oil temperature measured by sensors) with the simulated values ​​(theoretical values ​​calculated by the twin based on the load and mechanism model) in real time. Under normal circumstances, the two should be highly consistent; if a significant residual occurs and environmental factors are excluded, it indicates that the internal physical parameters of the equipment have changed (such as radiator blockage or inter-turn short circuit). At this time, the most likely fault cause causing the residual can be inferred by combining the fault knowledge graph.

[0038] The steps for deviation calculation and anomaly determination include: calculating the residual vector between physical observation values ​​and twin simulation values, and calculating the residual mean and residual variance of the residual sequence based on the sliding window statistical method to construct a dynamic adaptive threshold. When the residual satisfies the dynamic adaptive threshold and the number of sampling points that continuously satisfy the threshold is not less than a set number of frames, it is determined as a potential anomaly. Anomaly feature vectors are extracted and input into the fault knowledge graph for graph reasoning to output the fault type and fault location results.

[0039] The residual analysis method can be implemented as follows: establish a baseline twin model under normal operating conditions, and calculate the physical observations. Twin predictions The residual between physical observations and twin predictions is defined as follows:

[0040] in: Indicates at time Physical observations of the equipment obtained by sensors; Indicates at the same time The corresponding predicted value obtained from simulation calculations using a digital twin model; Indicates at time The residual magnitude is used to characterize the degree of deviation between the physical entity state and the twin model state. A sliding window statistical method is employed, whereby the residual sequence is analyzed within a sliding time window. Perform statistical analysis, calculate the mean and variance of the residuals within the window, and denot them as: mean residual residual variance The sliding window consists of the most recent consecutive... It consists of 1 sampling point. The window length parameter. A dynamic adaptive threshold is constructed based on historical statistics:

[0041] in: Indicates a dynamic adaptive threshold. This represents the historical mean of the residuals within the sliding window. This represents the historical standard deviation of the residuals within the sliding window. , These are weighting coefficients used to adjust the weights of the mean and fluctuation terms in the threshold calculation. When the real-time residual satisfies... Furthermore, the number of sampling points that continuously meet this condition is no less than the set number of frames. When this occurs, it is identified as a potential anomaly, and knowledge graph reasoning is triggered to perform root cause analysis of the failure.

[0042] In one implementation, physical observations and twin predictions can constitute a multidimensional state vector. The residual vector can be further used to calculate the Euclidean distance, and the anomaly threshold is dynamically updated by combining the statistical characteristics (mean, variance) of the sliding time window. Subsequently, anomaly feature vectors are extracted and input into the graph inference engine of the substation fault knowledge graph to locate the faulty component (such as windings, cores, bushings). Figure 5 As shown, the residual detection and fault location process may include: acquiring physical observation values ​​after monitoring begins. With twin simulation values Calculate the residual vector Calculate the dynamic adaptive threshold Determine the residual Is it greater than the threshold? If not, the status is considered normal and monitoring continues; if yes, it is determined to be a potential anomaly, anomaly feature vectors are extracted, a fault knowledge graph is invoked for graph reasoning and root cause analysis, and the fault type and location are output. The sliding window length can be defined as the nearest... A time window consisting of sampling points is defined, and the set frame number is defined as continuously satisfying... Minimum number of consecutive points To suppress false triggering caused by single-point noise; among which and As a configurable parameter for engineering, it can be configured according to device type and sampling frequency without changing the core mechanism of building thresholds based on window statistics and triggering based on the number of continuous frames.

[0043] In step S4, based on the fault type, fault location results, and historical operation and maintenance data, a pre-trained physical constraint deep learning prediction model is used to predict the health evolution trend of substation equipment and output the remaining life of the substation equipment. Specifically, based on the fault diagnosis results and historical operation and maintenance data, a life prediction architecture integrating a multi-physics coupling model and deep learning is constructed to predict the performance degradation trajectory and remaining life (RUL) of the equipment within a future time window, and output the health evolution trend and its confidence interval. This process can employ a Physically Guided Neural Network (PGNN), where a physical model provides the basic aging trend (such as the curve of the degree of polymerization (DP) of insulating paper changing with temperature), and a deep learning model captures the impact of nonlinear factors such as environmental fluctuations and sudden shocks on life, thereby outputting a predicted RUL value with a confidence interval.

[0044] The construction steps of the physical constraint deep learning prediction model include: extracting historical health index (HI), cumulative aging rate, and environmental stress factor as input features based on historical operation and maintenance data, constructing a deep learning network for time series prediction, and introducing physical constraint terms during training to form a physical constraint deep learning prediction model. The historical operation and maintenance data includes historical operation records from the SCADA system and factory parameters and maintenance records from the PMS ledger. The deep learning network includes a Transformer encoder and an LSTM network. The physical constraint terms are constructed based on Arrhenius's law.

[0045] like Figure 6 As shown, step S4 specifically includes: S401, feature engineering, extracting the historical health index (HI), cumulative aging rate and environmental stress factor of the equipment as feature vectors. The health factor may include gas growth rate, hot spot temperature rise rate, etc. S402, construct a hybrid temporal prediction model based on Transformer and LSTM, use multi-head attention mechanism to capture long-order dependencies and use temporal network to represent the health evolution; S403 introduces a physically guided joint loss function during the training of the lifetime prediction model:

[0046] in: Represents the total loss function; This represents the prediction error loss term based on historical sample data; This represents the physical consistency loss term based on physical mechanism constraints; These are the physical constraint weighting coefficients, used to balance data fitting accuracy and physical consistency constraints. Wherein... It can be built based on the Arrhenius Law to constrain the aging evolution trend of the model output, so that the prediction results conform to the thermal aging mechanism of insulating materials with temperature change, and avoid the deep learning model from producing prediction results that violate physical laws. S404 outputs the device's health score and confidence interval at each future time step to determine the remaining lifetime (RUL), which is the time difference between when the health curve reaches the failure threshold. The input layer of the lifetime prediction architecture can include the cumulative aging rate, historical health index (HI), and environmental stress factor; the model layer includes a Transformer encoder and an LSTM network, and incorporates physical constraints and a loss function based on Arrhenius's law. The output layer outputs the future health curve, remaining lifetime (RUL), and confidence interval.

[0047] In step S5, based on the remaining lifetime and grid operation safety constraints, the operation and maintenance strategy is output using a pre-trained reinforcement learning strategy model. In step S5, based on the remaining lifetime prediction results and grid operation safety constraints, and with operation and maintenance cost and reliability as objective functions, the optimal differentiated operation and maintenance strategy or emergency response plan is generated using a deep reinforcement learning (DRL) algorithm.

[0048] The steps for constructing the reinforcement learning policy model include: training the reinforcement learning agent using the near-end policy optimization (PPO) algorithm under the condition of satisfying the power grid operation safety constraints. The state space is defined to include at least: equipment health, power grid load level, and spare parts inventory, serving as the input to the reinforcement learning agent. The action space is defined to include at least: maintenance, load reduction operation, enhanced monitoring, and maintaining the status quo, serving as the output of the reinforcement learning agent. During training, the policy parameters are updated based on experience samples formed by the interaction between the reinforcement learning agent and the environment to maximize long-term cumulative rewards. The following reward function is used during training:

[0049] in, For reliability benefits, For maintenance costs, , The weighting coefficients are used to determine the operation and maintenance strategy. When outputting the operation and maintenance strategy, the strategy is output by the reinforcement learning strategy model based on the remaining lifetime and the power grid operation safety constraints.

[0050] In one specific implementation, the strategy generation can employ a proximal policy optimization (PPO) algorithm to train the agent, and, under the premise of satisfying the N-1 safety criterion of the power grid, find the sequence of operation and maintenance actions that maximizes the long-term cumulative rewards (reliability benefits, maintenance costs); specifically, it may include: defining the state space as equipment health, power grid load level, and spare parts inventory; defining the action space as "immediate maintenance," "load reduction operation," "enhanced monitoring," and "maintaining the status quo"; and defining the reward function during the strategy optimization process:

[0051] in: This represents the immediate reward value obtained by the reinforcement learning agent in the current state; Indicators representing the benefits of equipment operational reliability; This indicates the cost of operation and maintenance. , These are weighting coefficients used to adjust the relative importance of reliability benefits and operational costs in the reward function. The aim is to maximize equipment reliability while minimizing operational costs; after training the agent using the PPO algorithm, it outputs the optimal operational action for the current state. Figure 7 As shown, the strategy optimization process on the left uses RUL prediction results and grid constraints as inputs, and Max (reliability) and Min (cost) as rewards or objectives. A deep reinforcement learning agent (PPO) outputs the optimal operation and maintenance strategy, which may include maintenance, load reduction, or component replacement, and is then distributed to the execution of the operation and maintenance operations. In one embodiment, the... and Each item can be normalized according to a preset benchmark, mapped to a dimensionless quantity, and then substituted into the reward function to ensure comparability between reward items; weight coefficients , It can be set as a configurable parameter according to the preferences of the operation and maintenance side.

[0052] In step S6, after executing the operation and maintenance strategy, feedback data generated from the operation and maintenance execution results is obtained, and the digital twin model is modified and evolved based on the feedback data. Specifically, the equipment operating status, fault handling process, and post-maintenance effects are continuously recorded, and the feedback data is input into the digital twin model for physical parameter modification and model evolution, achieving closed-loop management throughout the entire lifecycle. The model evolution may include using Bayesian update methods to modify physical parameters (such as winding DC resistance, insulation coefficient, and oil flow velocity) to ensure that the digital twin evolves synchronously with the aging of the physical entity.

[0053] The parameter correction and model evolution of the digital twin model based on the feedback data includes: after executing the operation and maintenance strategy, obtaining maintenance records, handling process records, and post-maintenance measured data as the feedback data, and using the feedback data as observation data (Data), performing Bayesian updates on the model parameters in the digital twin model used to calculate the twin simulation values ​​of the target state parameters to obtain the posterior parameter distribution of the model parameters. The Bayesian update satisfies the following formula:

[0054] in: This represents the updated posterior distribution of the parameters; This represents the prior distribution of the parameters before the update; This represents the likelihood function of observed data given certain parameters; The posterior distribution is proportional to the product of the prior distribution and the likelihood function. The feedback data, the fault type and fault location results, and the execution records of the operation and maintenance strategy are hashed and stored on the blockchain to ensure data immutability and full-process traceability.

[0055] Combination Figure 7 The closed-loop feedback process on the right collects maintenance feedback data (actual wear or aging measured values) after performing maintenance operations, updates Bayesian parameters and corrects the twin physical parameters, completing model evolution and updating parameters for the next cycle of diagnosis, prediction, and strategy optimization, thus forming a closed-loop iteration of strategy, execution, feedback, update, and re-optimization. In a further implementation, this method can also establish a blockchain notarization mechanism to hash and store key status data, fault diagnosis reports, and maintenance operation records throughout the entire lifecycle on the blockchain, ensuring data immutability and full-process traceability.

[0056] Combination Figure 8The system architecture and data flow shown can be implemented into a corresponding system: The physical entity layer includes substation equipment entities (such as transformer entities) and sensors such as UHF, infrared, and fiber optic sensors; the data transmission and edge layer includes edge computing gateways, PTP time synchronization, and data cleaning and reduction modules, which are used to form standardized data streams from collected data and write them into the panoramic time series database; the digital twin platform layer includes a mechanism model library, a fault knowledge graph, and a virtual-real interaction simulation engine. The virtual-real interaction simulation engine interacts with the panoramic time series database and provides twin simulation and diagnostic reasoning capabilities upwards; the intelligent application layer includes intelligent fault diagnosis, RUL life prediction, operation and maintenance strategy generation, and a 3D visualization interface. Among them, intelligent fault diagnosis can output feedback correction information to the platform side, and operation and maintenance strategy generation can issue control commands or work orders to the field and send the execution results back to the digital twin platform layer to achieve feedback correction and continuous iteration.

[0057] The following is a specific embodiment (taking a 220 kV main transformer as an example) to illustrate the above steps: First, partial discharge signals are collected by an ultra-high frequency (UHF) sensor array installed on the transformer body. Temperature distribution of the transformer body and bushings is collected by distributed fiber optic temperature measurement or fiber optic grating sensors, with sampling frequencies set to 100 MHz (partial discharge) and 1 Hz (temperature). The edge gateway performs wavelet threshold denoising on the data and aligns it with the load current data of the SCADA system at the microsecond level. Then, the system loads the digital twin of the transformer. The twin integrates a winding temperature field calculation model. When the transformer load rate increases from 60% to 90%, the twin calculates in real time that the hot spot temperature of the A-phase low-voltage winding should rise to 85 ℃. Furthermore, the virtual-real interaction module detected that the actual temperature reading of the fiber optic thermometer installed at the same location was 92 ℃, resulting in a residual of +7 ℃. The residual analysis algorithm determined that this deviation exceeded the dynamic threshold (3 ℃) and identified it as abnormal. The system automatically invoked the fault knowledge graph and, combined with oil chromatography data (trace increase in acetylene), deduced that the fault type was "local overheating caused by slight damage to the inter-turn insulation of the low-voltage winding." Correspondingly, the life prediction module extrapolated the future evolution trend from the current fault state. The prediction results showed that if the current 90% load operation was maintained, the degree of polymerization of the insulation paper would decrease to 250 (the failure threshold) within 3 months. Then, the strategy generation module, based on the prediction results and the constraints that the power grid was in the critical period of peak summer demand and that unplanned power outages were not allowed, generated an optimized strategy: "Immediately start the air-cooling system to full-speed mode, limit the load to within 80%, and schedule a power outage for maintenance during the off-peak period 20 days later," and pushed it to the operation and maintenance side for execution. Finally, the maintenance personnel perform the inspection and enter the feedback information "Replace the A-phase low-voltage coil and find carbonization of the insulation paper". The system uses this feedback to correct the physical parameters (such as insulation aging-temperature sensitivity coefficient) in the twin model of this type of transformer, complete the model evolution, and thus enter the closed-loop management of the next cycle.

[0058] Example 2 like Figure 9 As shown, based on the same inventive concept as the above embodiments, the present invention also provides a digital twin-based substation equipment full lifecycle health management device, comprising: The twin simulation module is used to acquire multi-source real-time operating data of substation equipment, and based on the pre-built digital twin model, use the multi-source real-time operating data to drive the digital twin model to update the simulation state, and output the twin simulation value of at least one target state parameter of the substation equipment. The fault determination module is used to calculate the deviation and determine the anomaly between the twin simulation value of the target state parameter and the physical observation value of the target state parameter. When an anomaly is determined, the module extracts the anomaly feature vector and inputs the anomaly feature vector into the fault knowledge graph for graph reasoning, and outputs the fault type and fault location result. The lifespan prediction module is used to predict the health evolution trend of the substation equipment and output the remaining lifespan of the substation equipment based on the fault type, fault location results and historical operation and maintenance data, using a pre-trained physical constraint deep learning prediction model. The physical constraint deep learning prediction model includes a deep learning network for time-series prediction of the health evolution trend and physical constraint terms for constraining the prediction results to conform to the aging mechanism. The strategy output module is used to output an operation and maintenance strategy based on the remaining lifetime and power grid operation safety constraints, using a pre-trained reinforcement learning strategy model. The operation and maintenance strategy includes at least one of maintenance, load reduction operation, enhanced monitoring, or maintaining the status quo. The model evolution module is used to obtain feedback data formed by the operation and maintenance execution results after the operation and maintenance strategy is executed, and to perform parameter correction and model evolution on the digital twin model based on the feedback data.

[0059] Example 3 like Figure 10 As shown, the present invention also provides an electronic device 100 for implementing a digital twin-based method for full lifecycle health management of substation equipment; The electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on at least one processor 102, and at least one communication bus 104.

[0060] The memory 101 can be used to store computer programs 103. The processor 102 implements the digital twin-based substation equipment full life cycle health management method of Embodiment 1 by running or executing the computer programs stored in the memory 101 and calling the data stored in the memory 101.

[0061] The memory 101 may primarily include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the electronic device 100 (such as audio data), etc. In addition, the memory 101 may include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0062] At least one processor 102 may be a Central Processing Unit (CPU), or 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. Processor 102 may be a microprocessor or any conventional processor. Processor 102 is the control center of electronic device 100, connecting various parts of electronic device 100 via various interfaces and lines.

[0063] The memory 101 in the electronic device 100 stores multiple instructions to implement a digital twin-based substation equipment full life cycle health management method. The processor 102 can execute multiple instructions to achieve: acquiring multi-source real-time operating data of the substation equipment, using the pre-built digital twin model to drive the digital twin model to update the simulation state based on the multi-source real-time operating data, and outputting the twin simulation value of at least one target state parameter of the substation equipment. The deviation between the twin simulation value of the target state parameter and the physical observation value of the target state parameter is calculated and anomaly is determined. When an anomaly is determined, an anomaly feature vector is extracted and input into the fault knowledge graph for graph reasoning, and the fault type and fault location result are output. Based on the fault type, fault location results, and historical operation and maintenance data, a pre-trained physical constraint deep learning prediction model is used to predict the health evolution trend of the substation equipment and output the remaining life of the substation equipment. The physical constraint deep learning prediction model includes a deep learning network for time-series prediction of the health evolution trend and physical constraint terms for constraining the prediction results to conform to the aging mechanism. Based on the remaining lifetime and power grid operation safety constraints, a pre-trained reinforcement learning strategy model is used to output an operation and maintenance strategy, which includes at least one of maintenance, load reduction operation, enhanced monitoring, or maintaining the status quo. After executing the operation and maintenance strategy, the feedback data formed by the operation and maintenance execution results is obtained, and the parameters of the digital twin model are corrected and the model is evolved based on the feedback data.

[0064] Example 4 If the modules / units integrated in the electronic device 100 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).

[0065] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0066] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0067] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0068] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0069] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for full lifecycle health management of substation equipment based on digital twins, characterized in that, include: The system acquires multi-source real-time operating data of substation equipment, and based on a pre-built digital twin model, uses the multi-source real-time operating data to drive the digital twin model to update the simulation state, and outputs the twin simulation value of at least one target state parameter of the substation equipment. The deviation between the twin simulation value of the target state parameter and the physical observation value of the target state parameter is calculated and anomaly is determined. When an anomaly is determined, an anomaly feature vector is extracted and input into the fault knowledge graph for graph reasoning, and the fault type and fault location result are output. Based on the fault type, fault location results, and historical operation and maintenance data, a pre-trained physical constraint deep learning prediction model is used to predict the health evolution trend of the substation equipment and output the remaining life of the substation equipment. The physical constraint deep learning prediction model includes a deep learning network for time-series prediction of the health evolution trend and physical constraint terms for constraining the prediction results to conform to the aging mechanism. Based on the remaining lifetime and power grid operation safety constraints, a pre-trained reinforcement learning strategy model is used to output an operation and maintenance strategy, which includes at least one of maintenance, load reduction operation, enhanced monitoring, or maintaining the status quo. After executing the operation and maintenance strategy, the feedback data formed by the operation and maintenance execution results is obtained, and the parameters of the digital twin model are corrected and the model is evolved based on the feedback data.

2. The method for full lifecycle health management of substation equipment based on digital twins according to claim 1, characterized in that, The steps for constructing the digital twin model include: Geometric data of the substation equipment is obtained based on laser point cloud scanning, BIM modeling, or design drawings to construct a geometric twin; Based on the finite element analysis theory, a multi-physics simulation kernel integrating electromagnetics, thermodynamics, and fluid mechanics is used to construct a physical twin. Based on an expert knowledge base, equipment operation procedures, or finite state machine, the state transition rules of the substation equipment under different operating conditions are defined to construct a rule twin. The geometric twin, physical twin, and rule twin are integrated through a twin engine and a real-time data-driven interface is provided to use the multi-source real-time running data to drive the digital twin model to update the simulation state and output the twin simulation value of at least one target state parameter.

3. The method for full lifecycle health management of substation equipment based on digital twins according to claim 1, characterized in that, The steps for acquiring multi-source real-time operating data of substation equipment include: The system receives communication messages from the sensor array and the SCADA system respectively, and performs protocol parsing and field extraction on the communication messages to obtain the measurement point values ​​and timestamps corresponding to the sensor array and the SCADA system. It also performs ledger record reading and field extraction on the information from the PMS ledger. The obtained measurement point values ​​and timestamps are synchronized and aligned in time and space. The time synchronization includes synchronizing the timestamps of heterogeneous data using the IEEE 1588 PTP protocol. The space-time alignment includes spatial alignment using GIS and BIM coordinate mapping. Data cleaning and reduction are performed on the synchronized and aligned data. The data cleaning includes using Kalman filtering and wavelet transform to remove noise and outliers, and filling in missing data. The reduction includes converting multi-source data into a unified JSON or Protobuf format.

4. The method for full lifecycle health management of substation equipment based on digital twins according to claim 1, characterized in that, The steps for calculating deviation and determining anomalies include: Calculate the physical observation values ​​using the following formula. With twin simulation values residuals between : Based on the following formula, the sliding window statistical method is used to calculate the statistics of the residuals within the window and construct a dynamic threshold. : in, , These are weighting coefficients. The historical mean of the residuals within the window. The historical standard deviation of the residuals within the window; When the residual is greater than the dynamic threshold, it is determined to be abnormal, and the extraction of the abnormal feature vector and the graph inference are triggered.

5. The method for full lifecycle health management of substation equipment based on digital twins according to claim 1, characterized in that, The steps for constructing the physical constraint deep learning prediction model include: Based on the historical operation and maintenance data, historical health index HI, cumulative aging rate and environmental stress factor are extracted as input features to construct the deep learning network for time-series prediction of the health evolution trend. The historical operation and maintenance data includes historical operation records from the SCADA system and factory parameters and maintenance records from the PMS ledger. The deep learning network includes a Transformer encoder and an LSTM network. The physical constraint term is introduced during model training, and the following loss function is used: in, This is a physical constraint term for thermal aging constructed based on Arrhenius's law. For data-driven loss terms, These are the weighting coefficients; The step of outputting the remaining lifespan of the substation equipment includes: The historical feature sequence is input into the physical constraint deep learning prediction model, and the physical constraint deep learning prediction model outputs the health curve and its confidence interval for future time. The time point at which the health curve reaches the failure threshold is determined based on a preset failure threshold, and the time difference between the predicted start time and the time point is used as the remaining lifespan of the substation equipment.

6. The method for full lifecycle health management of substation equipment based on digital twins according to claim 1, characterized in that, The steps for constructing the reinforcement learning policy model include: Under the condition of satisfying the power grid operation safety constraints, a near-end policy optimization (PPO) algorithm is used to train a reinforcement learning agent. The state space is defined to include at least: equipment health, power grid load level, and spare parts inventory, serving as the input to the reinforcement learning agent. The action space is defined to include at least: maintenance, load reduction operation, enhanced monitoring, and maintaining the status quo, serving as the output to the reinforcement learning agent. During training, policy parameters are updated based on experience samples formed by the interaction between the reinforcement learning agent and the environment to maximize long-term cumulative rewards. The following reward function is used during training: in, For reliability benefits, For maintenance costs, , The weighting coefficients are used to determine the operation and maintenance strategy. When outputting the operation and maintenance strategy, the strategy is output by the reinforcement learning strategy model based on the remaining lifetime and the power grid operation safety constraints.

7. The method for full lifecycle health management of substation equipment based on digital twins according to claim 1, characterized in that, The parameter correction and model evolution of the digital twin model based on the feedback data includes: After executing the aforementioned operation and maintenance strategy, maintenance records, handling process records, and post-maintenance measured data are acquired as feedback data. This feedback data is then used as observation data to perform Bayesian updates on the model parameters in the digital twin model used to calculate the target state parameters, thereby obtaining the posterior parameter distribution of the model parameters. The Bayesian update satisfies the following formula: in, Corresponding to the observation data, Corresponding to the model parameters, The prior parameter distribution of the model parameters before the update. Let be the likelihood function of the observed data given the model parameters; The feedback data, the fault type and fault location results, and the execution records of the operation and maintenance strategy are hashed and stored on the blockchain.

8. A digital twin-based substation equipment lifecycle health management device, characterized in that, include: The twin simulation module is used to acquire multi-source real-time operating data of substation equipment, and based on the pre-built digital twin model, use the multi-source real-time operating data to drive the digital twin model to update the simulation state, and output the twin simulation value of at least one target state parameter of the substation equipment. The fault determination module is used to calculate the deviation and determine the anomaly between the twin simulation value of the target state parameter and the physical observation value of the target state parameter. When an anomaly is determined, the module extracts the anomaly feature vector and inputs the anomaly feature vector into the fault knowledge graph for graph reasoning, and outputs the fault type and fault location result. The lifespan prediction module is used to predict the health evolution trend of the substation equipment and output the remaining lifespan of the substation equipment based on the fault type, fault location results and historical operation and maintenance data, using a pre-trained physical constraint deep learning prediction model. The physical constraint deep learning prediction model includes a deep learning network for time-series prediction of the health evolution trend and physical constraint terms for constraining the prediction results to conform to the aging mechanism. The strategy output module is used to output an operation and maintenance strategy based on the remaining lifetime and power grid operation safety constraints, using a pre-trained reinforcement learning strategy model. The operation and maintenance strategy includes at least one of maintenance, load reduction operation, enhanced monitoring, or maintaining the status quo. The model evolution module is used to obtain feedback data formed by the operation and maintenance execution results after the operation and maintenance strategy is executed, and to perform parameter correction and model evolution on the digital twin model based on the feedback data.

9. An electronic device, characterized in that, It includes a processor and a memory, the processor being used to execute a computer program stored in the memory to implement the digital twin-based substation equipment lifecycle health management method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction, which, when executed by a processor, implements the digital twin-based substation equipment lifecycle health management method as described in any one of claims 1 to 7.

Images