A wind-solar-storage fault diagnosis and self-healing system based on digital twinning

Abstract

The application discloses a kind of fusion digital twin's wind light storage fault diagnosis and self-healing system, it is related to intelligent power grid technical field, including: twin coupling modeling module, for analyzing multiple heterogeneous equipment in wind light storage system, construct the equipment coupling model of dynamic graph neural network and knowledge graph reasoning fusion, generate the energy interaction graph and fault propagation path graph between multiple heterogeneous equipment.The application can analyze the complex energy interaction relationship and fault propagation path between multiple heterogeneous equipment in wind light storage system in real time by constructing the equipment coupling model of dynamic graph neural network and knowledge graph fusion, not only realizes the quantitative characterization of dynamic coupling strength between equipment, but also can infer potential fault propagation chain based on real-time data, significantly improve the accuracy and timeliness of cross-device fault correlation analysis, provide reliable topology and logic foundation for accurate positioning of complex system fault.

Description

Technical Field

[0001] This invention relates to the field of smart grid technology, specifically to a wind, solar, and energy storage fault diagnosis and self-healing system that integrates digital twins. Background Technology

[0002] Wind-solar-storage systems are integrated energy systems composed of wind energy, solar energy, and energy storage equipment. They aim to improve the utilization efficiency and reliability of renewable energy. However, these systems may face various failures during operation, such as equipment failure, communication interruption, and environmental changes. These problems can affect the stable supply of energy and the overall performance of the system. Therefore, fault diagnosis and self-healing technologies for wind-solar-storage systems are particularly important.

[0003] For example, the collaborative optimization scheduling method for an integrated wind-solar-storage system, as disclosed in Chinese Patent Publication No. CN119253610A, can achieve accurate prediction at multiple time scales, dynamic optimization of multiple objectives, and adaptive control strategies. It also considers multiple factors such as demand-side response, fault diagnosis and self-healing, and grid support services, thereby improving the overall operating efficiency and reliability of the integrated wind-solar-storage system.

[0004] In existing technologies, wind-solar-storage integrated systems include various heterogeneous devices such as wind power generation, photovoltaic power generation, and energy storage equipment. There are complex dynamic coupling and energy interaction relationships between these devices. Existing digital twin models are difficult to accurately characterize and extrapolate cross-device fault propagation paths in the fault diagnosis stage, resulting in delayed fault location and inaccurate root cause judgment. Furthermore, after fault identification, the system's self-healing control strategy often ignores the influence of time-varying factors such as equipment health degradation and environmental cumulative effects. This may cause system operation oscillations, performance degradation, or even repeated faults during the recovery process, affecting the long-term reliable operation of the system. To address these issues, a wind-solar-storage fault diagnosis and self-healing system integrating digital twins is proposed. Summary of the Invention

[0005] To solve the above-mentioned technical problems, the present invention is implemented through the following technical solution: a wind, solar, and energy storage fault diagnosis and self-healing system integrating digital twins, comprising:

[0006] The twin coupling modeling module is used to analyze multiple heterogeneous devices in the wind-solar-storage system, construct a device coupling model that integrates dynamic graph neural network and knowledge graph reasoning, generate energy interaction relationship diagrams and fault propagation path diagrams among multiple heterogeneous devices, provide accurate coupling relationship representation and reasoning basis for fault diagnosis, and realize accurate characterization of dynamic energy interaction relationship among heterogeneous devices and real-time representation of fault propagation path.

[0007] The fault propagation simulation module, based on the energy interaction relationship diagram and fault propagation path diagram output by the equipment coupling model, simulates the propagation process of cross-equipment faults in real time, quickly locates complex faults and performs root cause diagnosis, realizes cross-equipment simulation of complex faults, and improves the efficiency and accuracy of fault location.

[0008] The status assessment module, based on the fault propagation simulation results, assesses the health status and environmental cumulative effects of each heterogeneous device in real time, constructs a health status optimization model, designs time-varying parameters and safety boundary conditions for the self-healing strategy, realizes dynamic assessment of device health status, and provides time-varying safety boundaries for the self-healing strategy.

[0009] The self-healing strategy optimization module combines deep reinforcement learning and adaptive fruit fly optimization algorithm to generate the optimal self-healing control sequence under the constraints of the health state optimization model, thereby achieving high efficiency and reliability in the recovery process, realizing efficient self-healing strategy generation under complex constraints, and ensuring the smooth and reliable recovery process of the system.

[0010] The strategy execution module is used to simulate and execute the self-healing strategies covered in the optimal self-healing control sequence in a digital twin environment, evaluate their recovery effect, system stability and equipment load changes, optimize strategy parameters, ensure their feasibility and stability in the actual system, and verify the feasibility of the strategy through twin simulation.

[0011] Preferably, the twin coupling modeling module includes a dynamic graph construction unit and a knowledge graph reasoning unit;

[0012] The dynamic graph construction unit is used to collect the operating data of each heterogeneous device in the wind-solar-storage system in real time, construct a dynamic interaction graph between devices based on a dynamic graph neural network, identify coupling strength and energy flow direction, infer potential fault propagation chains, improve the dynamic identification capability of coupling relationships between devices, and support the early detection of abnormal propagation chains.

[0013] The knowledge graph reasoning unit is used to integrate the physical characteristics, operating rules and historical fault knowledge of various heterogeneous devices, construct a fault reasoning knowledge graph, integrate the potential fault propagation chain of dynamic interaction graph reasoning between devices, construct a device coupling model, output energy interaction relationship graph and fault propagation path graph, represent the dynamic energy interaction and fault propagation path between multiple types of heterogeneous devices in real time, integrate real-time data and prior knowledge, and enhance the interpretability and coverage of fault reasoning.

[0014] Preferably, the dynamic graph construction unit performs the following steps:

[0015] Real-time acquisition of operational data from various heterogeneous devices in the wind-solar-storage system, including wind turbine generators, photovoltaic arrays, and energy storage devices, including power output, voltage, current, frequency, and temperature, enables high-frequency synchronous acquisition of multi-source heterogeneous data, providing a real-time data foundation for system status perception;

[0016] Based on the aforementioned operational data, a dynamic graph neural network model is constructed with nodes representing devices and edges representing energy interactions. The node features and edge weights are dynamically updated, thereby enabling quantitative modeling and continuous tracking of the dynamic coupling relationships between devices and improving the model's adaptability to system topology changes.

[0017] By calculating the coupling strength and energy flow direction between nodes through graph attention mechanism, the potential fault propagation chain between multiple devices is inferred and output as a structured dynamic interaction graph between devices, thereby enhancing the interpretability and inference accuracy of fault association paths and providing structured input for subsequent root cause localization.

[0018] Preferably, the knowledge graph reasoning unit performs the following steps:

[0019] By integrating the physical characteristics, operating rules, and historical failure cases of wind power, photovoltaic, and energy storage equipment, a fault reasoning knowledge graph containing equipment type, fault mode, and propagation logic is constructed, forming a unified representation system covering multi-source heterogeneous knowledge and improving the comprehensiveness of fault mode cognition.

[0020] The output potential fault propagation chain is semantically aligned and rule-integrated with the fault reasoning knowledge graph to form a device coupling model, realizing the deep integration of dynamic propagation logic and static knowledge structure, and enhancing the accuracy and interpretability of fault reasoning.

[0021] Based on the fused device coupling model, energy interaction relationship diagrams and fault propagation path diagrams are generated in real time to present the dynamic energy interaction and fault propagation path between heterogeneous devices. This intuitively displays the dynamic evolution process of energy and faults within the system, supports multi-dimensional situational awareness and decision support, and continuously updates the fault propagation patterns and reasoning rules in the graph to ensure the timeliness and adaptability of the knowledge graph, thereby promoting the autonomous evolution and continuous improvement of fault diagnosis capabilities.

[0022] Preferably, the fault propagation simulation module includes a path simulation unit and a root cause diagnosis unit;

[0023] The path simulation unit is used to simulate the propagation process of faults between different devices using a digital twin environment, combine real-time data to infer the scope and timing characteristics of the fault impact, output a path simulation dataset, provide high-fidelity fault propagation simulation, and support multi-dimensional impact analysis and verification.

[0024] The root cause diagnosis unit is used to combine path simulation datasets with energy interaction relationship graphs and fault propagation path graphs for collaborative reasoning, identify the source equipment and critical path of fault propagation, achieve accurate root cause localization, reduce false positives and false negatives, and realize root cause diagnosis through multi-source information fusion.

[0025] Preferably, the path simulation unit performs the following steps:

[0026] Import the energy interaction diagram and fault propagation path diagram output by the device coupling model into the digital twin environment to build a high-fidelity system simulation foundation and clarify the objects and paths of fault inference;

[0027] Simulated fault signals are injected based on real-time monitoring data, and the dynamic propagation process of faults between devices is deduced based on the physical simulation engine. The diffusion behavior of faults in the actual system is dynamically reproduced, and potential risk paths are identified.

[0028] Record the temporal characteristics, impact range, and energy flow changes of fault propagation, generate a path simulation dataset containing multi-dimensional simulation indicators, and form a traceable and quantifiable fault propagation archive to support subsequent root cause diagnosis.

[0029] Preferably, the root cause diagnosis unit performs the following steps:

[0030] By combining path simulation datasets with energy interaction graphs, key nodes and abnormal energy mutation points in the fault propagation path are extracted, significantly improving the ability to capture fault correlation features and enhancing the sensitivity of abnormal event identification.

[0031] By performing reverse reasoning and causal analysis based on the fault propagation path diagram, the source equipment of the fault and the critical path of propagation can be identified, the root cause of the fault can be quickly traced, and the diagnostic efficiency and accuracy can be improved.

[0032] By using graph neural networks and Bayesian inference methods to fuse multi-source information, a high-confidence root cause diagnosis report is output to locate the source and propagation mechanism of the fault, realize multi-source evidence fusion reasoning, and improve the credibility and interpretability of fault location.

[0033] Preferably, the state assessment module performs the following steps:

[0034] Based on the fault propagation simulation results, the real-time operating status parameters and historical degradation trends of each heterogeneous device are extracted to improve the real-time perception capability of health status and the accuracy of correlation analysis of historical degradation trends.

[0035] Assess the impact of cumulative environmental effects on equipment health status, establish a time-varying degradation model that incorporates aging, temperature, and load fluctuations, and achieve dynamic assessment and predictive maintenance of equipment health status.

[0036] Based on the equipment safety operation boundary and system stability constraints, a multi-objective and multi-constraint health state optimization model is constructed, and the time-varying parameters and safety boundary conditions of the self-healing strategy are output to ensure that the self-healing strategy is executed efficiently within the safety boundary and improve the reliability of the recovery process.

[0037] Preferably, the self-healing strategy optimization module performs the following steps:

[0038] Based on the time-varying parameters and safety boundary output by the state evaluation module, a deep reinforcement learning environment is constructed, defining the state space, action space and reward function to realize a highly adaptive learning framework with multi-dimensional state perception and reward drive.

[0039] The Soft Actor-Critic algorithm is used for policy exploration, and the adaptive fruit fly optimization algorithm is combined to search for the optimal control sequence in the continuous action space, which significantly improves the policy exploration efficiency and global optimization ability.

[0040] The self-healing strategy is iteratively optimized under multiple constraints of the health state optimization model to generate the optimal self-healing control sequence that takes into account recovery efficiency, equipment load and system stability, so as to ensure the safety and efficiency of the self-healing strategy under complex constraints.

[0041] Preferably, the strategy execution module performs the following steps:

[0042] Load the optimal self-healing control sequence in the digital twin environment, simulate the execution of the self-healing strategy at each stage, and ensure high-precision synchronization and reliable mapping between the control command and the physical system, laying the foundation for actual execution;

[0043] Real-time monitoring of system stability indicators, equipment load changes and recovery effects during simulation, recording key performance parameters, realizing digital tracking and quantitative evaluation of the entire process, and providing objective basis for strategy adjustment;

[0044] The parameters of the self-healing strategy are optimized based on simulation results feedback, and its feasibility and execution stability in the actual system are verified. A closed-loop optimization mechanism is formed to improve the adaptability and robustness of the self-healing strategy and realize the continuous autonomous optimization and evolution of the system.

[0045] This invention provides a wind, solar, and energy storage fault diagnosis and self-healing system integrating digital twins. It has the following beneficial effects:

[0046] (i) This wind, solar and energy storage fault diagnosis and self-healing system integrates digital twins. By constructing a device coupling model that integrates dynamic graph neural network and knowledge graph, it can analyze the complex energy interaction relationship and fault propagation path between multiple heterogeneous devices in the wind, solar and energy storage system in real time. It not only realizes the quantitative characterization of the dynamic coupling strength between devices, but also infers the potential fault propagation chain based on real-time data. It significantly improves the accuracy and timeliness of cross-device fault correlation analysis and provides a reliable topological and logical basis for the accurate location of faults in complex systems.

[0047] (II) This wind, solar and energy storage fault diagnosis and self-healing system integrates digital twins and introduces a fault propagation and inference mechanism based on digital twins. Through a high-fidelity physical simulation engine, it simulates the dynamic diffusion process of faults in a virtual environment. It can combine real-time monitoring data to accurately infer the spatiotemporal characteristics of fault impact and protection action effects, thereby achieving rapid and accurate diagnosis of fault root causes, effectively reducing misjudgments and missed detections caused by unclear fault propagation paths, and significantly improving the completeness and reliability of fault diagnosis.

[0048] (III) This wind, solar and energy storage fault diagnosis and self-healing system integrates digital twins. By establishing a time-varying degradation model with multiple coupled factors, it can achieve scientific assessment and quantitative prediction of equipment health status and environmental cumulative effects. It comprehensively considers multiple degradation mechanisms such as aging, temperature, and load fluctuations, and can dynamically calculate the comprehensive degradation index and remaining life of the equipment. It provides key health status boundary and life warning information for the formulation of self-healing strategies, and ensures that system operation and maintenance decisions are based on the actual degradation state of the equipment. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the workflow of a wind, solar and energy storage fault diagnosis and self-healing system integrating digital twins according to the present invention.

[0050] Figure 2 This is a modular structure diagram of a wind, solar, and energy storage fault diagnosis and self-healing system integrating digital twins according to the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Example 1, please refer to Figure 1 , Figure 2 This invention provides a technical solution: a wind, solar, and energy storage fault diagnosis and self-healing system integrating digital twins, comprising:

[0053] The twin coupling modeling module is used to analyze multiple heterogeneous devices in the wind-solar-storage system, construct a device coupling model that integrates dynamic graph neural network and knowledge graph reasoning, generate energy interaction relationship graphs and fault propagation path graphs among multiple heterogeneous devices, provide accurate coupling relationship representation and reasoning basis for fault diagnosis, and realize accurate characterization of dynamic energy interaction relationship among heterogeneous devices and real-time representation of fault propagation path. The twin coupling modeling module includes a dynamic graph construction unit and a knowledge graph reasoning unit.

[0054] The dynamic graph construction unit is used to collect real-time operating data of various heterogeneous devices in the wind-solar-storage system, construct a dynamic interaction graph between devices based on a dynamic graph neural network, identify coupling strength and energy flow direction to infer potential fault propagation chains, improve the dynamic identification capability of coupling relationships between devices, support the early detection of abnormal propagation chains, collect real-time operating data of various heterogeneous devices in wind turbine generators, photovoltaic arrays, and energy storage devices in the wind-solar-storage system, including power output, voltage, current, frequency, and temperature, realize high-frequency synchronous acquisition of multi-source heterogeneous data, provide a real-time data foundation for system status perception, construct a dynamic graph neural network model with nodes as devices and edges as energy interactions based on operating data, dynamically update node features and edge weights, thereby realizing quantitative modeling and continuous tracking of dynamic coupling relationships between devices, improving the model's adaptability to system topology changes, calculate the coupling strength and energy flow direction between nodes through graph attention mechanism, infer potential fault propagation chains between multiple devices, and output a structured dynamic interaction graph between devices, thereby enhancing the interpretability and inference accuracy of fault association paths, and providing structured input for subsequent root cause localization;

[0055] The specific work involves: During implementation, based on the actual hardware architecture and communication protocol of the integrated wind-solar-storage system, establishing a standardized real-time data acquisition interface. Through intelligent sensors and edge computing terminals deployed on various heterogeneous device nodes, synchronously collecting operating parameter data including power, voltage, current, frequency, and temperature at a sampling period of no less than 100ms. After preliminary anomaly detection and format standardization at the edge, the collected operating parameter data is uploaded to the central data platform via an encrypted communication channel. The central data platform performs timestamp alignment and dimension normalization on the operating parameter data of each heterogeneous device node, forming a structured multi-source time-series data stream. Based on the collected multi-source time-series data, a system is constructed with physical devices as nodes and actual energy... The model is a dynamic graph neural network model with transmission paths as edges. Each node's feature vector contains dimensions of the device's real-time power, voltage, current, frequency, and temperature. Edge weights are dynamically calculated based on the power exchange value and power direction between adjacent devices. The graph structure is initialized according to the actual system topology, and a dynamic update mechanism based on time sliding windows is adopted. The node features and edge weights are recalculated and updated every 5 minutes. The model uses a graph attention network layer for feature aggregation and calculates attention coefficients through a trainable parameter matrix to reflect the coupling strength between devices and the dominant energy flow direction. Through multi-layer graph convolution and attention mechanisms, potential cross-device fault association paths are inferred from real-time data, and a dynamic interaction graph between devices with weight labels, direction indicators, and time attributes is output.

[0056] The knowledge graph reasoning unit integrates the physical characteristics, operating rules, and historical fault knowledge of various heterogeneous devices to construct a fault reasoning knowledge graph. It merges potential fault propagation chains from dynamic interaction graphs between devices, builds device coupling models, and outputs energy interaction relationship graphs and fault propagation path graphs. This provides real-time representation of dynamic energy interactions and fault propagation paths among multiple types of heterogeneous devices, integrating real-time data and prior knowledge to enhance the interpretability and coverage of fault reasoning. It integrates the physical characteristics, operating rules, and historical fault cases of wind power, photovoltaic, and energy storage equipment to construct a fault reasoning knowledge graph containing device type, fault mode, and propagation logic, forming a unified representation system covering multi-source heterogeneous knowledge and improving fault mode representation. The comprehensiveness of cognition involves semantic alignment and rule fusion of the output potential fault propagation chain with the fault reasoning knowledge graph to form a device coupling model. This achieves deep integration of dynamic propagation logic and static knowledge structure, enhancing the accuracy and interpretability of fault reasoning. Based on the fused device coupling model, energy interaction relationship graphs and fault propagation path graphs are generated in real time to present the dynamic energy interaction and fault propagation path between heterogeneous devices. This intuitively demonstrates the dynamic evolution of energy and faults within the system, supports multi-dimensional situational awareness and decision support, and continuously updates the fault propagation patterns and reasoning rules in the graph to ensure the timeliness and adaptability of the knowledge graph, thereby promoting the autonomous evolution and continuous improvement of fault diagnosis capabilities.

[0057] The specific work involves: constructing an entity and relationship system for a fault reasoning knowledge graph; defining equipment entities, attributes, and their relationships based on equipment manuals, operating procedures, and historical maintenance records for wind turbines, photovoltaic inverters, and battery energy storage systems; classifying fault mode entities according to international standards and field fault statistics; defining fault propagation logic using directed rules; storing the knowledge graph using a resource description framework; and using SPARQL for querying and reasoning. The initial graph should cover no fewer than 3 main equipment categories and 15 typical types of equipment. Fault modes and 50 propagation rules; static attributes of wind turbines include rated power, voltage level, rated wind speed, blade type, manufacturer, installation date, etc., and dynamic attributes include real-time power, speed, pitch angle, nacelle temperature, etc.; static attributes of photovoltaic inverters include rated power, input voltage range, MPPT efficiency, protection level, communication protocol, etc., and dynamic attributes include input voltage / current, output power, temperature, efficiency, etc.; static attributes of battery energy storage systems include rated capacity, rated power, charge / discharge efficiency, cycle life, SOC range, etc., and dynamic attributes include SOC, SOH, charge / discharge power, battery temperature, etc.; static attributes of fault modes include Fault codes, fault levels (minor / moderate / critical), possible causes, and affected equipment types are listed. Dynamic attributes include occurrence time, duration, and current status. Static attributes of propagation rules include rule ID, preconditions (predecessor fault), conclusion (subsequent fault), confidence level, and historical frequency. Dynamic attributes include most recent trigger time and current confidence level. For fault mode classification: wind turbine faults include blade icing, yaw system anomalies, generator overheating, and gearbox vibration exceeding limits; photovoltaic faults include inverter overheating, MPPT failure, module hot spots, and DC arcing; and energy storage faults include overcharging / over-discharging, abnormal temperature, SOC jumps, and increased internal resistance. System-level faults include frequency overruns, voltage drops, communication interruptions, and protection malfunctions. The potential fault propagation chain inferred in real time from the dynamic graph neural network is structured and analyzed to extract the device entities, abnormal events, and time series features in the chain. Through a pre-set semantic mapping table, the device IDs and parameter anomaly types in the dynamic chain are accurately matched with the corresponding entities and fault modes in the knowledge graph. After successful matching, the temporal and intensity information in the dynamic chain is used as new evidence weights or constraints and injected into the corresponding propagation rules of the knowledge graph to form a unified device coupling model that integrates real-time data association and prior knowledge logic.Based on real-time data streams and updated inference rules, the device coupling model periodically derives the current energy interaction relationships and fault propagation paths of the system. The energy interaction relationship diagram is visualized in the form of a force-directed graph, where the node size represents the real-time power of the device, and the thickness and color intensity of the edges dynamically reflect the magnitude and direction of the interaction power. Based on the updated rules, forward causal inference is performed to generate a fault propagation path diagram, and the complete transmission chain from the potential fault source to the potentially affected device is marked in the form of a highlighted path. Each path is accompanied by the current confidence value. An automatic graph update mechanism is established. When the same type of dynamic propagation chain appears repeatedly within a set time window for more than a preset threshold and is not covered by the existing knowledge graph rules, the rule learning process is triggered. After review and confirmation by the operation and maintenance personnel, the new model is formally incorporated into the knowledge graph to realize the autonomous evolution and continuous improvement of the graph.

[0058] The fault propagation simulation module, based on the energy interaction relationship diagram and fault propagation path diagram output by the equipment coupling model, simulates the propagation process of cross-equipment faults in real time, quickly locates complex faults and performs root cause diagnosis, realizes cross-equipment simulation of complex faults, and improves the efficiency and accuracy of fault location. The fault propagation simulation module includes a path simulation unit and a root cause diagnosis unit.

[0059] The path simulation unit is used to simulate the propagation process of faults between different devices using a digital twin environment. It combines real-time data to infer the impact range and timing characteristics of faults, outputs a path simulation dataset, provides high-fidelity fault propagation simulation, supports multi-dimensional impact analysis and verification, imports the energy interaction relationship diagram and fault propagation path diagram output by the device coupling model into the digital twin environment, constructs a high-fidelity system simulation foundation, clarifies the objects and paths of fault inference, injects simulated fault signals based on real-time monitoring data, and infers the dynamic propagation process of faults between devices based on the physical simulation engine. It dynamically reproduces the diffusion behavior of faults in the actual system, identifies potential risk paths, records the timing characteristics, impact range and energy flow changes of fault propagation, generates a path simulation dataset containing multi-dimensional simulation indicators, and forms a traceable and quantifiable fault propagation archive to support subsequent root cause diagnosis.

[0060] The specific work involves: In a digital twin environment, importing the energy interaction diagram and fault propagation path diagram generated in real time from the device coupling model through a standard data interface. The energy interaction diagram is stored in JSON-LD format, with nodes containing device ID, real-time power, voltage level, and topology connections. Edge attributes include interaction power value, direction, and timestamp. The fault propagation path diagram is represented using RDF triples, with each path labeled with propagation sequence, confidence level, and rule ID. Subsequently, simulated fault signals are injected based on the real-time monitoring data stream. For key nodes such as wind turbines, photovoltaic inverters, and energy storage devices, the fault signals can simulate power surges. Typical anomaly modes such as voltage drop, voltage exceeding limits, frequency deviation, or temperature anomaly are simulated. The injection process follows the IEC61850 communication protocol, synchronizing fault events to the physical simulation engine in the digital twin environment via a publish-subscribe mechanism. A multi-domain physical simulation engine based on the Modelica language is employed, integrating a wind turbine aerodynamic-mechanical-electrical coupling model, a photovoltaic dual-diode equivalent circuit model, and a second-order RC equivalent circuit model for energy storage. The simulation step size is set to 10ms to meet the high-precision simulation requirements of dynamic processes. After fault injection, the simulation engine, based on the energy interaction relationships and fault propagation rules in the equipment coupling model, uses differential-algebraic methods to simulate the fault. The equation system solves the dynamic propagation process of faults between devices in real time. During the simulation, the temporal changes of electrical quantities (power, voltage, current) and state quantities (temperature, speed, SOC) at each node are continuously monitored, and the spatiotemporal diffusion characteristics of the fault impact are recorded, including fault propagation delay, number of affected devices, and energy flow interruption or distortion path. Simultaneously, by calling the built-in protection logic model, the system simulates the blocking or exacerbating effects of relay protection device actions on fault propagation in the actual system. Throughout the fault simulation, simulation results are synchronously collected at 100ms intervals to generate a structured path simulation dataset. This dataset utilizes a time-series database. The storage includes a timestamp, device identifier, multi-dimensional simulation metrics, and metadata tags for each record. The multi-dimensional simulation metrics cover electrical (power deviation rate, voltage sag depth, frequency change rate), thermal (equipment temperature rise rate, hot spot temperature), mechanical (vibration acceleration, torque fluctuation), and system (energy loss degree, power supply reliability index). The path simulation dataset also records the topological characteristics of fault propagation, including a list of affected devices, propagation path length, number of branch nodes, and key bottleneck devices. All data undergoes a data verification process to ensure time alignment, uniformity of units, and compliance with the IEEE 1159 power quality event classification standard.

[0061] The root cause diagnosis unit is used to combine path simulation datasets with energy interaction graphs and fault propagation path graphs for collaborative reasoning. This identifies the source equipment and critical paths of fault propagation, achieving accurate root cause localization, reducing false positives and false negatives, and realizing multi-source information fusion for root cause diagnosis. By combining path simulation datasets with energy interaction graphs, it extracts key nodes and abnormal energy mutation points in the fault propagation path, significantly improving the ability to capture fault correlation features and enhancing the sensitivity of abnormal event identification. Based on the fault propagation path graph, it performs reverse reasoning and causal analysis to identify the source equipment and critical propagation paths of the fault, quickly tracing the root cause of the fault and improving diagnostic efficiency and accuracy. It uses graph neural networks and Bayesian reasoning methods to fuse multi-source information and outputs a high-confidence root cause diagnosis report, locating the source of the fault and the propagation mechanism, realizing multi-source evidence fusion reasoning, and improving the credibility and interpretability of fault localization.

[0062] The specific work involves the following steps: In practical operation, the path simulation dataset stores the time series sequences of electrical and state variables of each device during fault propagation in 10ms increments. First, sampled values ​​of key indicators such as power, voltage, current, and temperature are extracted from this dataset by device dimension. Their first-order differences and standard deviations are analyzed to identify abnormal energy mutation points. Simultaneously, combined with a real-time updated energy interaction graph, graph algorithms are used to analyze the betweenness centrality and proximity centrality of each node, extracting key nodes located at hub positions in the fault propagation topology. The criteria for determining key nodes include: a connectivity change rate exceeding 30% within the simulation time window, or an average power flow change of more than 20% of its rated capacity on its connected edges. Devices that simultaneously meet the abnormal mutation conditions and key topological characteristics are marked as key abnormal nodes, and their timestamps, mutation magnitudes, and associated edge information are recorded to form a structured list of key events. Based on the extracted list of key abnormal nodes… The fault propagation path graph is invoked for reverse reasoning. The reasoning process is based on the predefined propagation rules in the graph. The SPARQL 1.1 update statement is used to perform reverse queries. Starting from the abnormal device at the end, the path is traversed along the ":hasCause" or ":precondition" relationship edge towards the source. During the traversal, the time sequence in the path simulation dataset is combined with the time window matching of the preconditions of the rules. The default time window is 5 seconds before and after the fault trigger. Potential causal chains that conform to the time logic are filtered out. At the same time, the protection action records in the simulation data, such as relay trip time and circuit breaker opening signal, are used to correct and prune the propagation path and eliminate invalid paths interrupted by protection actions. Finally, one or more device sequences that trace back from the abnormal phenomenon to the suspected source are output. Each path is accompanied by a comprehensive suspicion score calculated based on rule confidence and time sequence consistency, and the key transmission devices in the path are identified.

[0063] The expression for the overall suspiciousness score is as follows:

[0064] ;

[0065] ;

[0066] ;

[0067] In the formula: The higher the score, the greater the likelihood that the path is a genuine failure propagation path. This is the rule confidence weighting coefficient, representing the importance of rule confidence in the overall score, and is usually set to 0.5; The rule confidence score is derived from the predefined propagation rule confidence score field in the knowledge graph. This is the weighting coefficient for time series matching, representing the importance of the degree of time series fit, and is usually set to 0.3; For time series matching degree; This represents the number of nodes in the path. For the first in the path The timestamp of each node when an anomaly actually occurs in the simulation data; This is the expected timestamp calculated based on the propagation rules and energy interaction relationship; The standard deviation of the time tolerance window is set to 2.5 seconds by default. The weighting coefficient for protection actions represents the degree of influence of protection actions on path effectiveness, and is usually set to 0.2; To protect the impact of actions on the weight decay factor; This represents the number of protective actions involved in the path. For the first The blocking strength of a protective action is set as follows: if the action completely blocks the propagation, the value is 1; if it partially blocks the propagation, the value is 0.5; and if it has no effect, the value is 0.

[0068] A graph neural network and Bayesian inference are used to fuse multi-source information. The energy interaction graph, fault propagation path graph, and key node features are input into a three-layer graph attention network. The network node features have a dimension of 12, including real-time operating parameters and topological attributes. The higher-order representation of the nodes is learned through message passing and feature aggregation. Then, the node embedding vector output by the graph neural network is concatenated with the statistical features in the path simulation data as observational evidence of the Bayesian network. The Bayesian network structure is pre-constructed based on historical fault cases and expert knowledge. Its nodes correspond to equipment fault modes, and the edges represent conditional probability relationships. The posterior probability of each fault mode is calculated using an exact inference algorithm (connection tree algorithm) and verified by combining temporal constraints. Finally, a structured root cause diagnosis report is generated, which includes: the primary fault source equipment (posterior probability ≥ 0.7), the core propagation path (including key equipment and propagation delay), the main influencing mechanism, and the comprehensive confidence level (calculated based on probability and rule weight fusion). The report output format conforms to the IEEE C37.239 standard.

[0069] The expression for the posterior probability of the failure mode is as follows:

[0070] ;

[0071] ;

[0072] In the formula: In order to provide given observational evidence Under these conditions, failure modes The posterior probability; For the first Types of failure modes ; This represents the total number of failure modes. The observation evidence vector is composed of two concatenated parts: ; The node embedding vector output by the graph attention network has a dimension of 16. Statistical feature vectors extracted from path simulation data; This represents a vector concatenation operation; Fault mode The prior probability is set based on historical failure statistics and expert experience; In failure mode Evidence observed under the conditions of occurrence The likelihood probability is defined by the conditional probability table (CPT) in the Bayesian network; Evidence vector The first in Each feature dimension; In the Bayesian network structure The set of parent nodes; The total number of evidentiary features; This is the total probability, used to normalize the posterior probability.

[0073] The status assessment module, based on the fault propagation simulation results, assesses the health status and environmental cumulative effects of each heterogeneous device in real time, constructs a health status optimization model, designs time-varying parameters and safety boundary conditions for the self-healing strategy, realizes dynamic assessment of device health status, and provides time-varying safety boundaries for the self-healing strategy.

[0074] The self-healing strategy optimization module combines deep reinforcement learning and adaptive fruit fly optimization algorithm to generate the optimal self-healing control sequence under the constraints of the health state optimization model, thereby achieving high efficiency and reliability in the recovery process, realizing efficient self-healing strategy generation under complex constraints, and ensuring the smooth and reliable recovery process of the system.

[0075] The strategy execution module is used to simulate and execute the self-healing strategies covered in the optimal self-healing control sequence in a digital twin environment, evaluate their recovery effect, system stability and equipment load changes, optimize strategy parameters, ensure their feasibility and stability in the actual system, and verify the feasibility of the strategy through twin simulation.

[0076] Example 2, as Figure 1 , Figure 2 As shown, based on Embodiment 1, the present invention provides a technical solution: the status assessment module performs the following steps: based on the fault propagation inference results, extract the real-time operating status parameters and historical degradation trends of each heterogeneous device, improve the real-time perception capability of health status and the accuracy of correlation analysis of historical degradation trends, assess the impact of environmental cumulative effects on the health status of the equipment, establish a time-varying degradation model combining aging, temperature, and load fluctuations, realize dynamic assessment and predictive maintenance of equipment health status, construct a multi-objective, multi-constraint health status optimization model with equipment safe operation boundary and system stability constraints as conditions, and output the time-varying parameters and safety boundary conditions of the self-healing strategy to ensure that the self-healing strategy is executed efficiently within the safety boundary and improve the reliability of the recovery process;

[0077] The specific work involves extracting real-time operating status parameters of each device during the operation of the integrated wind-solar-storage system, based on simulation results output by the fault propagation simulation module. These parameters include electrical quantities (power, voltage, current, frequency), mechanical quantities (speed, vibration, torque), thermal quantities (nacelle temperature, bearing temperature, component temperature), and status indicators (such as SOC and SOH). Each operating status parameter is synchronously collected from physical sensors and the digital twin environment with a sampling period of no less than 100ms. After filtering and normalization, a structured time-series data stream is formed. Simultaneously, operating records of each device over the past 30 to 365 days are extracted from the historical database to establish a degradation trend curve with time as the horizontal axis. Key indicators include performance degradation rate, fault frequency, and cumulative operating time. Based on trend analysis, the health baseline of each device is identified. A time-varying degradation model with multiple coupled factors is used to quantitatively assess the impact of environmental cumulative effects on equipment health status. An aging effect is analyzed using a life decay model based on the Weibull distribution, with shape parameter β ranging from 1.2 to 2.5. The scale parameter η is set according to the equipment's design life (20 years for wind turbines, 25 years for photovoltaic modules, and 10 years for energy storage batteries). The cumulative effect of temperature is calculated using the Arrhenius acceleration factor model, with activation energy Ea ranging from 0.6 to 1.2 eV and a reference temperature of 25°C. Load fluctuation effects are analyzed by extracting the load spectrum using the rainflow counting method. Fatigue damage is calculated using Miner's linear cumulative damage theory, and the comprehensive degradation index D of each piece of equipment is dynamically calculated. A warning is triggered when D ≥ 0.8, and replacement is recommended when D ≥ 0.95. The model supports rolling predictions of the remaining service life of equipment under different operating scenarios.

[0078] The expression for the comprehensive degradation index is as follows:

[0079] ;

[0080] ;

[0081] ;

[0082] ;

[0083] In the formula: The comprehensive degradation index is a quantitative indicator representing the overall health status of the equipment. These are the weighting coefficients for three degradation factors: aging, temperature accumulation, and fatigue damage. The default value is [value missing]. , , ; For aging and degradation components based on the Weibull distribution; For the equipment during runtime The reliability function is given by the Weibull distribution. ; This refers to the cumulative operating time of the equipment. The shape parameter of the Weibull distribution reflects the time-varying characteristics of the failure mode; The scale parameter (characteristic lifetime) for the Weibull distribution is set according to the equipment design lifetime: 20 years for wind turbines, 25 years for photovoltaic modules, and 10 years for energy storage batteries. This is the degradation component caused by the cumulative effect of temperature. It is an Arrhenius accelerator factor. ; The activation energy reflects the material's sensitivity to temperature. Here is the Boltzmann constant, with a value of 8.617 × 10⁻⁶. -5 eV / K; For reference temperature, set to 25. o C (i.e., 298.15K); This refers to the actual operating temperature of the equipment. For temperature The cumulative running time; The design life of the equipment (in years) is the same as the design life of the equipment. The values ​​are consistent; This represents the fatigue damage and degradation component caused by load fluctuations. This represents the total number of different load levels in the load spectrum. In the first Each load level The actual number of loops that occurred; In the first Each load level The fatigue life cycle count is determined by the SN curve of the material or component; load level Rainflow counting is used to extract data from equipment operation (power, torque, temperature, etc.). Based on a time-varying degradation model, a multi-objective, multi-constraint health state optimization model is constructed with equipment safety and system stability as the core. The safety boundary conditions include: allowable fluctuation range of electrical parameters, maximum allowable temperature rise of equipment, and mechanical vibration limit. The system stability constraints cover power balance error ≤2%, frequency change rate ≤0.5Hz / s, and voltage recovery time ≤3s. The optimization objective function simultaneously minimizes the overall system degradation rate, maximizes the remaining life of key equipment, and minimizes the operating risk index. The time-varying control parameters of the self-healing strategy are output through a constrained multi-objective particle swarm optimization algorithm, including power regulation, voltage compensation value, protection setting correction coefficient, and control timing. Each parameter constitutes the safety execution boundary of the self-healing strategy and is embedded in the strategy optimization module as hard constraint conditions.

[0084] The self-healing strategy optimization module performs the following steps: Based on the time-varying parameters and safety boundary output by the state evaluation module, a deep reinforcement learning environment is constructed, defining the state space, action space, and reward function to realize a highly adaptive learning framework with multi-dimensional state perception and reward drive. The Soft Actor-Critic algorithm is used for policy exploration, and the adaptive fruit fly optimization algorithm is combined to search for the optimal control sequence in the continuous action space, significantly improving the policy exploration efficiency and global optimization capability. Under the multi-constraint conditions of the health state optimization model, the self-healing strategy is iteratively optimized to generate the optimal self-healing control sequence that takes into account recovery efficiency, equipment load, and system stability, ensuring the safety and efficiency of the self-healing strategy under complex constraints.

[0085] The specific work content is as follows: In practical applications, the deep reinforcement learning environment is constructed based on the time-varying parameters output by the state evaluation module and multi-dimensional safety boundaries. The state space is designed with 42 dimensions, specifically including: real-time electrical parameters of each device (power, voltage, current, frequency), thermodynamic state of key components (nacelle temperature, bearing temperature, component hot spot temperature), mechanical vibration characteristic values, state of charge and health status of energy storage devices, and comprehensive degradation index and remaining lifetime prediction values ​​from the health status optimization model. The action space adopts a continuous vector form with 18 dimensions, covering the active power regulation coefficient of wind turbines, reactive power output setpoints of photovoltaic inverters, charging and discharging power commands of energy storage systems, and the setting of key protection devices. The fine-tuning coefficients and reward function employ a piecewise weighted design, primarily including: a basic recovery reward of +10 for system frequency deviation recovery to within 50±0.2Hz and +8 for voltage recovery to within 0.95-1.05pu; equipment safety penalties of -15 for any equipment temperature exceeding its rated operating temperature limit by 5°C and -12 for vibration intensity exceeding 4.5mm / s for critical components; and an economic reward of +5 for every 1% reduction in total energy loss during the self-healing process. This environment is encapsulated through the OpenAIGym standard interface and interacts with the physics simulator in 100ms time steps. Policy exploration utilizes the SoftActor-Critic (SAC) algorithm. The hybrid architecture of the Adaptive Fruit Fly Optimization Algorithm (AFOA) employs a three-layer fully connected structure in the main network of the SAC algorithm, with 256, 256, and 128 hidden layer neurons respectively. The policy network outputs the mean and standard deviation of a Gaussian distribution. The temperature parameter α is initialized to 0.2 and automatically adjusted. AFOA is integrated into the action selection stage of SAC to perform global exploration in the continuous action space. At the beginning of each training round, AFOA uses the action output by the current policy network as the initial population center, generating a fruit fly population of 50. Its olfactory search step size is adaptively adjusted based on the recent reward variance, ranging from 0.05 to 0.15 times the action space. The visual search stage uses a Pa... The elite selection strategy based on reto ranking involves the SAC Critic network (a dual-Q network structure with 256 neurons per layer) evaluating the Q-values ​​of AFOA candidate action sequences in each iteration and guiding the policy network towards higher entropy and higher reward directions. During training, the experience replay buffer capacity is set to 1 million records, and a batch update is performed every 5,000 new data records with a batch size of 512 to ensure efficient searching of high-dimensional control sequences that satisfy complex constraints while maintaining action continuity. Iterative optimization is performed under multiple constraints derived from the health state optimization model, and the following hard constraints must be verified in each iteration: the system power balance error does not exceed 2% of the total load, and the frequency change rate is limited to 0.The optimization objective is to minimize the combined cost function consisting of the original action sequence output by the SAC algorithm and the AFOA-optimized sequence, given a voltage of 5 Hz / s and a voltage recovery time of no more than 3 seconds for any node. This function integrates the recovery time cost, the cumulative equipment damage cost, and the system transient instability risk cost. The iterative process employs a constrained trust domain optimization method, transforming the safety boundary into inequality constraints and incorporating them into the optimizer. Each iteration generates a 60-step control sequence (corresponding to a 6-second actual duration), where each control step contains the aforementioned 18-dimensional actions. After sequence generation, full dynamic verification is performed in a digital twin simulation environment. Only sequences that simultaneously satisfy all electrical safety constraints, equipment thermal stability constraints, and system transient stability constraints are ultimately adopted as the optimal self-healing control sequence. This sequence will explicitly provide the specific setpoints and allowable fluctuation boundaries of each controllable device at each time segment (100ms interval).

[0086] The strategy execution module performs the following steps: loading the optimal self-healing control sequence in the digital twin environment, simulating the execution of the self-healing strategy at each stage, ensuring high-precision synchronization and reliable mapping between control commands and the physical system, laying the foundation for actual execution, monitoring system stability indicators, equipment load changes and recovery effects in real time during the simulation process, recording key performance parameters, realizing digital tracking and quantitative evaluation of the entire process, providing objective basis for strategy adjustment, optimizing self-healing strategy parameters based on simulation results feedback, verifying its feasibility and execution stability in the actual system, forming a closed-loop optimization mechanism, improving the adaptability and robustness of the self-healing strategy, and realizing continuous autonomous optimization and evolution of the system;

[0087] The specific work involves: When loading the optimal self-healing control sequence in the digital twin environment, the sequence file is imported into the twin platform through a standard data interface. This sequence is a 60-step control instruction set with a duration of 6 seconds and a step size of 100 milliseconds. It includes the active power regulation coefficient of the wind turbine, the reactive power setpoint of the photovoltaic inverter, the charging and discharging power instructions of the energy storage, and the protection setting fine-tuning coefficient. During loading, the integrity of the sequence format and whether each parameter exceeds the equipment safety limit are verified. After verification, the twin engine initializes the simulation environment according to the actual system topology and maps each step instruction in the sequence to the control interface of the corresponding device, ensuring that the timing synchronization accuracy error is less than 10 milliseconds. During simulation execution, the self-healing strategy is gradually advanced with a step size of 100 milliseconds, and multiple key indicators are monitored in real time. In terms of electrical stability, the voltage of each node, system frequency, and power balance error are continuously collected. In terms of equipment status, the wind turbine bearing temperature, the hot spot temperature rise rate of the photovoltaic module, the state of charge of the energy storage battery, and the vibration intensity are monitored. All monitoring data are recorded at a sampling rate of 10 milliseconds and analyzed through time synchronization. The system stores data in a sequence database, forming a structured log containing timestamps, device identifiers, parameter types, and values. At the end of the simulation, a recovery process evaluation report is automatically generated, summarizing performance parameters such as voltage recovery time (≤3 seconds), frequency stabilization time (≤2 seconds), and maximum load rate of each device. Based on the simulation results, the self-healing strategy parameters are optimized in a closed-loop manner. If the voltage recovery time exceeds 3 seconds or the device temperature exceeds the limit, a parameter calibration process is initiated: the adjustment amount is calculated based on the magnitude and duration of the exceedance; secondly, a constrained multi-objective optimization algorithm is used to re-search for control sequences that satisfy all safety boundaries within a preset action space. Optimization objectives include shortening recovery time, reducing cumulative device damage, and reducing energy loss. Finally, the optimized sequence is reloaded into the digital twin environment for iterative verification until all indicators meet the operating specifications. Through this mechanism, the self-healing strategy has undergone sufficient verification and optimization before execution in the actual system, ensuring its feasibility and stability under actual operating conditions, while simultaneously forming a continuous accumulation and self-evolving capability of the strategy library.

[0088] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0089] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A wind, solar, and energy storage fault diagnosis and self-healing system integrating digital twins, characterized in that, include: The twin coupling modeling module is used to analyze multiple heterogeneous devices in the wind, solar and energy storage system, construct a device coupling model that integrates dynamic graph neural network and knowledge graph reasoning, and generate energy interaction relationship diagrams and fault propagation path diagrams among multiple heterogeneous devices. The fault propagation simulation module, based on the energy interaction relationship diagram and fault propagation path diagram output by the equipment coupling model, simulates the propagation process of faults across equipment in real time, quickly locates complex faults, and performs root cause diagnosis. The status assessment module, based on the fault propagation simulation results, assesses the health status and environmental cumulative effects of each heterogeneous device in real time, constructs a health status optimization model, and designs time-varying parameters and safety boundary conditions for the self-healing strategy. The self-healing strategy optimization module is used to combine deep reinforcement learning and adaptive fruit fly optimization algorithm to generate the optimal self-healing control sequence under the constraints of the health state optimization model. The strategy execution module is used to simulate and execute the self-healing strategies covered in the optimal self-healing control sequence in a digital twin environment, evaluate their recovery effect, system stability and equipment load changes, and optimize strategy parameters.

2. The wind, solar, and energy storage fault diagnosis and self-healing system integrating digital twins as described in claim 1, characterized in that: The twin coupling modeling module includes a dynamic graph construction unit and a knowledge graph reasoning unit; The dynamic graph construction unit is used to collect the operating data of each heterogeneous device in the wind-solar-storage system in real time, construct a dynamic interaction graph between devices based on a dynamic graph neural network, identify coupling strength and energy flow direction, and infer potential fault propagation chains. The knowledge graph reasoning unit is used to integrate the physical characteristics, operating rules and historical fault knowledge of various heterogeneous devices, construct a fault reasoning knowledge graph, integrate the potential fault propagation chain of dynamic interaction graph reasoning between devices, construct a device coupling model, output energy interaction relationship graph and fault propagation path graph, and represent the dynamic energy interaction and fault propagation path between multiple types of heterogeneous devices in real time.

3. The wind, solar, and energy storage fault diagnosis and self-healing system integrating digital twins according to claim 2, characterized in that: The dynamic graph construction unit performs the following steps: Real-time acquisition of operational data from various heterogeneous devices within the wind-solar-storage system, including power output, voltage, current, frequency, and temperature, from wind turbine generators, photovoltaic arrays, and energy storage devices. Based on the aforementioned operational data, a dynamic graph neural network model is constructed with nodes representing devices and edges representing energy interactions, dynamically updating node features and edge weights. By calculating the coupling strength and energy flow direction between nodes using the graph attention mechanism, the potential fault propagation chain between multiple devices is inferred, and the result is output as a structured dynamic interaction graph between devices.

4. The wind, solar, and energy storage fault diagnosis and self-healing system integrating digital twins according to claim 3, characterized in that: The knowledge graph reasoning unit performs the following steps: By integrating the physical characteristics, operating rules, and historical failure cases of wind power, photovoltaic, and energy storage equipment, a fault reasoning knowledge graph is constructed that includes equipment type, failure mode, and propagation logic. The output potential fault propagation chain is semantically aligned and rule-integrated with the fault reasoning knowledge graph to form a device coupling model; Based on the fused device coupling model, energy interaction relationship diagrams and fault propagation path diagrams are generated in real time to present the dynamic energy interaction and fault propagation path between heterogeneous devices, and the fault propagation patterns and inference rules in the diagrams are continuously updated.

5. A wind, solar, and energy storage fault diagnosis and self-healing system integrating digital twins as described in claim 2, characterized in that: The fault propagation simulation module includes a path simulation unit and a root cause diagnosis unit. The path simulation unit is used to simulate the propagation process of faults between different devices using a digital twin environment, and to infer the scope and timing characteristics of the fault impact by combining real-time data, and output a path simulation dataset. The root cause diagnosis unit is used to combine path simulation datasets with energy interaction relationship graphs and fault propagation path graphs for collaborative reasoning to identify the source equipment and critical path of fault propagation, thereby achieving accurate root cause localization.

6. A wind, solar, and energy storage fault diagnosis and self-healing system integrating digital twins as described in claim 5, characterized in that: The path simulation unit performs the following steps: Import the energy interaction diagram and fault propagation path diagram output by the device coupling model into the digital twin environment; Simulated fault signals are injected based on real-time monitoring data, and the dynamic propagation process of faults between devices is deduced based on a physical simulation engine. Record the temporal characteristics, impact range, and energy flow changes of fault propagation to generate a path simulation dataset containing multi-dimensional simulation indicators.

7. A wind, solar, and energy storage fault diagnosis and self-healing system integrating digital twins as described in claim 5, characterized in that: The root cause diagnosis unit performs the following steps: By combining the path simulation dataset and the energy interaction graph, key nodes and abnormal energy mutation points in the fault propagation path are extracted. Based on the fault propagation path diagram, reverse reasoning and causal analysis are performed to identify the fault source device and the critical propagation path. By using graph neural networks and Bayesian inference methods to fuse multi-source information, a high-confidence root cause diagnosis report is output to locate the source of the fault and the propagation mechanism.

8. A wind, solar, and energy storage fault diagnosis and self-healing system integrating digital twins as described in claim 5, characterized in that: The status assessment module performs the following steps: Based on the fault propagation simulation results, the real-time operating status parameters and historical degradation trends of each heterogeneous device are extracted. To assess the impact of cumulative environmental effects on equipment health, a time-varying degradation model combining aging, temperature, and load fluctuations was established. Based on the equipment safety operation boundary and system stability constraints, a multi-objective and multi-constraint health state optimization model is constructed, and the time-varying parameters and safety boundary conditions of the self-healing strategy are output.

9. A wind, solar, and energy storage fault diagnosis and self-healing system integrating digital twins as described in claim 8, characterized in that: The self-healing strategy optimization module performs the following steps: Based on the time-varying parameters and safety boundary output by the state evaluation module, a deep reinforcement learning environment is constructed, and the state space, action space and reward function are defined. The Soft Actor-Critic algorithm is used for policy exploration, and the optimal control sequence is searched in the continuous action space by combining the adaptive fruit fly optimization algorithm. Under multiple constraints of the health state optimization model, the self-healing strategy is iteratively optimized to generate the optimal self-healing control sequence that balances recovery efficiency, equipment load, and system stability.

10. A wind, solar, and energy storage fault diagnosis and self-healing system integrating digital twins according to claim 9, characterized in that: The strategy execution module performs the following steps: Load the optimal self-healing control sequence in the digital twin environment and simulate the execution of the self-healing strategy at each stage; Real-time monitoring of system stability indicators, equipment load changes and recovery effects during simulation, and recording of key performance parameters; The parameters of the self-healing strategy are optimized based on the feedback of simulation results, and its feasibility and execution stability in the actual system are verified, forming a closed-loop optimization mechanism.

Images