Multi-source data fusion anti-mislock intelligent decision and early warning method for wind farm booster station

Abstract

The application discloses a kind of wind farm booster station multi-source data fusion prevents false locking intelligent decision and early warning method, it is related to the technical field of power system intelligent false-proof, comprising the following steps: multi-source heterogeneous data acquisition, data is obtained by distributed sensor network, and preprocessed by edge computing;Data space-time alignment and fusion, equipment state evaluation, construct deep confidence network and bayesian network hybrid model to calculate health index;False operation rule modeling, operation logic verification is carried out based on Petri net and expert knowledge base;Risk early warning decision, fusion multi-source early warning information is divided into risk grade;Intelligent locking control, use reinforcement learning to optimize locking strategy;Decision support and visualization, construct three-dimensional digital twin model and show operation guidance and risk early warning in combination with AR technology.The application improves the accuracy and efficiency of false-proof locking through multi-source data fusion and intelligent decision, and the safety and operation level of booster station are significantly enhanced by 3 months early warning of equipment failure.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent error prevention for power systems, and in particular to an intelligent decision-making and early warning method for error prevention locking by fusion of multi-source data for a wind farm booster station. Background Art

[0002] As a key hub in the power system, wind farm booster stations undertake the core tasks of power collection, voltage conversion, and safe transmission. Their operational safety and equipment reliability directly impact the stable operation of wind farms and the security of the power grid. However, traditional booster station anti-error locking systems rely on single electrical quantity data and fixed logic rules, making it difficult to meet the needs of multi-source heterogeneous data integration and risk identification under complex operating conditions. For example, in severe weather conditions, error prevention judgments based solely on switch position signals fail to integrate the potential impact of environmental data on equipment status, leading to an increased risk of misoperation. Furthermore, equipment status assessment relies on regular inspections and offline analysis, which cannot capture subtle changes in equipment health in real time. For example, traditional monitoring methods cannot provide early warnings in the early stages of minor deformation of transformer windings, which can easily lead to cascading failures.

[0003] With the intelligent upgrade of wind farms, a large number of sensors and smart devices are connected to booster stations, generating multi-dimensional data including electrical quantities, equipment status, and environmental meteorology. However, the existing system lacks an efficient data fusion mechanism, and the problem of data silos is prominent. For example, sensors from different manufacturers use heterogeneous communication protocols, and the inconsistent data formats lead to insufficient fusion accuracy and the inability to form a complete awareness of the equipment's operating status. In addition, the modeling of anti-error operation rules relies on expert experience, making it difficult to dynamically adapt to changes in equipment parameters and adjustments to operating procedures. When new energy storage equipment is added to the booster station or a new power electronic device is connected, the traditional rule base cannot be updated quickly, resulting in loopholes in the anti-error logic.

[0004] At the risk warning and decision-making level, traditional systems use fixed thresholds to trigger alarms, lacking the ability to dynamically assess and predict risk levels. For example, when the main transformer oil temperature approaches a threshold, it's impossible to comprehensively determine the risk level based on multiple factors such as load trends and winding deformation, potentially leading to false or missed alarms. Furthermore, the blocking control strategy lacks adaptability and has fixed operating procedures. In complex operational scenarios, this rigidity can impact operational efficiency and even create safety hazards. Therefore, there is an urgent need for an anti-error blocking technology that integrates multi-source data, provides intelligent decision-making, and offers dynamic warning capabilities to improve the safety and O&M efficiency of substations. Summary of the Invention

[0005] The present invention proposes a wind farm booster station multi-source data fusion anti-error locking intelligent decision-making and early warning method to solve the problems mentioned in the above-mentioned prior art.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A wind farm booster station multi-source data fusion anti-error locking intelligent decision-making and early warning method includes the following steps:

[0008] S1: Multi-source heterogeneous data collection and preprocessing steps: Deploy edge computing nodes and IoT sensor arrays to collect data, and use abnormal data filtering algorithms combined with lightweight deep learning models to perform real-time noise reduction and feature extraction on the data;

[0009] S2: Cross-scale spatiotemporal data fusion step: Construct a fusion model of long short-term memory network and graph convolution network, synchronize the data through spatiotemporal alignment algorithm, and adopt the improved DS evidence theory fusion formula: , Assign a value to the basic probability of proposition A, 、 Proposition 、 The basic probability assignment of is the data source credibility weight, 、 It is an index parameter used to distinguish different propositions and is used to represent the proposition set elements in different data sources;

[0010] S3: Device health status in-depth assessment step: Use capsule networks to dynamically learn device features and transfer similar device health assessment knowledge to the target device. The device health index calculation formula is: , m is the number of evaluation indicators, is the indicator weight, is the actual value of the indicator, 、 are the maximum and minimum values ​​of the kth indicator respectively;

[0011] S4: Dynamic modeling of anti-misoperation rules: Create a dynamic colored Petri net to adjust the parameters and connection relationships between places and transitions based on the real-time status of the equipment and the operation tasks. The knowledge graph integrates equipment knowledge, performs knowledge reasoning through a graph neural network, and combines it with a temporal logic verification algorithm to formally verify the operation sequence.

[0012] S5: Risk warning and decision optimization steps: Propose a risk warning method based on an improved evidence neural network, integrate DS evidence theory into the neural network structure, and enhance the model's ability to process uncertain information;

[0013] S6: Intelligent locking strategy collaborative control steps: Design a hierarchical distributed intelligent locking control system and use the dual deep Q network algorithm in deep reinforcement learning to optimize the locking strategy;

[0014] S7: Visual decision support and virtual simulation steps: Build a digital twin model of the substation and combine it with MR equipment to provide immersive operation and training for operators.

[0015] Furthermore, the multi-source heterogeneous data collection and preprocessing steps also include: developing an adaptive protocol conversion gateway to support dynamic parsing and conversion of industrial communication protocols, using a protocol feature fingerprint recognition algorithm to automatically identify the communication protocol type of the access device, and building an edge node task scheduling model based on game theory to allocate data collection and preprocessing tasks according to the node status.

[0016] Furthermore, the cross-scale spatiotemporal data fusion step also includes: introducing a generative adversarial network to enhance the features of spatiotemporal data, the generator learns the distribution characteristics of the original data and generates supplementary data; the discriminator distinguishes between real data and generated data, and through adversarial training, increases the diversity and integrity of the data, and establishes a fusion quality evaluation index system based on fuzzy comprehensive evaluation. The evaluation formula is: , where Q is the comprehensive score of fusion quality, n is the number of evaluation indicators, is the weight of the i-th indicator, The score for the i-th indicator is determined by the analytic hierarchy process-fuzzy entropy weight method to ensure that the fused data meets the decision-making requirements.

[0017] Furthermore, the device health status in-depth assessment step further includes: predicting device failures and evaluating remaining life based on a hybrid model of a variational autoencoder and a long short-term memory network. The remaining life calculation formula is: ,in are the mean and standard deviation of the normal state characteristics of the equipment, In order to predict the equipment characteristic values, the equipment health status is dynamically divided into four levels according to the equipment health index and combined with the fuzzy C-means clustering algorithm, providing a scientific basis for equipment maintenance.

[0018] Furthermore, the dynamic modeling step of preventing misoperation rules also includes: dynamic update of operation rules: establishing an operation rule update mechanism based on incremental learning. When the equipment structure changes, the operation process is adjusted, or a new fault case occurs, the dynamic colored Petri net model and knowledge graph are automatically updated. The knowledge distillation technology is used to integrate new knowledge into the existing model, and an operation conflict resolution strategy based on a genetic algorithm is designed. The operation safety, efficiency, and cost are optimized to construct a conflict resolution model.

[0019] Furthermore, the risk warning and decision optimization steps construct a dynamic risk assessment matrix ,in represents the risk value of the i-th risk factor in the j-th scenario, m represents the number of risk factors, and n represents the number of scenarios. , is the risk correction factor, is the probability of risk occurrence, The severity of the risk consequences; this step also includes: dynamic deduction of risk situation: constructing a dynamic deduction model of risk situation based on cellular automaton and Monte Carlo simulation, cellular automaton simulates the process of risk propagation and diffusion in the substation equipment network, Monte Carlo simulation handles the uncertainty of risk factors, and a hybrid reasoning mechanism based on case reasoning and rule reasoning, according to the current risk situation, retrieves similar cases and rules from the historical case library and rule library to generate candidate decision plans.

[0020] Furthermore, the intelligent locking strategy collaborative control step is achieved through the reward function Maximize the overall performance of the system, where S is the operational safety score, E is the operational efficiency score, and T is the operational timeliness score. is the weight coefficient; this step also includes: adaptive evolution of the locking strategy: the cultural gene algorithm is used to adaptively evolve the locking strategy of the dual deep Q network. During the algorithm iteration process, local search and global search are combined to accelerate the strategy convergence speed, and a redundancy mechanism is designed in the hierarchical distributed intelligent locking control system.

[0021] Furthermore, the visual decision support and virtual simulation steps also include: integrating gesture recognition and voice recognition technology into the digital twin model to achieve natural interaction between operators and virtual equipment, simulating the mechanical properties and physical responses of the equipment during operation through the physical engine to provide operators with a realistic operating experience, using the digital twin model to verify the generated decision plan, and evaluating the feasibility and effectiveness of the decision plan by simulating the changes in equipment status and risk evolution during the implementation of the decision plan.

[0022] Furthermore, it also includes steps for continuous system optimization and knowledge accumulation: establishing a system performance evaluation system, using principal component analysis-grey correlation analysis method to comprehensively evaluate system performance, locating system weaknesses based on the evaluation results, providing direction for system optimization, and using natural language processing technology to extract knowledge from technical documents, operating manuals, and fault reports, combined with equipment operation data, to automatically build and update knowledge graphs.

[0023] Furthermore, it also includes emergency response and rapid recovery steps: building an emergency scenario recognition model based on convolutional neural networks and bidirectional long short-term memory networks to identify emergency scenarios. The model uses an attention mechanism to highlight key features. Once the emergency scenario is identified, a recovery strategy is generated based on the target particle swarm optimization algorithm.

[0024] Compared with the existing technology, the beneficial effects of the present invention are:

[0025] At the data fusion level, through the spatiotemporal attention mechanism and the improved DS evidence theory, high-precision fusion of multi-source data such as electrical quantities, equipment status, and environmental meteorology is achieved, with a fusion accuracy of over 98%. This technological breakthrough effectively solves the problem of data silos and forms a complete awareness of equipment operation status. For example, in lightning weather, the comprehensive lightning intensity, equipment grounding current and insulation monitoring data can provide an early warning of equipment insulation risks one hour in advance, giving operation and maintenance personnel sufficient time to deal with the situation.

[0026] In terms of equipment health status assessment, the model architecture based on capsule network and transfer learning achieves accurate quantification of the equipment health index (HI), with the assessment error controlled at ±5%. Through in-depth analysis of multi-dimensional data such as transformer oil temperature, vibration spectrum, oil chromatogram, etc., potential faults such as winding aging and bearing wear can be warned three months in advance, reducing the occurrence rate of equipment failures and extending equipment service life.

[0027] In the field of error prevention and risk warning, the combination of dynamic colored Petri nets and knowledge graphs has improved the coverage of error prevention rules and shortened the operation verification time to 5 seconds. The improved evidence neural network and reinforcement learning algorithm now enable dynamic classification of risk levels and adaptive adjustment of warning thresholds.

[0028] In terms of intelligent interlocking control, the application of hierarchical distributed architecture and deep reinforcement learning has shortened the interlocking response time to less than 80ms, an improvement of more than 80% compared to traditional mechanical interlocking. In emergency response scenarios, the rapid recovery strategy based on cellular automata and ant colony algorithms can shorten the fault location time to 30 seconds, significantly improving the power supply reliability and fault recovery efficiency of the wind farm.

[0029] In summary, the technology of this application has built a safer, more efficient and more intelligent anti-error locking system for substations through full-chain innovation of multi-source data fusion, intelligent modeling, dynamic early warning and adaptive control, providing core technical support for the digital transformation and inherent safety improvement of wind farms. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a schematic block diagram of the multi-source data fusion and error-proof intelligent decision-making and early warning method for wind farm booster stations proposed by the present invention;

[0031] Figure 2 A bar chart comparing the accuracy of different data fusion methods in wind farm booster station data processing;

[0032] Figure 3 A line chart comparing the early warning time of different methods in detecting insulation aging faults in the main transformer windings of a wind farm booster station;

[0033] Figure 4The bar chart shows the comparison of operational error rates for wind farm substation operators using different training methods. DETAILED DESCRIPTION

[0034] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0035] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like to indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0036] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the said features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined. In addition, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. The present invention will be further described in detail below with reference to the accompanying drawings.

[0037] Reference Figures 1 to 4 : A wind farm booster station multi-source data fusion anti-error locking intelligent decision-making and early warning method, comprising the following steps:

[0038] Multi-source heterogeneous data collection and preprocessing steps: Edge computing nodes and IoT sensor arrays are deployed in key equipment areas of the wind farm's booster station, such as transformers, circuit breakers, and busbars. Edge computing nodes utilize embedded hardware with real-time data processing capabilities, equipped with multi-core processors and high-speed cache to support real-time computation and caching of local data. The IoT sensor array includes current and voltage sensors, temperature sensors, and vibration sensors, respectively, to collect electrical parameters, equipment temperature, and mechanical vibration data. The collected data is transmitted to the edge nodes via Industrial Ethernet. During this process, an abnormal data filtering algorithm combined with a lightweight deep learning model is used for real-time noise reduction and feature extraction. The abnormal data filtering algorithm utilizes a dual detection mechanism based on statistical thresholds and sliding windows. First, normal range thresholds are set for each data type, and data exceeding the thresholds is initially marked. Then, a sliding window is used to analyze the temporal continuity of the data and remove isolated outliers. The lightweight deep learning model uses a convolutional neural network (CNN) that has undergone pruning optimization to extract local features based on the waveform characteristics of the electrical parameters to reduce data redundancy.

[0039] To address the communication protocol differences between devices from different manufacturers, an adaptive protocol conversion gateway was developed. This gateway integrates multiple industrial communication protocol parsing modules, including Modbus, IEC61850, and OPCUA, and automatically identifies the communication protocol type of connected devices through a protocol feature fingerprint recognition algorithm. This algorithm extracts features such as fixed field positions, data length identifiers, and checksum formats from protocol messages to construct a feature fingerprint library. When a new device is connected, the real-time message features are compared with the fingerprint library to achieve automatic matching and dynamic parsing and conversion of protocol types. Furthermore, a game-theory-based edge node task scheduling model was constructed, using parameters such as each edge node's processing power, current load, and data transmission delay as game factors. A payoff function was established between nodes, and the optimal task allocation strategy was solved through Nash equilibrium. This allows for dynamic allocation of data collection and preprocessing tasks, improving the overall processing efficiency of the system.

[0040] Cross-scale spatiotemporal data fusion steps: Construct a fusion model of long short-term memory network (LSTM) and graph convolutional network (GCN) to process the spatiotemporal correlation data of booster station equipment. The LSTM network captures the time series characteristics of the data and adapts to the dynamic characteristics of electrical parameters changing over time; the GCN network uses the equipment topology structure as a graph model, converts the physical connection relationship between devices into nodes and edges of the graph, and extracts spatial correlation features. The data of different sampling frequencies and different devices are synchronously processed through the spatiotemporal alignment algorithm. The algorithm first calibrates the timestamps of each data source, and uses linear interpolation to downsample high-frequency data or upsample low-frequency data to ensure that the data is aligned in the time dimension. At the same time, a spatial mapping matrix is ​​established based on the equipment topology relationship to achieve feature alignment in the spatial dimension. The improved DS evidence theory fusion formula is used to fuse multi-source data: ,in Assign the basic probability of proposition A after fusion, 、 Different data sources are used to formulate propositions 、 The basic probability assignment of is the data source credibility weight, 、 This is an index parameter used to distinguish different propositions and represents the elements of the proposition set from different data sources. It is determined by evaluating the accuracy and stability of historical data. By introducing data source weights, this formula strengthens the reliance on high-trustworthiness data and improves the reliability of the fusion results.

[0041] To improve data quality, a generative adversarial network (GAN) is introduced to enhance the features of spatiotemporal data. The generator uses a deep convolutional neural network to learn the distribution characteristics of the original data and generate supplementary data that conforms to the actual scenario. The discriminator distinguishes between real data and generated data through a binary classification task. Through adversarial training, the distribution of generated data is highly consistent with real data, increasing data diversity and integrity. At the same time, a fusion quality assessment indicator system based on fuzzy comprehensive evaluation is established. The evaluation formula is: Where Q is the comprehensive score of fusion quality, and n is the number of evaluation indicators, including data consistency, completeness, timeliness, etc. is the weight of the i-th indicator, which is determined by the hierarchical analysis method-fuzzy entropy weight method. The subjective weight is calculated by constructing a judgment matrix, and the objective weight is calculated by combining the data entropy value. Finally, the combined weight is obtained by weighted average; Score the i-th indicator and convert the original indicator value into a score through the fuzzy membership function. This system ensures that the fused data meets the needs of subsequent decision-making.

[0042] Steps for in-depth assessment of equipment health status: Use capsule networks to dynamically learn equipment features through routing. The input layer of the capsule network receives pre-processed multi-dimensional feature data, such as electrical parameters, temperature, vibration, etc., extracts local features through the primary capsule layer, and then aggregates similar features into advanced capsules through the dynamic routing algorithm to achieve a high-level abstract representation of the equipment health status. The network can transfer the health assessment knowledge of similar equipment to the target equipment, reduce the sample requirements for new equipment assessment through transfer learning, and improve assessment efficiency. The equipment health index calculation formula is: , where m is the number of evaluation indicators, is the weight of the kth indicator, which is determined by combining expert experience and historical data statistics; is the actual value of the kth indicator, 、 are the maximum and minimum values ​​of the kth indicator respectively. This formula normalizes each indicator, maps it to the interval [0,1], and then obtains the health index by weighted summation. The smaller the value, the better the health status of the device.

[0043] A hybrid model based on variational autoencoders (VAE) and long short-term memory networks (LSTM) is used to predict equipment failures and estimate remaining life. VAE is used to extract the potential distribution of equipment features, while LSTM captures the temporal variation of features. The two are combined to build a prediction model. The remaining life calculation formula is ,in are the mean and standard deviation of the normal state characteristics of the equipment, respectively, obtained through statistics of historical normal operation data; This formula estimates the remaining useful life of a device by calculating the deviation of the current characteristics from the normal state. Based on the device health index and combined with the fuzzy C-means clustering algorithm, the device health status is dynamically classified into four levels: normal, caution, abnormal, and critical. This provides a scientific basis for equipment maintenance, such as triggering emergency maintenance procedures for equipment rated "critical."

[0044] Dynamic modeling steps for preventing misoperation rules: Create a dynamic colored Petri net (DCPN) model. The model's place nodes represent device states, such as circuit breaker open / close and grounding switch status, and transition nodes represent operational actions, such as closing and opening. Color sets are used to distinguish different types of devices and operations. The parameters and connections between places and transitions are dynamically adjusted based on the real-time device status and operational tasks. For example, when a circuit breaker is under maintenance, the triggering of the related closing transition is prohibited. Simultaneously, a device knowledge graph is constructed, with device entities as nodes and device attributes, failure modes, and operational rules as edges. A graph neural network (GNN) is used for knowledge reasoning to extract implicit relationships between devices.

[0045] The operation sequence is formally verified using a temporal logic verification algorithm. This algorithm converts operation rules into linear temporal logic (LTL) formulas, such as "before closing the circuit breaker, the grounding switch must be in the open state." Model checking techniques are used to verify that the operation sequence satisfies all safety constraints, preventing incorrect operations. An incremental learning-based operation rule update mechanism is established. When the device structure changes, the operation process is adjusted, or a new fault case occurs, this change information is automatically collected. The incremental learning algorithm is used to update the DCPN model's place transition parameters and the node-edge relationships in the knowledge graph. Knowledge distillation techniques are used to transfer new knowledge from the teacher model to the student model, reducing the computational overhead of model updates.

[0046] A genetic algorithm-based conflict resolution strategy was designed, and a conflict resolution model was constructed with operational safety, efficiency, and cost as optimization goals. When multiple operational tasks simultaneously request equipment resources, the operation sequence is encoded as a chromosome. Using conflict probability, completion time, and labor cost as fitness functions, selection, crossover, and mutation operations are used to find the optimal operation sequence, resolve resource conflicts, and ensure safe and efficient execution.

[0047] Risk early warning and decision optimization steps: Building a dynamic risk assessment matrix ,in represents the risk value of the i-th risk factor in the j-th scenario, m represents the number of risk factors, and n represents the number of scenarios. ,in, is the risk correction factor, which takes into account the impact of equipment aging, environmental factors, etc. on risk; The probability of risk occurrence is calculated through historical data statistics and Bayesian updating; The severity of risk consequences is determined based on the importance of the equipment, the scope of the fault impact, etc. This matrix quantifies the potential impact of different risk factors in various scenarios and provides data support for risk warning.

[0048] A dynamic risk scenario deduction model based on cellular automata (CA) and Monte Carlo simulation is constructed. The cellular automata divides the network of substation equipment into regular grids, and each cell represents a device. The neighbor rules of cells are defined according to the electrical connection relationship between devices, and the propagation and diffusion process of risk in the device network is simulated, such as the spread of short circuit fault from a circuit breaker to adjacent devices. The Monte Carlo simulation processes the uncertainty of risk factors through random sampling, such as the randomness of device failure rate, and generates the probability distribution of risk propagation through multiple simulations. A hybrid reasoning mechanism based on case-based reasoning (CBR) and rule-based reasoning (RBR) is adopted. When the risk scenario is detected, the most similar historical scenario is matched by retrieving similar cases from the historical case base through case similarity calculation, and the corresponding decision scheme is extracted. If there is no matching case in the case base, the rule reasoning is triggered to generate candidate decision schemes from the rule base.

[0049] The intelligent interlocking strategy collaborative control steps are as follows: a hierarchical distributed intelligent interlocking control system is designed, which has three layers of architecture, i.e. station control layer, bay layer and device layer. The station control layer is responsible for global strategy optimization and coordination, the bay layer executes the interlocking logic in the region, and the device layer implements local operation interlocking. The double deep Q network (DDQN) algorithm in deep reinforcement learning is used to optimize the interlocking strategy, and the optimal interlocking decision is obtained through the interaction between the agent and the environment. The reward function is defined to maximize the comprehensive performance of the system: where S is the operation safety score, which is calculated based on whether the operation violates the anti-misoperation rules, risk assessment results, etc.; E is the operation efficiency score, which considers the operation process time consumption and resource utilization rate, etc.; T is the operation timeliness score, which measures the operation response speed; are weight coefficients, which are dynamically adjusted according to the operation requirements of the wind farm, such as prioritizing safety during normal operation and focusing on timeliness in emergency situations.

[0050] The cultural gene algorithm is used to adaptively evolve the interlocking strategy of DDQN. This algorithm combines the global search genetic algorithm with the local search hill climbing algorithm. In the algorithm iteration process, the crossover and mutation operations of the genetic algorithm are used to explore the solution space, and then the hill climbing optimization is performed on the local excellent solutions to speed up the strategy convergence. In the hierarchical distributed system, a redundancy mechanism is designed, such as hardware redundancy of key nodes and dual-channel backup of data transmission, to ensure the reliable operation of the system in the case of partial component failure and improve the reliability of the interlocking control.

[0051] Visual Decision Support and Virtual Simulation Steps: Build a digital twin model of the substation. Using 3D modeling technology, perform a 1:1 digital reconstruction of the substation equipment and site, integrating information such as the equipment's geometric model, electrical parameters, and control logic. Integrate gesture and voice recognition technologies into the digital twin model, allowing operators to interact naturally with virtual devices through gestures, such as waving to simulate switch operation, or voice commands, enhancing intuitiveness and convenience.

[0052] The physics engine simulates the mechanical characteristics and physical responses of equipment during operation, such as the mechanical impact of closing a circuit breaker and the arcing effect of switch contacts. This provides operators with a realistic operational experience, assisting with training and decision-making exercises. The generated decision plan is verified using a digital twin model. By inputting the decision plan's operational sequence, the system simulates changes in equipment status and the evolution of risks during implementation. For example, it simulates the power flow distribution and potential overload risks after equipment load transfer under a specific maintenance plan. This allows the feasibility and effectiveness of the decision plan to be evaluated, providing a basis for solution optimization.

[0053] The present invention also includes continuous system optimization and knowledge accumulation steps: establishing a system performance evaluation system and comprehensively evaluating system performance using principal component analysis (PCA) and grey correlation analysis. PCA is first used to reduce the dimensionality of multi-dimensional performance indicators, such as data processing delay, fault warning accuracy, and blocking strategy response time, to extract the key influencing factors. Grey correlation analysis is then used to calculate the correlation between each indicator and the ideal optimal value to obtain a comprehensive evaluation score. Based on the evaluation results, weak links in the system are identified. For example, if the processing delay of the data fusion module is found to be excessive, the model parameters or hardware configuration of the LSTM and GCN are optimized, providing a clear direction for system optimization.

[0054] Natural language processing technology is used to extract knowledge from technical documents, operating manuals, and fault reports. Named entity recognition (NER) technology is used to identify entities such as equipment names, fault types, and operating steps. The relationship between entities is extracted through relationship extraction algorithms. Combined with equipment operation data, the knowledge graph is automatically constructed and updated to achieve continuous accumulation and reuse of knowledge and improve the intelligence level of the system.

[0055] The present invention also includes emergency response and rapid recovery steps: Intelligent identification of emergency scenarios: Constructing an emergency scenario recognition model based on convolutional neural network (CNN) and bidirectional long short-term memory network (Bi-LSTM). CNN is used to extract spatial features such as waveforms and images during equipment failures, and Bi-LSTM is used to capture the bidirectional dependencies of time series data. The model uses an attention mechanism to automatically focus on the most critical features for scene recognition (such as current mutation features during short-circuit faults), thereby improving the recognition accuracy of emergency scenarios.

[0056] Once an emergency scenario is identified, a recovery strategy is generated based on the Targeted Particle Swarm Optimization (TOPSO) algorithm. This algorithm uses the recovery objective as the optimization goal, along with constraints such as device state and operating rules, to iteratively search through a particle swarm to find the optimal sequence of recovery operations. Particle positions are encoded as the recovery operation sequence, and the velocity update formula incorporates the objective function gradient information to accelerate convergence, ensuring rapid generation of a scientifically sound recovery strategy in emergency situations and minimizing losses.

[0057] The above are only preferred specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A wind farm booster station multi-source data fusion anti-error locking intelligent decision-making and early warning method, characterized by: The following steps are involved: S1: Multi-source heterogeneous data collection and preprocessing steps: Deploy edge computing nodes and IoT sensor arrays to collect data, and use abnormal data filtering algorithms combined with lightweight deep learning models to perform real-time noise reduction and feature extraction on the data; S2: Cross-scale spatiotemporal data fusion step: Construct a fusion model of long short-term memory network and graph convolution network, synchronize the data through spatiotemporal alignment algorithm, and adopt the improved DS evidence theory fusion formula: , Assign a value to the basic probability of proposition A, 、 Proposition 、 The basic probability assignment of is the data source credibility weight, 、 It is an index parameter used to distinguish different propositions and is used to represent the proposition set elements in different data sources; S3: Device health status in-depth assessment step: Use capsule networks to dynamically learn device features and transfer similar device health assessment knowledge to the target device. The device health index calculation formula is: , m is the number of evaluation indicators, is the indicator weight, is the actual value of the indicator, 、 are the maximum and minimum values ​​of the kth indicator respectively; S4: Dynamic modeling of anti-misoperation rules: Create a dynamic colored Petri net and adjust the parameters and connections between places and transitions based on the real-time status of the equipment and the operation tasks. Place nodes represent equipment status, and transition nodes represent operation actions. The knowledge graph integrates equipment knowledge, performs knowledge reasoning through a graph neural network, and combines it with a temporal logic verification algorithm to formally verify the operation sequence. S5: Risk warning and decision optimization steps: Propose a risk warning method based on an improved evidence neural network, integrate DS evidence theory into the neural network structure, and enhance the model's ability to process uncertain information; S6: Intelligent locking strategy collaborative control steps: Design a hierarchical distributed intelligent locking control system and use the dual deep Q network algorithm in deep reinforcement learning to optimize the locking strategy; S7: Visual decision support and virtual simulation steps: Build a digital twin model of the substation and combine it with MR equipment to provide immersive operation and training for operators.

2. The wind farm booster station multi-source data fusion anti-error locking intelligent decision-making and early warning method according to claim 1 is characterized in that: The multi-source heterogeneous data collection and preprocessing steps also include: developing an adaptive protocol conversion gateway to support dynamic parsing and conversion of industrial communication protocols, using a protocol feature fingerprint recognition algorithm to automatically identify the communication protocol type of the access device, and building an edge node task scheduling model based on game theory to allocate data collection and preprocessing tasks according to the node status.

3. The wind farm booster station multi-source data fusion anti-error locking intelligent decision-making and early warning method according to claim 1 is characterized in that: The cross-scale spatiotemporal data fusion step also includes: introducing a generative adversarial network to enhance the features of spatiotemporal data, the generator learns the distribution characteristics of the original data and generates supplementary data; the discriminator distinguishes between real data and generated data, and through adversarial training, increases the diversity and integrity of the data, and establishes a fusion quality evaluation index system based on fuzzy comprehensive evaluation. The evaluation formula is: , where Q is the comprehensive score of fusion quality, n is the number of evaluation indicators, is the weight of the i-th indicator, The score of the i-th indicator is obtained, and the weight is determined by the hierarchical analysis method-fuzzy entropy weight method to ensure that the fused data meets the decision-making requirements.

4. The wind farm booster station multi-source data fusion anti-error locking intelligent decision-making and early warning method according to claim 1 is characterized in that: The device health status in-depth assessment step also includes: predicting device failures and evaluating remaining life based on a hybrid model of a variational autoencoder and a long short-term memory network. The remaining life calculation formula is: ,in are the mean and standard deviation of the normal state characteristics of the equipment, In order to predict the equipment characteristic values, the equipment health status is dynamically divided into four levels according to the equipment health index and combined with the fuzzy C-means clustering algorithm, providing a scientific basis for equipment maintenance.

5. The wind farm booster station multi-source data fusion anti-error locking intelligent decision-making and early warning method according to claim 1 is characterized in that: The step of dynamic modeling of anti-misoperation rules also includes: dynamic updating of operation rules: establishing an operation rule updating mechanism based on incremental learning. When the equipment structure changes, the operation process is adjusted, or a new fault case occurs, the dynamic colored Petri net model and the knowledge graph are automatically updated. The knowledge distillation technology is used to integrate new knowledge into the existing model, and an operation conflict resolution strategy based on a genetic algorithm is designed. With operation safety, efficiency, and cost as the optimization goals, a conflict resolution model is constructed.

6. The wind farm booster station multi-source data fusion anti-error locking intelligent decision-making and early warning method according to claim 1 is characterized in that: The risk warning and decision optimization steps construct a dynamic risk assessment matrix ,in represents the risk value of the i-th risk factor in the j-th scenario, m represents the number of risk factors, and n represents the number of scenarios. , is the risk correction factor, is the probability of risk occurrence, The severity of the risk consequences; this step also includes: dynamic deduction of risk situation: constructing a dynamic deduction model of risk situation based on cellular automaton and Monte Carlo simulation, cellular automaton simulates the process of risk propagation and diffusion in the substation equipment network, Monte Carlo simulation handles the uncertainty of risk factors, and a hybrid reasoning mechanism based on case reasoning and rule reasoning, according to the current risk situation, retrieves similar cases and rules from the historical case library and rule library to generate candidate decision plans.

7. The wind farm booster station multi-source data fusion anti-error locking intelligent decision-making and early warning method according to claim 1 is characterized in that: The intelligent locking strategy collaborative control step is achieved through the reward function Maximize the overall performance of the system, where S is the operational safety score, E is the operational efficiency score, and T is the operational timeliness score. is the weight coefficient; this step also includes: adaptive evolution of the locking strategy: the cultural gene algorithm is used to adaptively evolve the locking strategy of the dual deep Q network. During the algorithm iteration process, local search and global search are combined to accelerate the strategy convergence speed, and a redundancy mechanism is designed in the hierarchical distributed intelligent locking control system.

8. The wind farm booster station multi-source data fusion anti-error locking intelligent decision-making and early warning method according to claim 1 is characterized in that: The visual decision support and virtual simulation steps also include: integrating gesture recognition and voice recognition technology into the digital twin model to achieve natural interaction between operators and virtual equipment, simulating the mechanical properties and physical responses of the equipment during operation through a physical engine to provide operators with a realistic operating experience, using the digital twin model to verify the generated decision plan, and evaluating the feasibility and effectiveness of the decision plan by simulating the changes in equipment status and risk evolution during the implementation of the decision plan.

9. The wind farm booster station multi-source data fusion anti-error locking intelligent decision-making and early warning method according to claim 1 is characterized in that: It also includes steps for continuous system optimization and knowledge accumulation: establishing a system performance evaluation system, using principal component analysis-grey correlation analysis method to comprehensively evaluate system performance, locating system weaknesses based on the evaluation results, providing direction for system optimization, and using natural language processing technology to extract knowledge from technical documents, operating manuals, and fault reports, combined with equipment operation data to automatically build and update knowledge graphs.

10. The wind farm booster station multi-source data fusion anti-error locking intelligent decision-making and early warning method according to claim 1 is characterized in that: It also includes emergency response and rapid recovery steps: building an emergency scenario recognition model based on convolutional neural networks and bidirectional long short-term memory networks to identify emergency scenarios. The model uses an attention mechanism to highlight key features. Once the emergency scenario is identified, a recovery strategy is generated based on the target particle swarm optimization algorithm.

Images