Intelligent decision-making method for preventing misoperation locking of booster station of wind power plant based on multi-source data fusion

Abstract

The invention discloses a wind power plant booster station misoperation-preventive locking intelligent decision-making method based on multi-source data fusion, and relates to the technical field of misoperation-preventive locking decision-making. Multi-source data such as electrical equipment states, environmental parameters and equipment appearance images are collected and preprocessed; extracting multi-dimensional features, and constructing a space-time incidence matrix; constructing a power equipment knowledge graph, and converting safety regulations into rules and reasoning; a three-branch neural network is constructed, multi-modal features are fused, and a decision model is trained; according to the risk assessment value and the dynamic threshold value, an anti-misoperation locking decision is executed, a permission or locking instruction is output, and an operation log is recorded. By fusing multi-source data and combining the knowledge graph, deep learning and other technologies, the wind power plant booster station anti-misoperation locking intelligent decision is realized, the fault early warning accuracy and decision efficiency are improved, the system flexibility and interpretability are enhanced, and safe and reliable operation of the booster station is guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mistaken lockout decision-making, and particularly relates to a mistaken lockout intelligent decision-making method for a wind farm booster station based on multi-source data fusion. BACKGROUND

[0002] With the increasing demand for clean energy worldwide, wind farms are expanding in size, and the safe and stable operation of booster stations, which are the core hubs for power collection and transmission in wind farms, is crucial. Traditional mistaken lockout systems for wind farm booster stations mainly rely on single SCADA system data or fixed logic lockout rules, which are difficult to cope with complex and variable operating conditions. For example, relying solely on electrical parameters for operation lockout judgment cannot timely detect potential risks caused by environmental factors (such as high temperature, high humidity, and abnormal vibration) of equipment, resulting in increased risks of misoperation or missed operation. At the same time, fixed logic rules lack flexibility and scalability, making it difficult to adapt to scenarios such as the access of new equipment and changes in operating modes.

[0003] Existing mistaken lockout technologies have many limitations. On the one hand, the data sources are single, and multi-source information such as environmental parameters and equipment appearance states is not fully fused. Environmental factors have a significant impact on the insulation performance and mechanical strength of electrical equipment, but traditional systems do not consider these factors in decision-making, making it difficult to capture abnormal signs before equipment failure. On the other hand, the decision-making model lacks intelligence and is mainly based on static rule libraries, which cannot adapt to the dynamic changes in equipment states during wind farm operation. For example, in extreme weather, the electrical and mechanical stresses on equipment change, and traditional rules cannot adjust the mistaken lockout strategy in real time, resulting in insufficient timeliness and accuracy of mistaken lockout. In addition, existing systems lack explainability of decision results, making it difficult for operation and maintenance personnel to understand the basis for lockout instructions, increasing the risk of operation and difficulty in troubleshooting.

[0004] In recent years, artificial intelligence and Internet of Things technologies have been gradually applied to power systems, but there are still deficiencies in their application to mistaken lockout in wind farm booster stations. Some mistaken lockout systems based on machine learning have problems such as poor model generalization ability and insufficient training data, and perform poorly in actual complex operating conditions. At the same time, the fusion of multi-source heterogeneous data faces challenges such as data synchronization difficulty and complex feature correlation, and there is no effective solution yet. With the increasing demand for intelligent development of wind farms, there is an urgent need for a mistaken lockout method that can fuse multi-source data, have intelligent decision-making capabilities, and be explainable, to ensure the safe and reliable operation of booster stations. SUMMARY

[0005] The mistaken lockout intelligent decision-making method for a wind farm booster station based on multi-source data fusion proposed by the present application solves the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent decision-making method for preventing misoperation during interlocking of wind farm booster stations based on multi-source data fusion, comprising:

[0007] Multi-source heterogeneous data acquisition and preprocessing steps: Acquire real-time status data of electrical equipment through the SCADA system. scada =[I t U t P t Q t ], where I t U is the current value. t P is the voltage value. t For active power, Q t Reactive power; environmental parameters D are collected by an IoT sensor network installed on the surface of the equipment. env =[T t H t S t V t ], where T t For temperature, H t For humidity, S t V is the vibration frequency. t Noise level; Equipment appearance image data D is acquired through the inspection robot's vision system. img ={F1, F2, ..., F n Design a data preprocessing algorithm to synchronize the data in time using the formula t. sync =t ref +Δt i Implementation, where t ref For reference timestamp, Δt i The time offset of the i-th data source;

[0008] Multi-dimensional feature extraction and spatiotemporal correlation steps: Constructing electrical feature tensors Among them W elec A learnable weight matrix is ​​used; an environmental feature correlation function R is designed. env =σ(W env ·D env +b env ), where σ is the activation function; target detection and feature extraction are performed on the image data, and the device status is identified using the YOLOv5 model to extract the visual feature vector F. vis Constructing a spatiotemporal correlation matrix Where t i , t j For timestamps, x i x j Let λ be the spatial coordinates, and λ be the time decay factor.

[0009] Knowledge graph construction and rule reasoning steps: based on IEC61850 standard, the knowledge graph G=(V, E) of power equipment is constructed, wherein the node set V={v device , v state , v operation , v constraint} and the edge set E represent the semantic relationship between nodes; a misoperation prevention rule expression framework is designed, and the power safety regulations are converted into executable rules R={r1, r2, …, r n}, each rule is formalized as r i :(C1∧C2∧…∧C m )→A, wherein C j is the condition, and A is the action; a rule reasoning engine is constructed, and rule reasoning is realized through Datalog language, and rule satisfaction degree is calculated wherein w i is the rule weight, and δ(r i ) is the rule matching function;

[0010] Multi-modal fusion decision model training steps: a three-branch neural network architecture is constructed, including an electrical data branch NN elec , an environmental data branch NN env and a visual data branch NN vis ; a feature fusion layer is designed, and multi-modal feature fusion is realized through the formula F fusion =W1·F elec +W2·F env +W3·F vis +β·(F elec ⊙F env ⊙F vis ), wherein F elec , F env , F vis are the output features of each branch, and ⊙ is Hadamard product; a decision output layer is constructed, and decision probability distribution P action =Softmax(W out ·F fusion +b out ) is calculated through Softmax function; a hybrid loss function L total =α1L CE +α2L reg +α3L graph is designed, wherein the cross-entropy loss the regularization loss the graph structure loss

[0011] Adaptive misoperation prevention locking decision execution steps: according to the current device state and operation request, the risk assessment value is calculated wherein S iis the score of the ith risk factor; a self-adaptive threshold adjustment mechanism is designed to dynamically adjust the decision threshold through the formula τ adaptive base +Δτ, where Δτ=γ·σ(F fusion ), and σ is the feature variance; when R risk >τ adaptive , the false operation prevention locking mechanism is triggered to generate the locking instruction I lock ={device id , operation type , reason code}; when R risk ≤τ adaptive , the operation permission instruction I permit is generated, and the operation log Log op =[t op , device id , operation type , F fusion , R risk ] is recorded.

[0012] Further, it further comprises:

[0013] A multi-scale time sequence feature extraction step: multi-scale time windows W={w1, w2, w3} are constructed, where w1=5s, w2=30s, and w3=5min; in each time window, time sequence features are extracted through a bidirectional LSTM network A time attention mechanism is designed, and the attention weights of different time scales are calculated through the formula Multi-scale time sequence features are fused

[0014] A device failure prediction and early warning step: a device health index HI is constructed; a gradient boosting tree algorithm is used to realize it, where f is a mapping function; a failure development trend prediction model is designed, and the health index at future time k is predicted through the formula , where ΔHI i is the health index change amount; a set of early warning thresholds {τ warn1 , τ warn2 , τ warn3} is set, and when HI t <τ warn3 , an emergency warning is triggered to generate a repair work order Work order ={device id , fault type , priority level , suggestion}.

[0015] ​Furthermore, an anomaly detection mechanism is designed in the multi-scale temporal feature extraction step, which calculates the Mahalanobis distance between the current temporal features and the historical normal patterns. Anomaly detection, where μ is the mean vector and S is the covariance matrix; when D M (x)>τ anomaly The time stamp is marked as an abnormal state and the data cleaning process is triggered. The abnormal data is corrected by a sliding window mid-value filtering algorithm.

[0016] Furthermore, a dynamic weight adjustment mechanism is introduced into the knowledge graph construction and rule reasoning steps, dynamically adjusting the rule weights according to the device operating status, through formulas. Implementation, in which As the initial weights, Let be the state change of the i-th rule; design a rule conflict resolution algorithm, and select the rule when both rules are triggered simultaneously. The maximum value rule is executed.

[0017] Furthermore, in the equipment fault prediction and early warning step, a fault propagation path model is constructed, and the fault propagation probability P is calculated using a graph convolutional network. propagate =GCN(A fault F fault ), where A fault Let F be the fault correlation matrix. fault The fault feature vector is used to generate a cascaded early warning strategy based on the fault propagation path, which issues early warnings to the affected equipment in advance.

[0018] Furthermore, a federated learning framework is introduced into the training steps of the multimodal fusion decision model, deploying client models M on edge nodes. edge The central server deploys the global model M. center Design federated learning aggregation strategies Where n i Let ΔM be the data volume of the i-th edge node, N be the total data volume, and ΔM be the data volume of the i-th edge node. prior For prior knowledge gradients in the power sector; Laplace noise is added through a differential privacy mechanism. Noise parameters ∈ through formula Dynamic adjustment, R data This is a data sensitivity indicator.

[0019] Furthermore, in the adaptive anti-misoperation interlocking decision execution step, a decision interpretation module is constructed to calculate the contribution φ of each feature to the decision using SHAP values. i =E[f(z) S∪{i} )-f(z S )], where z S For feature subsets; generate natural language interpretation report Expl report= {feature i , φ i , impact type , threshold i}, to provide decision-making basis for operation and maintenance personnel.

[0020] Further, it also includes:

[0021] Self-learning and evolution step: design operation feedback mechanism, collect the correction information Feedback = {op id , correct label , confidence score} of operation and maintenance personnel on decision result; build incremental learning model, update model parameters through formula θ new = θ old + η·Δθ feedback , wherein Δθ feedback is the feedback gradient; regularly evaluate the model, when the model performance index decreases by more than the threshold, trigger the model retraining process, and update the model with the latest data.

[0022] Visualization and interaction step: build three-dimensional visualization interface, map device state, decision result and fault warning information to three-dimensional model; design data visualization components, including electrical parameter real-time curve, device state heat map and fault propagation path animation; realize man-machine interaction decision function, when the system generates locking instruction, provide alternative operation scheme Alt ops = {op1, op2, …, op n}, and calculate the risk score of each alternative for reference decision of operation and maintenance personnel.

[0023] Compared with the existing technology, the beneficial effects of the present application are:

[0024] By fusing electrical parameters, environmental data, device appearance images and other multi-source information, the device running state can be comprehensively perceived, and potential risks can be captured in advance. For example, by including environmental parameters such as device vibration and temperature in decision analysis, abnormality caused by mechanical failure or overheating can be effectively identified. Compared with traditional methods, the fault warning accuracy is improved by more than 30%, and the unplanned downtime caused by device failure is reduced.

[0025] The method adopts knowledge graph and rule reasoning technology, converts power safety regulations into executable rules, and realizes adaptive optimization of rules through dynamic weight adjustment. It can quickly adapt to device changes, operation mode adjustment and other scenarios, and the applicability and flexibility of the anti-misoperation locking rule are greatly improved. When new devices are connected, the system can automatically update the knowledge graph and generate matching anti-misoperation strategy, without the need for manual large-scale modification of rule library, and the system debugging period is shortened by more than 50%.

[0026] The multi-modal fusion decision model combines a deep learning algorithm, fully extracts multi-source data features through a multi-branch network and a feature fusion layer, and the operation decision accuracy is improved by 25% compared with a single data model. The adaptive anti-misoperation locking decision mechanism can dynamically adjust the decision threshold according to real-time risks, reduce unnecessary locking operations under the premise of ensuring safety, and improve the operation efficiency of the equipment. At the same time, the decision explanation module provides intuitive decision basis for operation and maintenance personnel through SHAP value analysis, reduces the risk of operation misunderstanding, and improves the fault troubleshooting efficiency.

[0027] In addition, the system has self-learning and evolution ability, collects operation and maintenance feedback and incremental learning, continuously optimizes the decision model, and continuously improves the performance with running time. The visualization and interaction function displays the system information in a three-dimensional interface and multi-dimensional charts, and provides alternative operation schemes to assist operation and maintenance personnel to make scientific decisions, and promote the development of the wind power plant booster station in the direction of intelligentization and safety. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 A schematic block diagram of the wind power plant booster station anti-misoperation locking intelligent decision method based on multi-source data fusion proposed by the present application is shown in the figure.

[0029] Figure 2 A schematic diagram of the fault early warning accuracy comparison of the wind power plant booster station anti-misoperation locking intelligent decision method based on multi-source data fusion proposed by the present application is shown in the figure.

[0030] Figure 3 A schematic diagram of the operation response time of different methods of the wind power plant booster station anti-misoperation locking intelligent decision method based on multi-source data fusion proposed by the present application with data volume change is shown in the figure. DETAILED DESCRIPTION

[0031] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0032] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, which is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application.

[0033] In addition, the terms "first", "second" are only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited. In addition, the terms "mounting", "connection", "connection" should be broadly understood, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be the communication between two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances, and the present application will be further described in detail with reference to the drawings.

[0034] Reference Figure 1 and Figure 3 : A specific embodiment of an intelligent decision-making method for preventing misoperation locking of a wind power plant booster station based on multi-source data fusion.

[0035] The specific embodiment of the method of the present application takes a certain 50MW wind power plant booster station as the application scenario, which is described in detail as follows:

[0036] Multi-source heterogeneous data acquisition and preprocessing

[0037] Thirty-two electrical equipment state monitoring terminals are deployed in the booster station, and real-time data such as current I t , voltage U t , active power P t , and reactive power Q t are collected from the SCADA system through the Modbus protocol, and the sampling frequency is 10Hz. Temperature and humidity sensors, vibration sensors and noise sensors are installed on the surface of key equipment to collect environmental parameters D env =[T t , H t , S t , V t, the sampling frequency is 1 Hz. Two inspection robots are configured, carrying an 8 million pixel camera, which collects an appearance image of the whole station equipment every 30 minutes. img .

[0038] The data preprocessing adopts a time synchronization algorithm sync = t ref + Δt i , taking the time output by the GPS timing module as the reference t ref , and calculating the time offset Δt i of each data source through the cross-correlation function. The electrical data is filtered using a sliding window with a window size of 5 sampling points; the image data is preprocessed using histogram equalization and Gaussian filtering.

[0039] Multi-dimensional feature extraction and spatio-temporal correlation: Constructing an electrical feature tensor where W elec is a 64x128 learnable weight matrix. The environmental feature correlation function R env = σ(W env ·D env + b env ), and the activation function σ uses ReLU. The image feature extraction uses a pre-trained YOLOv5 model to extract a 1024-dimensional feature vector F vis .

[0040] Spatio-temporal correlation matrix where the time decay factor λ = 0.1, and the spatial distance is calculated by the device physical location coordinates. Through this matrix, the correlation between device states in the spatio-temporal dimension is established.

[0041] Knowledge graph construction and rule reasoning: Based on the IEC61850 standard, a knowledge graph G = (V, E) is constructed, containing 500 device nodes, 200 state nodes, 100 operation nodes, and 80 constraint nodes. The "Electric Power Safety Work Procedures" are converted into 120 anti-misoperation rules R, such as:

[0042] r1: (C1 ∧ C2 ∧ C3) → A1 where C1 is "the circuit breaker is in the closed state", C2 is "the line is live", C3 is "the grounding knife switch is not pulled open", and A1 is "prohibit operating the line side grounding knife switch".

[0043] The rule reasoning engine is implemented using the Datalog language, and the rule satisfaction calculation is where the rule weight w i is determined through expert evaluation and historical data statistics, and the matching function δ(r i ) returns 0 or 1.

[0044] Multi-modal fusion decision model training: The model adopts a three-branch architecture, the electrical branch NN elec contains 3 fully connected layers, the environmental branch NN env contains 2 LSTM layers, the visual branch NN vis Based on ResNet50 modification. The feature fusion formula is:

[0045] F fusion = W1·F elec + W2·F env + W3·F vis + β·(F elec ⊙F env ⊙F vis )

[0046] Where W1, W2, W3 are fusion weight matrices, β = 0.2. The decision output layer calculates the operation probability distribution P action by the Softmax function operation. The hybrid loss function is: L total = α1L CE + α2L reg + α3L graph , where α1 = 0.6, α2 = 0.2, α3 = 0.2. The model is trained using the Adam optimizer, the learning rate is 0.001, and the training period is 100 epochs.

[0047] Adaptive false alarm prevention locking decision execution: The risk assessment value is calculated as Where the risk factors include device state abnormalities, environmental parameter overruns, etc. The adaptive threshold adjustment formula is τ adaptive = τ base + Δτ, the basic threshold τ base = 0.7, Δτ = γ·σ(F fusion ), γ = 0.3.

[0048] When R risk > τ adaptive , the system generates a locking instruction I lock , and notifies the operation and maintenance personnel through sound and light alarm. When R risk ≤ τ adaptive , the operation permission instruction I permit is generated, and the operation log Log op is recorded.

[0049] Implementation effect data representation

[0050]

[0051]

[0052] The above data show that the method has significant advantages in fault early warning, operation safety, system response speed, etc. The fault early warning accuracy is improved by nearly 40%, effectively reducing the risk of equipment damage; the number of misoperations is greatly reduced, improving the operation reliability of the booster station; the operation response time is shortened by 70%, meeting the rapid decision-making demand; the rule maintenance cost is significantly reduced, reflecting the advantages of dynamic adjustment of the knowledge graph; the adaptability of extreme working conditions is improved by nearly one time, proving the effectiveness of multi-source data fusion. These improvements provide a strong guarantee for the safe and stable operation of the booster station of the wind farm, and have important engineering application value.

[0053] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can make equivalent replacements or changes within the technical range disclosed by the present application according to the technical solution and inventive concept of the present application, which should be covered within the protection scope of the present application.

Claims

1. A wind farm booster station anti-misoperation locking intelligent decision method based on multi-source data fusion, characterized in that, Comprise: Multi-source heterogeneous data collection and preprocessing steps: Obtain real-time state data D of electrical equipment through SCADA system scada = [I t , U t , P t , Q t ], where I t is the current value, U t is the voltage value, P t is the active power, and Q t is the reactive power; Install Internet of Things sensor network on the surface of the equipment to collect environmental parameters D env = [T t , H t , S t , V t ], where T t is the temperature, H t is the humidity, S t is the vibration frequency, and V t is the noise level; Obtain equipment appearance image data D img = {F1, F2,..., F n} through a visual inspection robot system; Design a data preprocessing algorithm to time synchronize each data, and achieve it through the formula t sync = t ref + Δt i , where t ref is the reference timestamp, and Δt i is the time offset of the ith data source; Multi-dimensional feature extraction and space-time correlation step: construct electrical feature tensor wherein W elec is a learnable weight matrix; design environment feature correlation function R env = σ(W env ·D env +b env ), wherein σ is an activation function; target detection and feature extraction are performed on image data, device status is identified through a YOLOv5 model, and a visual feature vector F vis is extracted; construct a space-time correlation matrix wherein t i , t j are time stamps, x i , x j are spatial position coordinates, and λ is a time decay factor; Knowledge graph construction and rule reasoning steps: based on IEC61850 standard to construct power equipment knowledge graph G=(V, E), wherein node set V={v device , v state , v operation , v constraint}, edge set E represents the semantic relationship between nodes; design a misoperation prevention rule expression framework, convert the power safety regulations into executable rules r={r1, r2, …, r n}, each rule is formalized as r i :(C1∧C2∧…∧C m )→A, wherein C j is the condition, A is the action; construct a rule reasoning engine, implement rule reasoning through Datalog language, calculate rule satisfaction degree wherein w i is the rule weight, δ(r i ) is the rule matching function; Multimodal fusion decision model training steps: a three-branch neural network architecture is constructed, including an electrical data branch NN elec , an environmental data branch NN env , and a visual data branch NN vis ; A design feature fusion layer is fused by formula F fusion = W1·F elec + W2·F env + W3·F vis + β·(F elec ⊙ F env ⊙ F vis ) to realize multi-modal feature fusion, wherein F elec , F env , F vis are respectively each branch output feature, and ⊙ is Hadamard product; a decision output layer is constructed, a decision probability distribution P action = Softmax(W out ·F fusion + b out ) is calculated by a Softmax function; a hybrid loss function L total = α1L CE + α2L reg + α3L graph is designed, wherein a cross-entropy loss a regularization loss and a graph structure loss 2. The intelligent decision method for preventing misoperation locking of a wind farm booster station based on multi-source data fusion according to claim 1, characterized in that, Also comprise: Adaptive anti-misoperation lock decision execution step: according to the current device state and operation request, calculate risk assessment value Where S i is the score of the i th risk factor; design adaptive threshold adjustment mechanism, dynamically adjust the decision threshold through the formula τ adaptive = τ base + Δτ, where Δτ = γ·σ(F fusion ), where σ is the characteristic variance; when R risk > τ adaptive , trigger the anti-misoperation lock mechanism, generate lock instruction I lock = {device id , operation type , reason code}; when R risk ≤ τ adaptive , generate operation permission instruction I permit , and record operation log Log op = [t op , device id , operation type , F fusion , R risk ]; A multi-scale time sequence feature extraction step: a multi-scale time window W = {w1, w2, w3} is constructed, wherein w1 = 5s, w2 = 30s, and w3 = 5min; in each time window, a time sequence feature is extracted through a bidirectional LSTM network A time attention mechanism is designed, and the attention weight of different time scales is calculated through the formula ​ Fusing multi-scale temporal features 3.The intelligent decision method for preventing misoperation of a wind farm booster station based on multi-source data fusion according to claim 1, characterized in that, Also comprise: Device failure prediction and early warning step: constructing a device health index implemented by a gradient boosting tree algorithm, where f is a mapping function; A fault development trend prediction model is designed, and a formula is provided to predict the health index at a future k time point, where ΔHI i is a health index change amount; a plurality of early warning threshold values {τ warn1 , τ warn2 , τ warn3} are set, an emergency early warning is triggered when HI t < τ warn3 , and a maintenance work order Work order ={device id , fault type , priority level , suggestion} is generated.

4. The intelligent decision method for preventing misoperation locking of a wind farm booster station based on multi-source data fusion according to claim 2, characterized in that, The multi-scale time sequence feature extraction step designs an anomaly detection mechanism by calculating Mahalanobis distance between current time sequence feature and historical normal mode detecting the anomaly, wherein μ is a mean vector, S is a covariance matrix; when D M (x) > τ anomaly When the time sequence feature is marked as an abnormal state and a data cleaning process is triggered, the abnormal data is corrected by a sliding window median filtering algorithm.

5. The intelligent decision method for preventing misoperation locking of a wind farm booster station based on multi-source data fusion according to claim 1, characterized in that, The knowledge graph construction and rule reasoning steps incorporate a dynamic weight adjustment mechanism, which dynamically adjusts rule weights based on device operating status using formulas. Implementation, in which As the initial weights, Let be the state change of the i-th rule; design a rule conflict resolution algorithm, and select the rule when both rules are triggered simultaneously. The maximum value rule is executed.

6. The intelligent decision method for preventing misoperation locking of a wind farm booster station based on multi-source data fusion according to claim 3, characterized in that, The fault propagation path model is constructed in the device fault prediction and early warning step, and the fault propagation probability P is calculated through a graph convolution network propagate = GCN(A fault , F fault ), wherein A fault is a fault correlation matrix, and F fault is a fault feature vector; Generate a cascade early warning strategy based on the failure propagation path, and issue an early warning to the affected device.

7. The intelligent decision method for preventing misoperation locking of a wind farm booster station based on multi-source data fusion according to claim 1, characterized in that, The multi-modal fusion decision model training step introduces a federal learning framework, and a client model M is deployed on the edge node edge , a global model M is deployed on the center server center ; a federal learning aggregation strategy is designed Where n i is the data volume of the i th edge node, N is the total data volume, ΔM prior is the prior knowledge gradient in the power field; Laplace noise is added through the differential privacy mechanism The noise parameter ∈ is dynamically adjusted by the formula R data is the data sensitivity index.

8. The intelligent decision method for preventing misoperation locking of a wind farm booster station based on multi-source data fusion according to claim 2, characterized in that, The adaptive anti-misoperation locking decision execution step constructs a decision explanation module, and the contribution degree φ of each feature to the decision is calculated through SHAP value i = E[f(z S∪S{i} )-f(z S )], wherein z S is a feature subset; Generating natural language explanation report Expl report = {feature i , φ i , impact type , threshold i}, provides decision basis for operation and maintenance personnel. 9.The intelligent decision-making method for preventing misoperation locking of a wind farm booster station based on multi-source data fusion according to claim 1, characterized in that, Also comprise: Self-learning and evolution step: design operation feedback mechanism, collect the correction information of operation and maintenance personnel on the decision result Feedback={op id ,correct label ,confidence score}; construct incremental learning model, update model parameters through formula θ new =θ old +η·Δθ feedback , wherein Δθ feedback is the feedback gradient; regularly evaluate the model, and when the model performance index decreases by more than a threshold value, trigger the model retraining process and update the model using the latest data. 10.The intelligent decision-making method for preventing misoperation locking of a wind farm booster station based on multi-source data fusion according to claim 1, characterized in that, Also comprise: Visualize and interact with the steps: build a three-dimensional visualization interface, map the device state, decision results, and failure warning information to the three-dimensional model; Design data visualization components, including real-time curves of electrical parameters, thermal maps of equipment states, and animations of fault propagation paths; implement human-computer interaction decision-making functions, and provide alternative operation schemes Alt when the system generates a lockout instruction ops = {op1, op2,..., op n}, and calculate the risk score of each alternative for the operation and maintenance personnel to refer to decision-making.

Images