Intelligent operation and maintenance management and control method and system based on digital twin wind power plant

Abstract

The invention discloses an intelligent operation and maintenance management and control method and system based on a digital twinning wind power plant, and relates to the technical field of digital twinning, and the method comprises the steps: obtaining a wind power plant fan time sequence signal, constructing a feature vector, carrying out the Hash mapping, and storing the feature vector into a segmented Hash table; judging an operation mode through a Bloom filter, generating a label triple stream and summarizing abnormal hash keys; counting the Hash value frequency in the time window and marking a high-risk value, and generating a Hash frequency distribution vector; calculating mutual information between nodes, and constructing an adjacent forest through a minimum generation forest; carrying out sampling and low-frequency signal inflection point storage on the sample flow; constructing an exception decision-making tree, and judging exceptional nodes; extracting weighted sub-graphs and dividing the weighted sub-graphs into fault mode clusters; and calculating a fault urgency degree score and generating an operation and maintenance work order. Through a multi-source time sequence signal, Hash mapping, a decision tree and a spectrum division algorithm, wind power plant fault detection, mode recognition and operation and maintenance scheduling are optimized, and the operation efficiency and sustainability are improved.

Description

Technical Field

[0001] The present invention relates to the field of digital twin technology, and specifically to an intelligent operation and maintenance control method and system based on a digital twin wind farm. Background Art

[0002] With the rapid development of renewable energy, wind power has become a vital energy source. As the infrastructure for converting wind energy into electricity, the operation and maintenance of wind farms directly impacts their operational efficiency and power generation costs. Traditional wind farm operation and maintenance relies heavily on manual inspections and scheduled maintenance, which is not only inefficient but also difficult to cope with complex and changing operating conditions.

[0003] In recent years, with the advancement of the Internet of Things (IoT), sensor technology, and big data analytics, the application of digital twin technology in wind farms has garnered widespread attention. Digital twin wind farms use sensors to collect multi-source time-series data from wind turbines and their environment, build virtual models based on this data, and update the virtual models with real-time data, enabling real-time monitoring and intelligent analysis of the wind farm. However, existing digital twin wind farm technologies remain a challenge in efficiently and accurately extracting meaningful information from massive amounts of sensor data for turbine status monitoring, fault prediction, and decision support.

[0004] While existing technologies have made progress in wind farm fault prediction and monitoring through machine learning and data mining, significant challenges remain in integrating diverse signal data with adaptive adjustment algorithms to achieve intelligent O&M management, improve wind farm operational efficiency, and reduce fault rates. In particular, challenges remain in effectively detecting anomalies, providing fault warnings, and optimizing dispatch, as well as dynamically adjusting wind farm O&M strategies and decision-making. Summary of the Invention

[0005] Based on the above-mentioned shortcomings of the prior art, the purpose of the present invention is to provide an intelligent operation and maintenance control method and system based on a digital twin wind farm to solve the above-mentioned technical problems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent operation and maintenance control method based on a digital twin wind farm, comprising: Obtain multi-source time series signals of wind turbines in the digital twin wind farm, construct feature vectors based on the time series signals, perform hash mapping, generate hash values, and store them in the corresponding hash buckets of the segmented hash table; The hash buckets are tested for their operating modes using a preset Bloom filter to generate labeled triple streams. Hash buckets labeled as known anomalies or newly suspected anomalies are aggregated into an anomaly hash key list. Based on the segmented hash table and triple stream, the frequency of hash value occurrences within a preset time window is counted. High-risk hash values are marked according to the abnormal hash key list, and a hash frequency distribution vector with risk weights is generated. Taking wind turbine IDs as nodes and hash frequency distribution vectors as node features, the mutual information correlation between nodes is calculated, and a dynamic adjacency forest is constructed using the pruned minimum spanning forest algorithm. For the sliding window sample stream, a reservoir sampling strategy is driven by importance in a fixed-capacity sample pool to perform representative resampling of high-frequency window samples, and a piecewise linear linked list is used to store inflection point information for low-frequency trend signals. Construct a hierarchical anomaly decision tree, perform multi-level anomaly judgment based on hash error rate, neighbor forest weight difference, and Reservoir sample kNN distance metric, and output a set of confirmed anomaly nodes; A weighted subgraph containing only confirmed abnormal nodes is extracted from the dynamic adjacency forest, and the wind turbine cluster is divided into multiple fault mode clusters using a recursive spectral partitioning algorithm. The fault urgency score is calculated based on the abnormal node deviation and the shortest path distance of the adjacent forest, and pushed into the double-ended priority queue. When the fault urgency score is greater than the preset threshold, an operation and maintenance work order is generated.

[0007] The present invention is further configured such that the multi-source timing signal includes: three-axis acceleration of the blade root, thermocouple output voltage, RMS value of three-phase current of the generator, and wind speed pulse count value; The sliding window is divided according to the preset time window and time step, and the numerical sequence of the multi-source time series signal in the sliding window is calculated, including the maximum value, minimum value, mean value and variance; The numerical sequences of multi-source time series signals are sequentially spliced into fixed-length feature vectors, and the wind turbine ID and feature vector are hashed using the MurmurHash3 algorithm to generate a hash value. The hash value is modulo the total number of buckets N in the hash table to determine the hash bucket number, where the segmented hash table is pre-divided into S segments, and each segment contains M=N / S hash buckets.

[0008] The present invention is further configured such that the Bloom filter includes a normal mode filter and a known anomaly filter, the normal mode filter is used to store verified normal hash keys, and the known anomaly filter is used to store historical known anomaly hash keys; The generation of the triplet stream includes: mapping the hash value in the hash bucket according to a preset hash function to obtain a hash key, determining whether it belongs to the hash key stored in the normal mode filter or the known anomaly filter, generating a normal label when it belongs to the hash key stored in the normal mode filter; generating a known anomaly label when it belongs to the hash key stored in the known anomaly filter; generating a new suspected anomaly label when it does not belong to the normal mode filter and does not belong to the hash key stored in the known anomaly filter; setting the wind turbine ID, feature vector and label as a triplet stream; The abnormal hash key list aggregation includes: aggregating the hash values of the hash buckets labeled as known abnormalities or new suspected abnormalities into the abnormal hash key list, and inserting the known abnormality filter into the hash value of the new suspected abnormality.

[0009] The present invention is further configured to traverse all records in the triple stream within a preset time window, count the hash keys in each record, and store the number of occurrences of each hash key in a hash table; According to the abnormal hash key list, check whether each hash key belongs to a known abnormality or a new suspected abnormality. If so, mark it as a high-risk hash value.

[0010] According to the high-risk hash value labeling and hash key count, the frequency of high-risk hash values is calculated and normalized to generate risk weights. The wind turbine ID, feature vector and risk weight are set as the hash frequency distribution vector.

[0011] The present invention is further configured to use wind turbine IDs as nodes, hash frequency distribution vectors as node features, and calculate the mutual information correlation between nodes based on the risk weights of the nodes; A weighted graph between wind turbines is constructed using the calculated mutual information correlation. The weight of each edge is the mutual information correlation. The weighted graph is processed using the pruned minimum spanning forest algorithm to remove edges below a preset threshold and retain wind turbine node pairs with high correlation. Generate multiple connected subgraphs representing different wind turbine groups to form a dynamic adjacency forest.

[0012] The present invention is further configured to process the sliding window sample stream, perform representative resampling of high-frequency window samples according to an importance-driven reservoir sampling strategy, and store the resampled high-frequency samples in a fixed-capacity sample pool; For low-frequency trend signals, a piecewise linear linked list is used to store data points, and the inflection point is determined according to the set change amplitude threshold. When the trend change exceeds the preset threshold, it is recorded as an inflection point and stored as a linked list node to form an inflection point information linked list of the low-frequency trend.

[0013] The present invention is further configured to construct a hierarchical abnormality decision tree, perform preliminary judgment on the wind turbine node by using the hash error rate, and mark it as a potential abnormal node when the hash error rate is greater than a preset threshold; and mark it as a normal node when the hash error rate is less than or equal to the preset threshold; For potential abnormal nodes, the weight difference of the nodes in the adjacent forest is calculated. When the weight difference exceeds the preset threshold, it is marked as a high-risk wind turbine. When the weight difference is less than or equal to the preset threshold, the wind turbine continues to be monitored. For high-risk wind turbines, the average distance to the nearest k neighbors is calculated based on the kNN distance metric in the Reservoir sample pool. When the average distance is greater than the preset threshold, the wind turbine is confirmed as an abnormal node and entered into the abnormal node set.

[0014] The present invention is further configured to extract a weighted subgraph containing only a set of confirmed abnormal nodes from the dynamic adjacency forest, wherein the weighted subgraph includes all abnormal nodes and connecting edges; By calculating the Laplace matrix of the weighted subgraph, the eigenvalues and eigenvectors of the graph are obtained. The subgraph is spectrally partitioned according to the Laplace matrix, and the eigenvectors corresponding to the first k smallest eigenvalues are calculated. Based on the calculated eigenvectors, the weighted subgraph is subdivided into multiple subgraphs using a recursive spectral partitioning algorithm. Each subgraph represents a group of wind turbines with similar failure modes. Output multiple fault mode clusters, each cluster including wind turbine nodes belonging to the cluster.

[0015] The present invention is further configured to obtain the abnormal node deviation by calculating the distance between the characteristic vector of the abnormal node and the characteristic vector of the normal wind turbine; Use the shortest path algorithm to calculate the shortest path from the abnormal node to the adjacent normal node, and obtain the shortest path distance of the adjacency forest; The fault urgency score of the abnormal node is calculated based on the abnormal node deviation and the shortest path distance of the adjacent forest. The calculation logic is: , For the The fault urgency score of each abnormal node, and For the and The abnormal node deviation of abnormal nodes, is the set of abnormal nodes, is the index value, For the The shortest path distance of the adjacent forest of abnormal nodes; Push the fault urgency score and abnormal node information into a double-ended priority queue, which is sorted by the fault urgency score. When the fault urgency score of a node is greater than the preset threshold, an operation and maintenance work order is generated, recording the wind turbine ID and urgency score, and reminding the user to perform maintenance or processing operations.

[0016] The present invention also provides an intelligent operation and maintenance control system based on a digital twin wind farm, which is used to implement the above-mentioned intelligent operation and maintenance control method based on a digital twin wind farm, including: Hash mapping module: obtains the multi-source timing signals of wind turbines in the digital twin wind farm, constructs feature vectors based on the timing signals, performs hash mapping, generates hash values, and stores them in the corresponding hash buckets of the segmented hash table; Label generation module: This module uses a preset Bloom filter to determine the operating mode of hash buckets, generates a labeled triple stream, and aggregates hash buckets labeled as known anomalies or newly suspected anomalies into an anomaly hash key list. Weight generation module: Based on the segmented hash table and triple stream, it counts the frequency of hash values within a preset time window, marks high-risk hash values according to the abnormal hash key list, and generates a hash frequency distribution vector with risk weights; Forest construction module: Using wind turbine IDs as nodes and hash frequency distribution vectors as node features, it calculates the mutual information correlation between nodes and constructs a dynamic adjacency forest using the pruned minimum spanning forest algorithm. Sampling module: For the sliding window sample stream, a reservoir sampling strategy is driven by importance in a fixed-capacity sample pool to perform representative resampling of high-frequency window samples, and a piecewise linear linked list is used to store inflection point information for low-frequency trend signals; Anomaly confirmation module: Builds a hierarchical anomaly decision tree, performs multi-level anomaly judgment based on hash error rate, neighbor forest weight difference, and reservoir sample kNN distance metric, and outputs a set of confirmed anomaly nodes; Fault clustering module: extracts a weighted subgraph containing only confirmed abnormal nodes from the dynamic adjacency forest, and divides the wind turbine group into multiple fault mode clusters using a recursive spectral partitioning algorithm; Operation and maintenance generation module: Calculates the fault urgency score based on the abnormal node deviation and the shortest path distance in the adjacent forest, pushes it into the double-ended priority queue, and generates an operation and maintenance work order when the fault urgency score exceeds the preset threshold.

[0017] The present invention provides an intelligent operation and maintenance control method and system based on a digital twin wind farm. The method obtains multi-source time series signals of wind turbines in the digital twin wind farm, constructs a feature vector according to the time series signals, performs hash mapping, generates a hash value, and stores it in the corresponding hash bucket of the segmented hash table; the operation mode of the hash bucket is judged by a preset Bloom filter, a triple stream with a label is generated, and the hash buckets labeled as known anomalies or new suspected anomalies are summarized into an abnormal hash key list; based on the segmented hash table and the triple stream, the frequency of occurrence of hash values is counted within a preset time window, high-risk hash values are marked according to the abnormal hash key list, and a hash frequency distribution vector with a risk weight is generated; the wind turbine ID is used as a node and the hash frequency distribution vector is used as a node feature, the mutual information correlation between nodes is calculated, and the pruned minimum spanning forest algorithm is used to calculate the correlation between the mutual information between nodes. A dynamic adjacency forest is constructed. A representative resampling of high-frequency window samples is performed in a fixed-capacity sample pool using an importance-driven reservoir sampling strategy. A piecewise linear linked list is used to store inflection point information for low-frequency trend signals. A hierarchical anomaly decision tree is constructed, which performs multi-level anomaly determination based on hash error rate, adjacency forest weight difference, and reservoir sample kNN distance metric, outputting a set of confirmed anomaly nodes. A weighted subgraph containing only the set of confirmed anomaly nodes is extracted from the dynamic adjacency forest, and a recursive spectral partitioning algorithm is used to divide the wind turbine fleet into multiple fault mode clusters. A fault urgency score is calculated based on the abnormal node deviation and the adjacency forest shortest path distance, and the score is pushed into a two-ended priority queue. When the fault urgency score exceeds a preset threshold, an operation and maintenance work order is generated. The beneficial effects include: 1. Improve the accuracy of fault detection and early warning: By combining multi-source time series signals from wind turbines and using hash mapping to generate feature vectors, it is possible to accurately model the operating status of wind turbines in large-scale wind farms. The two-stage judgment mechanism of the Bloom filter can effectively identify normal modes and known abnormal modes, and promptly detect new suspected abnormalities, significantly improving the early warning capabilities of wind turbine failures and reducing the probability of failures. 2. Optimize wind farm operation and maintenance decisions: By constructing a hierarchical anomaly decision tree and combining multiple judgment methods such as hash error rate, neighbor forest weight difference, and reservoir sample kNN distance measurement, multi-level anomaly judgment is performed to accurately determine the operating status of wind turbines, optimize the operation and maintenance decision-making process, and improve the operating efficiency of wind farms.

[0018] 3. Improve wind farm fault pattern recognition and clustering capabilities: By dividing the wind turbine group into different fault pattern clusters through the recursive spectral partitioning algorithm, wind turbine nodes with similar fault patterns in the wind turbine group can be better identified, thereby achieving accurate fault pattern recognition.

[0019] 4. Dynamic O&M Work Order Generation and Priority Scheduling: By calculating fault urgency scores and utilizing a two-ended priority queue, dynamic generation and scheduling of wind farm O&M work orders are achieved. When a wind turbine fault urgency score exceeds a preset threshold, an O&M work order is immediately generated, ensuring that high-risk faulty turbines are prioritized.

[0020] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without inventive efforts. In the drawings: Figure 1 This is a flow chart showing an intelligent operation and maintenance control method based on a digital twin wind farm according to an exemplary embodiment of the present invention; Figure 2 The figure is a structural diagram of an intelligent operation and maintenance control system based on a digital twin wind farm, showing an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following describes the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art will readily appreciate the other advantages and benefits of the present invention from the disclosure herein. The present invention may also be implemented or applied through various other specific embodiments, and the various details in this specification may be modified or altered based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are intended only to illustrate the present invention and are not intended to limit the scope of protection of the present invention.

[0023] It should be noted that the illustrations provided in the following embodiments are merely schematic illustrations of the basic concept of the present invention. Therefore, the illustrations only show components related to the present invention and are not drawn according to the number, shape, and size of components in actual implementation. In actual implementation, the type, quantity, and proportion of each component may be changed arbitrarily, and the component layout may also be more complex.

[0024] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the embodiments of the present invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring the embodiments of the present invention.

[0025] Example 1: An intelligent operation and maintenance control method based on digital twin wind farm, such as Figure 1 Shown, including: Obtain multi-source time series signals of wind turbines in the digital twin wind farm, construct feature vectors based on the time series signals, perform hash mapping, generate hash values, and store them in the corresponding hash buckets of the segmented hash table; The hash buckets are tested for their operating modes using a preset Bloom filter to generate labeled triple streams. Hash buckets labeled as known anomalies or newly suspected anomalies are aggregated into an anomaly hash key list. Based on the segmented hash table and triple stream, the frequency of hash value occurrences within a preset time window is counted. High-risk hash values are marked according to the abnormal hash key list, and a hash frequency distribution vector with risk weights is generated. Taking wind turbine IDs as nodes and hash frequency distribution vectors as node features, the mutual information correlation between nodes is calculated, and a dynamic adjacency forest is constructed using the pruned minimum spanning forest algorithm. For the sliding window sample stream, a reservoir sampling strategy is driven by importance in a fixed-capacity sample pool to perform representative resampling of high-frequency window samples, and a piecewise linear linked list is used to store inflection point information for low-frequency trend signals. Construct a hierarchical anomaly decision tree, perform multi-level anomaly judgment based on hash error rate, neighbor forest weight difference, and Reservoir sample kNN distance metric, and output a set of confirmed anomaly nodes; A weighted subgraph containing only confirmed abnormal nodes is extracted from the dynamic adjacency forest, and the wind turbine cluster is divided into multiple fault mode clusters using a recursive spectral partitioning algorithm. The fault urgency score is calculated based on the abnormal node deviation and the shortest path distance of the adjacent forest, and pushed into the double-ended priority queue. When the fault urgency score is greater than the preset threshold, an operation and maintenance work order is generated.

[0026] The present invention is further configured such that the multi-source timing signal includes: three-axis acceleration of the blade root, thermocouple output voltage, three-phase current RMS value of the generator, and wind speed pulse count value; specifically, in the multi-source timing signal, the three-axis acceleration of the blade root is used to detect the vibration of the fan blade root, and is measured by an acceleration sensor, and the data includes acceleration values in the three directions of X, Y, and Z; the output voltage of the thermocouple is temperature measurement data, and the thermocouple sensor converts the temperature change into a voltage signal and outputs the temperature information inside the fan; the three-phase current RMS value of the generator is used to reflect the current load of the fan generator. By calculating the root mean square value (RMS) of the three-phase current, the current change trend is obtained to reflect the operating status of the generator; the wind speed pulse count value measures the wind speed change through the wind speed sensor, and uses the pulse signal to convert it into wind speed data for monitoring the wind environment in which the fan is operating; The sliding window is divided according to the preset time window and time step, and the numerical sequence of the multi-source time series signals in the sliding window is calculated, including the maximum value, minimum value, mean value and variance. Specifically, the time window specifies a fixed time length (such as 10 seconds), and each window includes the sampling data of multiple time series signals. The time step is set to a subset of the window length (5 seconds), then: the time window is 10 seconds, the step is 5 seconds, and the data window is updated every 5 seconds. The numerical value in each window is calculated, including: calculating the maximum value of each time series signal in the window, representing the peak value of the signal; calculating the minimum value of each time series signal in the window, representing the valley value of the signal; calculating the average value of each time series signal in the window, representing the overall level of the signal; calculating the variance of each time series signal, representing the fluctuation of the signal; The numerical sequences of multi-source time series signals are sequentially spliced into fixed-length feature vectors. The wind turbine ID and feature vector are hashed using the MurmurHash3 algorithm to generate a hash value. Specifically, the maximum, minimum, mean, and variance calculated in each time window are sequentially spliced into a fixed-length feature vector. There are four time series signals (three-axis acceleration at the blade root, thermocouple output voltage, generator three-phase current RMS value, and wind speed pulse count value). The characteristics of each signal include maximum, minimum, mean, and variance. Therefore, the feature vector length of each time window is: feature vector length = 4 × 4 = 16. The wind turbine ID and feature vector are spliced into a string as the input of the hash function. The spliced data is hashed using the MurmurHash3 algorithm to obtain a fixed-length hash value. The generated hash value is used to store in the segmented hash table. The hash value modulo the total number of hash table buckets N is used to determine the hash bucket number, where the segmented hash table is pre-divided into S segments, each segment contains M=N / S hash buckets; specifically, the hash value modulo the total number of hash table buckets N is used to determine the hash bucket number: h bucket =h hashmod N,h hash is the hash value.

[0027] The present invention is further configured such that the Bloom filter includes a normal mode filter and a known anomaly filter, the normal mode filter is used to store verified normal hash keys, and the known anomaly filter is used to store historical known anomaly hash keys; The generation of a triple stream involves mapping hash values in a hash bucket according to a preset hash function to obtain a hash key. The hash key is then determined to determine whether it belongs to a hash key stored in a normal mode filter or a known anomaly filter. If the hash key belongs to a hash key stored in a normal mode filter, a normal label is generated. If the hash key belongs to a hash key stored in a known anomaly filter, a known anomaly label is generated. If the hash key does not belong to a normal mode filter and does not belong to a hash key stored in a known anomaly filter, a new suspected anomaly label is generated. The wind turbine ID, feature vector, and label are set as a triple stream. Specifically, the normal mode filter stores verified normal hash keys. This filter is validated by historical data and stores hash values of all verified normal wind turbine operating states. The known anomaly filter stores historical known anomaly hash keys. This filter records all hash values marked as known faults or anomalies and is used to determine in real time whether a newly input hash key belongs to a known anomaly mode. Each filter uses a standard Bloom filter structure, mapping hash values using multiple hash functions and setting the corresponding positions to 1. The hash values in each hash bucket are mapped according to a preset hash function (MurmurHash3). For each wind turbine's feature vector and wind turbine ID, a unique hash value is generated. For each generated hash value, it is determined whether it belongs to the hash key in the normal mode filter or the known anomaly filter. If the hash value is in the normal mode filter, a "normal" label is generated; if the hash value is in the known anomaly filter, a "known anomaly" label is generated; if the hash value is neither in the normal mode filter nor in the known anomaly filter, a "new suspected anomaly" label is generated. Based on the wind turbine ID, feature vector, and label, a triple stream is generated. The anomaly hash key list aggregation process involves aggregating the hash values of hash buckets labeled "known anomalies" or "new suspected anomalies" into the anomaly hash key list and inserting the hash values of new suspected anomalies into the known anomaly filter. Specifically, hash values labeled "known anomaly" or "new suspected anomaly" are aggregated into the anomaly hash key list. This ensures that subsequent processing focuses only on abnormal wind turbines, optimizing computational and storage efficiency. Hash values labeled "new suspected anomaly" are further inserted into the known anomaly filter. This way, over time, new anomaly hash keys are continuously accumulated into the known anomaly list, enhancing the accuracy of the filter.

[0028] The present invention is further configured to traverse all records in the triple stream within a preset time window, count the hash keys in each record, and store the number of occurrences of each hash key in a hash table. Specifically, each record in the triple stream is traversed, and the hash keys in each record are counted. The hash keys are generated by a hash function using a characteristic vector of the wind turbine to identify the operating status of the wind turbine. The number of occurrences of each hash key is stored in the hash table to form a hash frequency statistic for each wind turbine within the given time window. According to the abnormal hash key list, check whether each hash key belongs to a known abnormality or a new suspected abnormality. If so, mark it as a high-risk hash value.

[0029] Based on the high-risk hash value labeling and hash key count, the frequency of high-risk hash values is calculated, normalized, and risk weights are generated. The wind turbine ID, feature vector, and risk weight are set as the hash frequency distribution vector. Specifically, the frequency of each high-risk hash key is calculated. For each high-risk hash key, its occurrence count is obtained from the hash table, divided by the total number of times to obtain the frequency of the high-risk hash value. The frequency of the high-risk hash value is normalized so that its value is within a reasonable range. The normalization method can be selected as maximum value normalization or Z-score normalization, which is not restricted here. The normalized frequency is set as the risk weight, and each high-risk hash value and its normalized frequency (risk weight) are concatenated with the corresponding wind turbine ID and feature vector to generate the final hash frequency distribution vector. The hash frequency distribution vector can be used as the state feature of the wind turbine within a given time window for subsequent fault diagnosis and operation and maintenance decisions.

[0030] The present invention further configures the wind turbine ID as a node and the hash frequency distribution vector as a node feature to calculate the mutual information correlation between nodes based on the node's risk weight; and calculates the mutual information correlation between wind turbines based on the node's hash frequency distribution vector and risk weight. The mutual information metric reflects the similarity between two nodes (wind turbines) in the feature space: ,in, denote the frequencies of the kth hash key of fan i and fan j respectively, is the joint probability, , is the marginal probability; by combining risk weight with mutual information correlation, the model can focus on high-risk nodes. A higher risk weight can increase the influence of the corresponding node on the adjacency relationship; The weighted graph between wind turbines is constructed by calculating the mutual information correlation. The weight of the edge is the mutual information correlation. The weighted graph is processed by the pruned minimum spanning forest algorithm to remove the edges below the preset threshold and retain the wind turbine node pairs with high correlation. Specifically, the minimum spanning tree (MST) algorithm (such as Kruskal algorithm) is used to construct a preliminary spanning tree, retaining the edge with the smallest weight, and then selecting the best node based on the preset threshold T. cut , remove edges with weights lower than the threshold. Only those node pairs with high similarity are retained to form a new wind turbine group; Generate multiple connected subgraphs representing different wind turbine groups to form a dynamic adjacency forest. Specifically, using a graph connectivity algorithm (such as depth-first search or breadth-first search), the graph is divided into multiple connected subgraphs. Each subgraph represents a wind turbine group, and the wind turbines within a group have strong connections. The multiple connected subgraphs are then combined to form a dynamic adjacency forest, where each tree represents a wind turbine group, the nodes in the tree represent wind turbines, and the edges represent the connections between wind turbines.

[0031] The present invention further processes a sliding window sample stream, representatively resampling high-frequency window samples using an importance-driven reservoir sampling strategy, and storing the resampled high-frequency samples in a fixed-capacity sample pool. Specifically, the wind turbine's time series data is segmented according to a preset time window and step size. The data within each window represents the wind turbine's status within that time period. The importance of each sample is assessed based on its deviation (i.e., the degree to which the data deviates from the normal data interval, quantified as (data - normal data area) / normal data area), and a weight is assigned to each sample based on the importance metric. Samples with higher importance have a greater probability of being sampled during the resampling process. Reservoir sampling is used to retain representative samples in the fixed-capacity sample pool. When each new sample arrives, it is added to the pool with a certain probability. If the pool is full, the sample's weight determines whether to replace it. If the sample pool is not full, the sample is added directly. If the sample pool is full, the sample's replacement probability is calculated based on its weight (data deviation), and the replacement probability is determined based on this probability.

[0032] For low-frequency trend signals, a piecewise linear linked list is used to store data points. Inflection points are determined based on a set threshold for change amplitude. When the trend change exceeds the preset threshold, it is recorded as an inflection point and stored as a linked list node, forming a linked list of inflection point information for the low-frequency trend. Specifically, the low-frequency trend signal is stored segment by segment, with each segment storing trend data for a time period. The change amplitude of the data for each time period (the difference from the previous segment) is calculated. If the change amplitude exceeds the preset threshold, it is considered an inflection point. An inflection point represents a significant change in the trend and can serve as an early warning signal for wind turbine anomalies. When a trend change exceeding the threshold is detected, the current inflection point is recorded and stored in a linked list node. Each node in the linked list stores information about an inflection point, including a timestamp, signal value, and change amplitude. All inflection points are stored in chronological order in the linked list, forming a linked list of inflection point information for the low-frequency trend signal. This linked list provides important trend changes during wind turbine operation, which can be used for subsequent status analysis and fault diagnosis.

[0033] The present invention is further configured to construct a hierarchical abnormality decision tree and make a preliminary judgment on the wind turbine node by using the hash error rate. When the hash error rate is greater than a preset threshold, it is marked as a potential abnormal node; when the hash error rate is less than or equal to the preset threshold, it is marked as a normal node. Specifically, the hash error rate is a measure of the difference between the wind turbine feature vector and the historical normal wind turbine feature. By comparing the hash value of the wind turbine's current state with the hash value of the historical normal wind turbine, the error rate of the wind turbine state can be obtained; e hash =|h current −h normal ∣ / h normal , where h current is the hash value of the current wind turbine, h normal is the hash value of a normal fan; if the hash error rate e hash Greater than the preset threshold T hash , then mark the wind turbine as a potential abnormal node, otherwise mark it as a normal node; For potential abnormal nodes, the weight difference of the nodes in the adjacent forest is calculated. When the weight difference exceeds the preset threshold, it is marked as a high-risk wind turbine; when the weight difference is less than or equal to the preset threshold, the wind turbine continues to be monitored. Specifically, the weight difference in the adjacent forest is used to measure the correlation between wind turbine nodes. The weight difference Δw refers to the difference in connection strength between a wind turbine node and its neighbors, Δw=|w node −w neighbor ∣, where w node is the weight of the current node, w neighbor is the weight of the adjacent node; if Δw is greater than the preset threshold T weight , then mark the node as a high-risk wind turbine, otherwise, continue to monitor the node; For high-risk wind turbines, the average distance to the k nearest neighbors is calculated based on the kNN distance metric in the Reservoir sample pool. When the average distance is greater than the preset threshold, the wind turbine is confirmed as an abnormal node and input into the abnormal node set. Specifically, the k-nearest neighbor algorithm (kNN) is used to calculate the distance between the current wind turbine and the k nearest neighbors in the sample pool. The degree of abnormality is measured by comparing the similarity between the current wind turbine and its neighbors. The Euclidean distance between the wind turbine and its nearest neighbors is calculated as the similarity metric for the wind turbine. The average distance to the k nearest neighbors is calculated. If the average distance is greater than the preset threshold T kNN , the wind turbine is confirmed as an abnormal node. Otherwise, the node continues to be monitored. The wind turbine marked as an abnormal node is added to the abnormal node set. This set is used to store all wind turbine nodes that are determined to be abnormal.

[0034] The present invention is further configured to extract a weighted subgraph containing only the set of confirmed abnormal nodes from the dynamic adjacency forest, where the weighted subgraph includes all abnormal nodes and connecting edges. Specifically, the subgraph containing only the set of confirmed abnormal nodes is extracted from the dynamic adjacency forest. Only abnormal nodes and the edges between them are retained in the subgraph, and the edge weights represent the strength of the association between the abnormal nodes, which is the mutual information correlation. By calculating the Laplace matrix of the weighted subgraph, the eigenvalues and eigenvectors of the graph are obtained. The subgraph is spectrally partitioned according to the Laplace matrix, and the eigenvectors corresponding to the first k smallest eigenvalues are calculated. Specifically, the adjacency matrix A of the weighted subgraph is constructed, where A ij represents the weight of the edge between node i and node j. If there is no edge between node i and node j, then A ij =0; the degree matrix D is a diagonal matrix, where each diagonal element D ii represents the degree of node i, that is, the number of neighbors of node i. The Laplacian matrix L is calculated from the degree matrix D and the adjacency matrix A: L = D − A. Perform eigendecomposition on the Laplacian matrix to calculate the first k smallest eigenvalues and their corresponding eigenvectors. These eigenvectors reflect the similarity between nodes. Based on the size of the eigenvalues, the eigenvectors corresponding to the first k smallest eigenvalues are selected. Each element in the eigenvector represents the position of a node on that feature dimension. Based on the calculated eigenvectors, the weighted subgraph is subdivided into multiple subgraphs using a recursive spectral partitioning algorithm. Each subgraph represents a group of wind turbines with similar failure modes. Specifically, the weighted subgraph is further subdivided using the recursive spectral partitioning algorithm by calculating the eigenvectors, dividing each subgraph into multiple smaller subgraphs. Each subgraph represents a group of wind turbines with similar failure modes. Nodes are partitioned based on the eigenvector values. Low eigenvalues correspond to nodes with high similarity, and these nodes are usually grouped into the same cluster, ultimately resulting in multiple clusters with similar failure modes. Nodes with known failure modes can be clustered for rapid identification of specific failure modes. Output multiple fault mode clusters, each cluster including wind turbine nodes belonging to the cluster.

[0035] The present invention is further configured to obtain the abnormal node deviation by calculating the distance between the characteristic vector of the abnormal node and the characteristic vector of the normal wind turbine; specifically, the distance between the characteristic vector of the abnormal node and the characteristic vector of the normal wind turbine is calculated using the Euclidean distance to obtain the abnormal node deviation; The shortest path algorithm is used to calculate the shortest path from the abnormal node to the adjacent normal node to obtain the adjacency forest shortest path distance. Specifically, the shortest path algorithm (such as the Dijkstra algorithm) is used to calculate the shortest path distance from the abnormal node to the adjacent normal node. The shortest path represents the structural distance between the abnormal node and the normal node. The fault urgency score of the abnormal node is calculated based on the abnormal node deviation and the shortest path distance of the adjacent forest. The calculation logic is: , For the The fault urgency score of each abnormal node, and For the and The abnormal node deviation of abnormal nodes, is the set of abnormal nodes, is the index value, For the The shortest path distance of the adjacent forest of abnormal nodes; specifically, through Normalize the deviation of the nodes. This normalization ensures that the fault urgency scores of different wind turbines are comparable, and prevents some nodes with large deviations but a small proportion in the abnormal node set from excessively influencing the results. The effect of distance is smoothed. The logarithmic function can suppress the excessive contribution of abnormal nodes with large distances to the score, avoid excessively high scores due to long paths, and ensure the stability of the calculation results. By combining the deviation of abnormal nodes and the shortest path distance, a comprehensive fault urgency score is obtained, which takes into account both the severity of the individual node fault and the node's propagation potential in the system. Push the fault urgency score and abnormal node information into a double-ended priority queue, which is sorted by the fault urgency score. When the fault urgency score of a node is greater than the preset threshold, an operation and maintenance work order is generated, the wind turbine ID and urgency score are recorded, and a reminder is given to perform maintenance or processing operations. Specifically, by calculating the hash error rate and the shortest path distance of the adjacent forest, the difference between the wind turbine and the normal wind turbine, as well as the correlation between the wind turbine and its neighboring wind turbines, can be accurately evaluated. This multi-dimensional measurement method improves the detection accuracy of abnormal nodes and can effectively identify potential faulty wind turbines, especially those with significant deviations from normal behavior; the fault urgency score is calculated based on the deviation of the abnormal node and the shortest path distance of the adjacent forest, and the priority of fault handling can be dynamically adjusted according to actual conditions. Wind turbines with high urgency scores will be handled first to ensure that critical faults are handled in the shortest time, thereby reducing fault expansion and downtime.

[0036] Example 2: See also Figure 2 This exemplary intelligent operation and maintenance control system based on a digital twin wind farm is used to implement the above-mentioned intelligent operation and maintenance control method based on a digital twin wind farm, including: Hash mapping module: obtains the multi-source timing signals of wind turbines in the digital twin wind farm, constructs feature vectors based on the timing signals, performs hash mapping, generates hash values, and stores them in the corresponding hash buckets of the segmented hash table; Label generation module: This module uses a preset Bloom filter to determine the operating mode of hash buckets, generates a labeled triple stream, and aggregates hash buckets labeled as known anomalies or newly suspected anomalies into an anomaly hash key list. Weight generation module: Based on the segmented hash table and triple stream, it counts the frequency of hash values within a preset time window, marks high-risk hash values according to the abnormal hash key list, and generates a hash frequency distribution vector with risk weights; Forest construction module: Using wind turbine IDs as nodes and hash frequency distribution vectors as node features, it calculates the mutual information correlation between nodes and constructs a dynamic adjacency forest using the pruned minimum spanning forest algorithm. Sampling module: For the sliding window sample stream, a reservoir sampling strategy is driven by importance in a fixed-capacity sample pool to perform representative resampling of high-frequency window samples, and a piecewise linear linked list is used to store inflection point information for low-frequency trend signals; Anomaly confirmation module: Builds a hierarchical anomaly decision tree, performs multi-level anomaly judgment based on hash error rate, neighbor forest weight difference, and reservoir sample kNN distance metric, and outputs a set of confirmed anomaly nodes; Fault clustering module: extracts a weighted subgraph containing only confirmed abnormal nodes from the dynamic adjacency forest, and divides the wind turbine group into multiple fault mode clusters using a recursive spectral partitioning algorithm; Operation and maintenance generation module: Calculates the fault urgency score based on the abnormal node deviation and the shortest path distance in the adjacent forest, pushes it into the double-ended priority queue, and generates an operation and maintenance work order when the fault urgency score exceeds the preset threshold.

[0037] It should be noted that the intelligent operation and maintenance control system based on a digital twin wind farm provided in the above embodiment and the intelligent operation and maintenance control method based on a digital twin wind farm provided in the above embodiment belong to the same concept, wherein the specific manner in which each module and unit performs operations has been described in detail in the method embodiment and will not be repeated here. In actual applications, the intelligent operation and maintenance control system based on a digital twin wind farm provided in the above embodiment can allocate the above functions to different functional modules as needed, that is, divide the internal structure of the system into different functional modules to complete all or part of the functions described above, and this is not limited here.

[0038] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0039] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0040] In this application, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0041] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0042] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0043] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0044] In the several embodiments provided in this application, it should be understood that the disclosed system can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0045] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0046] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0047] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0048] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. An intelligent operation and maintenance control method based on digital twin wind farm, characterized in that: include: Obtain multi-source time series signals of wind turbines in the digital twin wind farm, construct feature vectors based on the time series signals, perform hash mapping, generate hash values, and store them in the corresponding hash buckets of the segmented hash table; The hash buckets are tested for their operating modes using a preset Bloom filter to generate labeled triple streams. Hash buckets labeled as known anomalies or newly suspected anomalies are aggregated into an anomaly hash key list. Based on the segmented hash table and triple stream, the frequency of hash value occurrences within a preset time window is counted. High-risk hash values are marked according to the abnormal hash key list, and a hash frequency distribution vector with risk weights is generated. Taking wind turbine IDs as nodes and hash frequency distribution vectors as node features, the mutual information correlation between nodes is calculated, and a dynamic adjacency forest is constructed using the pruned minimum spanning forest algorithm. For the sliding window sample stream, a reservoir sampling strategy is driven by importance in a fixed-capacity sample pool to perform representative resampling of high-frequency window samples, and a piecewise linear linked list is used to store inflection point information for low-frequency trend signals. Construct a hierarchical anomaly decision tree, perform multi-level anomaly judgment based on hash error rate, neighbor forest weight difference, and Reservoir sample kNN distance metric, and output a set of confirmed anomaly nodes; A weighted subgraph containing only confirmed abnormal nodes is extracted from the dynamic adjacency forest, and the wind turbine cluster is divided into multiple fault mode clusters using a recursive spectral partitioning algorithm. The fault urgency score is calculated based on the abnormal node deviation and the shortest path distance of the adjacent forest, and pushed into the double-ended priority queue. When the fault urgency score is greater than the preset threshold, an operation and maintenance work order is generated.

2. The intelligent operation and maintenance control method based on digital twin wind farm according to claim 1 is characterized in that: The multi-source timing signals include: blade root triaxial acceleration, thermocouple output voltage, generator three-phase current RMS value and wind speed pulse count value; The sliding window is divided according to the preset time window and time step, and the numerical sequence of the multi-source time series signal in the sliding window is calculated, including the maximum value, minimum value, mean value and variance; The numerical sequences of multi-source time series signals are sequentially spliced into fixed-length feature vectors, and the wind turbine ID and feature vector are hashed using the MurmurHash3 algorithm to generate a hash value. The hash value is modulo the total number of buckets N in the hash table to determine the hash bucket number, where the segmented hash table is pre-divided into S segments, and each segment contains M=N / S hash buckets.

3. The intelligent operation and maintenance control method based on digital twin wind farm according to claim 1 is characterized in that: Bloom filters include normal mode filters and known anomaly filters. Normal mode filters are used to store verified normal hash keys, and known anomaly filters are used to store historical known anomaly hash keys. The generation of the triplet stream includes: mapping the hash value in the hash bucket according to a preset hash function to obtain a hash key, determining whether it belongs to the hash key stored in the normal mode filter or the known anomaly filter, generating a normal label when it belongs to the hash key stored in the normal mode filter; generating a known anomaly label when it belongs to the hash key stored in the known anomaly filter; generating a new suspected anomaly label when it does not belong to the normal mode filter and does not belong to the hash key stored in the known anomaly filter; setting the wind turbine ID, feature vector and label as a triplet stream; The abnormal hash key list aggregation includes: aggregating the hash values of the hash buckets labeled as known abnormalities or new suspected abnormalities into the abnormal hash key list, and inserting the known abnormality filter into the hash value of the new suspected abnormality.

4. The intelligent operation and maintenance control method based on digital twin wind farm according to claim 1 is characterized in that: In the preset time window, traverse all records in the triple stream, count the hash keys in each record, and store the number of occurrences of each hash key in the hash table; According to the abnormal hash key list, check whether each hash key belongs to a known abnormality or a new suspected abnormality. If so, mark it as a high-risk hash value. According to the high-risk hash value labeling and hash key count, the frequency of high-risk hash values is calculated and normalized to generate risk weights. The wind turbine ID, feature vector and risk weight are set as the hash frequency distribution vector.

5. The intelligent operation and maintenance control method based on digital twin wind farm according to claim 4 is characterized in that: Taking the wind turbine ID as the node and the hash frequency distribution vector as the node feature, the mutual information correlation between nodes is calculated according to the risk weight of the node; A weighted graph between wind turbines is constructed using the calculated mutual information correlation. The weight of each edge is the mutual information correlation. The weighted graph is processed using the pruned minimum spanning forest algorithm to remove edges below a preset threshold and retain wind turbine node pairs with high correlation. Generate multiple connected subgraphs representing different wind turbine groups to form a dynamic adjacency forest.

6. The intelligent operation and maintenance control method based on digital twin wind farm according to claim 1, characterized in that: The sliding window sample stream is processed, and the high-frequency window samples are representatively resampled according to the importance-driven reservoir sampling strategy, and the resampled high-frequency samples are stored in a fixed-capacity sample pool; For low-frequency trend signals, a piecewise linear linked list is used to store data points, and the inflection point is determined according to the set change amplitude threshold. When the trend change exceeds the preset threshold, it is recorded as an inflection point and stored as a linked list node to form an inflection point information linked list of the low-frequency trend.

7. The intelligent operation and maintenance control method based on digital twin wind farm according to claim 1 is characterized in that: A hierarchical anomaly decision tree is constructed to make preliminary judgments on wind turbine nodes based on the hash error rate. When the hash error rate is greater than the preset threshold, it is marked as a potential anomaly node. When the hash error rate is less than or equal to the preset threshold, it is marked as a normal node; For potential abnormal nodes, the weight difference of the nodes in the adjacent forest is calculated. When the weight difference exceeds the preset threshold, it is marked as a high-risk wind turbine; When the weight difference is less than or equal to the preset threshold, continue to monitor the wind turbine; For high-risk wind turbines, the average distance to the nearest k neighbors is calculated based on the kNN distance metric in the Reservoir sample pool. When the average distance is greater than the preset threshold, the wind turbine is confirmed as an abnormal node and entered into the abnormal node set.

8. The intelligent operation and maintenance control method based on digital twin wind farm according to claim 1 is characterized in that: Extract a weighted subgraph containing only the set of confirmed abnormal nodes from the dynamic adjacency forest. The weighted subgraph includes all abnormal nodes and connecting edges. By calculating the Laplace matrix of the weighted subgraph, the eigenvalues and eigenvectors of the graph are obtained. The subgraph is spectrally partitioned according to the Laplace matrix, and the eigenvectors corresponding to the first k smallest eigenvalues are calculated. Based on the calculated eigenvectors, the weighted subgraph is subdivided into multiple subgraphs using a recursive spectral partitioning algorithm. Each subgraph represents a group of wind turbines with similar failure modes. Output multiple fault mode clusters, each cluster including wind turbine nodes belonging to the cluster.

9. The intelligent operation and maintenance control method based on digital twin wind farm according to claim 1, characterized in that: By calculating the distance between the characteristic vector of the abnormal node and the characteristic vector of the normal wind turbine, the deviation degree of the abnormal node is obtained; Use the shortest path algorithm to calculate the shortest path from the abnormal node to the adjacent normal node, and obtain the shortest path distance of the adjacency forest; The fault urgency score of the abnormal node is calculated based on the abnormal node deviation and the shortest path distance of the adjacent forest. The calculation logic is: , For the The fault urgency score of each abnormal node, and For the and The abnormal node deviation of abnormal nodes, is the set of abnormal nodes, is the index value, For the The shortest path distance of the adjacent forest of abnormal nodes; Push the fault urgency score and abnormal node information into a double-ended priority queue, which is sorted by the fault urgency score. When the fault urgency score of a node is greater than the preset threshold, an operation and maintenance work order is generated, recording the wind turbine ID and urgency score, and reminding the user to perform maintenance or processing operations.

10. An intelligent operation and maintenance control system based on a digital twin wind farm, used to implement the intelligent operation and maintenance control method based on a digital twin wind farm according to any one of claims 1 to 9, characterized in that: include: Hash mapping module: obtains the multi-source timing signals of wind turbines in the digital twin wind farm, constructs feature vectors based on the timing signals, performs hash mapping, generates hash values, and stores them in the corresponding hash buckets of the segmented hash table; Label generation module: This module uses a preset Bloom filter to determine the operating mode of hash buckets, generates a labeled triple stream, and aggregates hash buckets labeled as known anomalies or newly suspected anomalies into an anomaly hash key list. Weight generation module: Based on the segmented hash table and triple stream, it counts the frequency of hash values within a preset time window, marks high-risk hash values according to the abnormal hash key list, and generates a hash frequency distribution vector with risk weights; Forest construction module: Using wind turbine IDs as nodes and hash frequency distribution vectors as node features, it calculates the mutual information correlation between nodes and constructs a dynamic adjacency forest using the pruned minimum spanning forest algorithm. Sampling module: For the sliding window sample stream, a reservoir sampling strategy is driven by importance in a fixed-capacity sample pool to perform representative resampling of high-frequency window samples, and a piecewise linear linked list is used to store inflection point information for low-frequency trend signals; Anomaly confirmation module: Builds a hierarchical anomaly decision tree, performs multi-level anomaly judgment based on hash error rate, neighbor forest weight difference, and reservoir sample kNN distance metric, and outputs a set of confirmed anomaly nodes; Fault clustering module: extracts a weighted subgraph containing only confirmed abnormal nodes from the dynamic adjacency forest, and divides the wind turbine group into multiple fault mode clusters using a recursive spectral partitioning algorithm; Operation and maintenance generation module: Calculates the fault urgency score based on the abnormal node deviation and the shortest path distance in the adjacent forest, pushes it into the double-ended priority queue, and generates an operation and maintenance work order when the fault urgency score exceeds the preset threshold.

Images