A wind turbine safety operation monitoring method based on digital twinning

Abstract

This invention discloses a method for monitoring the safe operation of wind turbine units based on digital twins, belonging to the field of safety monitoring technology. The method includes: performing data quality assessment and credibility labeling on wind turbine unit status data, outputting a unit status vector; inputting the unit status vector into a pre-trained wind power digital twin model for coupled state evolution simulation, outputting a fault propagation path; performing spatiotemporal encoding on operator behavior data, outputting a risk behavior warning area; performing virtual-real space risk coupling identification on the fault propagation path and the risk behavior warning area, generating a spatiotemporal risk coupling map; mapping the spatiotemporal risk coupling map to warning levels, outputting dynamic warning commands, and performing safety intervention on the wind turbine unit according to the dynamic warning commands, thus completing the monitoring of wind turbine unit safe operation. This invention achieves wind turbine unit fault prediction and personnel risk monitoring through the synergistic effect of coupled state evolution simulation and virtual-real space risk coupling identification.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring technology, and in particular to a method for monitoring the safe operation of wind turbine units based on digital twins. Background Technology

[0002] In recent years, wind turbine safety monitoring methods have evolved from traditional sensor threshold alarms to intelligent methods that integrate multi-source information. In the field of wind power safety monitoring, a basic data acquisition network is constructed by deploying vibration sensors, temperature sensors, and positioning sensors. This is combined with digital twin methods to establish a three-dimensional visualization model of the turbine, signal processing algorithms to detect abnormal wind turbine conditions, and electronic fences and trajectory tracking methods to monitor personnel's work area. This has gradually formed a monitoring system with data-driven operation at its core and virtual-real mapping as its means. The evolution of this monitoring system has significantly improved the safety monitoring coverage in complex terrain environments of wind farms, providing methodological support for responding to various emergencies.

[0003] However, existing methods have shortcomings in achieving accurate prediction of turbine failures and dynamic coupling analysis of personnel behavioral risks. Current monitoring systems struggle to overcome model inaccuracies caused by inconsistent data quality, there is a response delay between multiphysics coupled simulations and real-time operational data, fault propagation path deduction lacks a linkage analysis mechanism with personnel spatiotemporal trajectories, and a closed-loop control has not yet been formed between early warning decisions and on-site intervention measures. In particular, for high-risk scenarios such as high-altitude operations and live-line operations, existing methods are unable to achieve synchronous perception and intelligent assessment of wind turbine state evolution and personnel behavioral risks in both virtual and physical spaces. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for monitoring the safe operation of wind turbine units based on digital twins to address the shortcomings in achieving accurate prediction of unit faults and dynamic coupling analysis of personnel behavior risks.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for monitoring the safe operation of wind turbines based on digital twins. The method includes: collecting wind turbine status data and operator behavior data; performing data quality assessment and credibility labeling on the wind turbine status data to generate a quality-labeled data stream; performing multi-source status feature analysis on the quality-labeled data stream to output a turbine status vector; inputting the turbine status vector into a pre-trained wind power digital twin model for coupled state evolution simulation to output a state evolution map; performing fault chain analysis on the state evolution map to output a fault propagation path; performing spatiotemporal encoding on the operator behavior data to generate a personnel behavior trajectory sequence; performing dynamic risk coupling analysis on the personnel behavior trajectory sequence to output a risk behavior warning area; performing virtual-real space risk coupling identification on the fault propagation path and the risk behavior warning area to generate a spatiotemporal risk coupling map; mapping the spatiotemporal risk coupling map to warning levels and outputting dynamic warning commands; and performing safety intervention on the wind turbines according to the dynamic warning commands to complete the monitoring of wind turbine safe operation.

[0007] As a preferred embodiment of the wind turbine safety operation monitoring method based on digital twins described in this invention, the specific steps for performing data quality assessment and credibility labeling on the wind turbine status data to generate a quality-labeled data stream are as follows: Perform multi-dimensional data quality verification on wind turbine status data to generate verified turbine status data; The verification unit status data is quantified and labeled with a reliable measure, and the labeled unit status data is output. The unit status data is serialized and reconstructed according to timestamps to generate a quality-marked data stream.

[0008] As a preferred embodiment of the wind turbine safe operation monitoring method based on digital twins described in this invention, the specific steps for performing multi-source state feature analysis on the quality-marked data stream and outputting the turbine state vector are as follows: Perform feature association and multi-source data collaborative processing on the quality-labeled data stream to output a collaborative state dataset; Time-frequency domain features are extracted from the cooperative state dataset, and a normalization algorithm is used to unify the scale of the extracted time-frequency domain features, outputting a standard state feature set; Principal component analysis is used to reduce the dimensionality of the standard state feature set and output the unit state vector.

[0009] As a preferred embodiment of the wind turbine safe operation monitoring method based on digital twins described in this invention, the specific steps of inputting the turbine state vector into a pre-trained wind power digital twin model for coupled state evolution simulation and outputting a state evolution map are as follows: The unit state vector is input into a pre-trained wind power digital twin model for multi-physics coupling simulation, and the physical field coupling state data is output. Perform temporal state extrapolation analysis on the physical field coupling state data and output spatiotemporal evolution data; Spatiotemporal evolution data are graphically integrated to form a state evolution map.

[0010] As a preferred embodiment of the wind turbine safe operation monitoring method based on digital twins described in this invention, the specific steps for performing fault chain analysis on the state evolution graph and outputting the fault propagation path are as follows: A pattern matching algorithm is used to identify fault modes in the state evolution graph. The pre-stored graph pattern template is used to perform subgraph isomorphic matching on the state evolution graph to output a set of potential fault points. Perform causal correlation analysis on each fault point in the potential fault point set and output a fault propagation relationship diagram; Traverse the propagation path of the fault propagation graph, calculate the cumulative impact intensity of each propagation path in the graph, filter the path in the graph according to the cumulative impact intensity, and output the fault propagation path.

[0011] As a preferred embodiment of the wind turbine safe operation monitoring method based on digital twins described in this invention, the causal correlation analysis refers to using graph theory algorithms to analyze the causal relationship and propagation direction between each fault point, and performing directed edge connection and influence intensity quantification on each fault point according to the causal relationship and propagation direction.

[0012] As a preferred embodiment of the wind turbine safety operation monitoring method based on digital twins described in this invention, the specific steps for spatiotemporal encoding of operator behavior data to generate a personnel behavior trajectory sequence are as follows: The data on worker behavior is processed by timestamp alignment and spatial coordinate normalization to output standardized spatiotemporal data; Sliding window segmentation and motion pattern extraction are performed on standardized spatiotemporal data to output spatiotemporal coded fragments. Sequence recombination and trajectory smoothing are then performed on the spatiotemporal coded fragments to generate a sequence of human behavior trajectories.

[0013] As a preferred embodiment of the wind turbine safety operation monitoring method based on digital twins described in this invention, the specific steps of performing dynamic risk coupling analysis on personnel behavior trajectory sequences and outputting risk behavior early warning areas are as follows: Clustering algorithms are used to identify abnormal behavior patterns in personnel behavior trajectory sequences and output a set of risk behavior points. The risk behavior point set is correlated and coupled with the wind turbine status data to output risk coupling assessment data. The risk coupling assessment data is spatially divided to form risk behavior early warning zones.

[0014] As a preferred embodiment of the wind turbine safe operation monitoring method based on digital twins described in this invention, the specific steps for identifying the virtual and real spatial risk coupling of fault propagation paths and risk behavior early warning areas to generate a spatiotemporal risk coupling map are as follows: The fault propagation path and risk behavior warning area are aligned in time and space and fused with data to generate a fused risk dataset. Perform virtual-real space risk coupling identification on the fused risk dataset and output the risk coupling strength distribution; The risk coupling intensity distribution is integrated and mapped in a spatiotemporal dimension to generate a spatiotemporal risk coupling map.

[0015] As a preferred embodiment of the wind turbine safety operation monitoring method based on digital twins described in this invention, the specific steps for mapping early warning levels to the spatiotemporal risk coupling map and outputting dynamic early warning commands are as follows: The spatiotemporal risk coupling map and the preset early warning rule table are matched for early warning levels, and early warning level classification data is output. The warning level classification data is mapped and formatted for instruction encapsulation, and dynamic warning instructions are output.

[0016] The beneficial effects of this invention are as follows: By coordinating the dual mechanisms of coupled state evolution simulation and virtual-real space risk coupling identification, wind turbine fault prediction and personnel risk monitoring are achieved. By inputting the turbine state vector into a pre-trained digital twin model for multi-physics coupled simulation and temporal state deduction, the generated state evolution map reveals the spatiotemporal propagation patterns of fault chains, enabling in-depth insight and advanced early warning of potential wind turbine faults. Through the alignment of the fault propagation path with the spatiotemporal coordinates of the risk behavior warning area and the quantification of coupling strength, the generated spatiotemporal risk coupling map constructs a human-machine-environment risk linkage perception network. Dynamic instructions that can guide on-site intervention are formed through warning level mapping. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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 creative effort.

[0018] Figure 1 This is a flowchart of a method for monitoring the safe operation of wind turbine units based on digital twins.

[0019] Figure 2This is a flowchart for outputting the unit's state vector.

[0020] Figure 3 This is a flowchart to output the fault propagation path.

[0021] Figure 4 This is a flowchart for outputting dynamic early warning commands. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a method for monitoring the safe operation of wind turbine units based on digital twins, including the following steps: S1. Collect wind turbine status data and operator behavior data, perform data quality assessment and credibility labeling on wind turbine status data, generate quality-labeled data stream, perform multi-source status feature parsing on quality-labeled data stream, and output turbine status vector.

[0026] Collect wind turbine status data and operator behavior data, perform multi-dimensional data quality verification on the wind turbine status data, and generate verification turbine status data.

[0027] Specifically, wind turbine status data is collected in real time by vibration, temperature, speed, oil and electrical sensors deployed on wind turbine components (such as gearbox, main shaft, tower and blade root), and aggregated via edge gateway; worker behavior data is collected through smart wearable devices worn by workers (including smart safety helmets, smart bracelets or UWB / Bluetooth beacon tags with positioning, posture perception and identity recognition functions) to collect information such as location coordinates, movement trajectory, dwell time and operation actions.

[0028] The system performs an integrity check on the wind turbine status data by using a sliding time window to detect missing values ​​in the time interval between adjacent sampling points. Missing values ​​are filled using linear interpolation, where a straight line is connected between two known data points before and after the missing point, and the median value of the connection is used as the filler value according to the time ratio, generating a continuous time series. A consistency check is performed on the wind turbine status data by comparing the values ​​from different sensor channels at the same time point to see if they satisfy the physical constraints of the wind turbine (such as the monotonically increasing function relationship between generator speed and active power, the adjustment logic of pitch angle and wind speed, the positive correlation between temperature and load, and the balance of three-phase current). If the physical constraints are not satisfied, the data at the current time point is marked as inconsistent and discarded. Finally, a repeatability check is performed on the wind turbine status data by comparing records with identical values ​​in all dimensions under consecutive timestamps, retaining the earliest timestamp record, deleting duplicate records, and generating verification turbine status data.

[0029] The verification unit status data is quantified and labeled with a reliable metric, and the labeled unit status data is output.

[0030] Specifically, based on the type of sensors collected by the wind turbine, the stability of the installation location, and the consistency of historical data, a weighting factor is assigned to each record in the verification unit status data. A fuzzy comprehensive evaluation method is used to weight and synthesize the weighting factor with the verification unit status data based on three indicators: time continuity, physical rationality, and channel coordination, to calculate the credibility score of each record. According to the credibility score, each record is marked as high credibility, medium credibility, or low credibility. The marking results are embedded into the verification unit status data as additional fields, and the marked unit status data is output.

[0031] Furthermore, the credibility level classification is based on the normalized credibility score range: high credibility (≥0.8), medium credibility (0.5–0.8), and low credibility (<0.5). The classification is based on historical data statistical analysis. In a large number of normal operation samples, the fuzzy comprehensive evaluation score of more than 90% of the valid data is concentrated above 0.8, while the data under interference or edge conditions is between 0.5 and 0.8. Data below 0.5 is usually accompanied by sensor drift, communication anomalies, or physical contradictions, and the false alarm rate is significantly increased. The high credibility, medium credibility, and low credibility level settings take into account both reliability and practicality.

[0032] The unit status data is serialized and reconstructed according to timestamps to generate a quality-marked data stream.

[0033] Specifically, the timestamp field of each record in the unit status data is extracted through data query, and all records are sorted in ascending order of timestamp; the sorted records are resampled, and if there is no corresponding record at the sampling time, the data at the sampling time is generated by linear interpolation of adjacent records before and after, and the confidence mark is retained; the resampled records are encapsulated into a continuous and equally spaced data sequence to generate a quality mark data stream.

[0034] Perform feature association and multi-source data collaborative processing on the quality-labeled data stream to output a collaborative state dataset.

[0035] Specifically, vibration signals, temperature signals, speed signals, oil parameter signals, and electrical parameter signals are extracted from the quality marker data stream. Based on the mutual information method, the degree of dependence among the vibration signals, temperature signals, speed signals, oil parameter signals, and electrical parameter signals is statistically analyzed (i.e., the pattern of simultaneous occurrence of any two signals under different values ​​is statistically analyzed, and the amount of information shared between any two signals is measured as the degree of dependence). Based on a preset mutual information threshold, it is determined whether an association relationship is established (an association relationship refers to a statistically significant information dependency relationship between two signals). If the degree of dependence is less than or equal to the preset mutual information threshold, it is determined that the association condition is not met; if the degree of dependence is greater than the preset mutual information threshold, it is determined that the association condition is met.

[0036] For vibration, temperature, speed, oil parameter, and electrical parameter signals that meet the correlation conditions, timestamp alignment is performed, and weighted fusion is carried out according to the normalization ratio of their respective confidence labels (the normalization ratio refers to using the confidence labels of each signal as relative weights and normalizing the relative weights so that the sum of the weights is 1, thereby determining the proportion of each signal in the fusion). Vibration, temperature, speed, oil parameter, and electrical parameter signals that do not meet the correlation conditions are kept as independent channels. A sliding window is used to divide all fused signals and independent channel signals into a unified time step. Within each sliding window, each signal is weighted and combined according to its confidence label to form the collaborative feature vector of the current window. The collaborative feature vectors of all windows are arranged in chronological order, and the collaborative state dataset is output.

[0037] Furthermore, the preset mutual information threshold is determined by analyzing the mutual information distribution of each signal pair under historical normal operating conditions, and 75% of the dependence of all signals is taken as the threshold. The 75% value can effectively distinguish between strongly correlated signals with stable physical coupling relationships (such as rotational speed and vibration) and weakly correlated signals affected by noise or independent disturbances. Using a 75% mutual information threshold can avoid information confusion caused by excessive fusion of irrelevant signals, while retaining multi-source features with mechanistic correlations, and improving the physical consistency of the cooperative state dataset and the reliability of subsequent modeling.

[0038] Time-frequency domain features are extracted from the cooperative state dataset, and a normalization algorithm is used to unify the scale of the extracted time-frequency domain features, outputting a standard state feature set.

[0039] Specifically, a short-time Fourier transform is performed on each dimension of the signal in the cooperative state dataset to obtain the spectral amplitude sequence and phase sequence. That is, the signal in each dimension of the cooperative state dataset is framed using a Hanning window, and a discrete Fourier transform is performed on each frame to obtain the complex spectrum corresponding to the current frame. The complex spectra of all frames are arranged in chronological order, and the magnitude of the complex spectrum is taken to form the spectral amplitude sequence, and the argument of the complex spectrum is taken to form the phase sequence. The center frequency, frequency band energy ratio, and spectral entropy are extracted from the spectral amplitude sequence, and the phase difference change rate is extracted from the phase sequence. The center frequency, frequency band energy ratio, spectral entropy, and phase difference change rate are concatenated into a feature vector. The global mean and standard deviation of each feature component in all feature vectors are calculated, and the Z-score normalization algorithm is used to unify the scale of each feature component. The normalized feature vectors are organized to output the standard state feature set.

[0040] Principal component analysis is used to reduce the dimensionality of the standard state feature set and output the unit state vector.

[0041] Specifically, covariance statistics are performed on all eigenvectors in the standard state feature set to obtain the covariance matrix; eigenvalue decomposition is performed on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors; principal components are selected in descending order of eigenvalues ​​to form the projection matrix; each eigenvector in the standard state feature set is fused with the projection matrix to obtain the dimension-reduced representation vector, i.e., the unit state vector.

[0042] S2. Input the unit state vector into the pre-trained wind power digital twin model to perform coupled state evolution simulation, output the state evolution map, and perform fault chain analysis on the state evolution map to output the fault propagation path.

[0043] The unit state vector is input into a pre-trained wind power digital twin model for multi-physics coupling simulation, and the physical field coupling state data is output.

[0044] Specifically, after inputting the unit state vector into the pre-trained wind power digital twin model, the wind power digital twin model uses the unit state vector as a representation of the current operating state. Combining the structural parameters of the wind turbine with the multiphysics coupling mechanism, it synchronously simulates multiple interrelated physical responses within the same time step: estimating aerodynamic loads based on information such as rotational speed, power, and pitch angle, and calculating the deformation displacement of the blades in space based on the aerodynamic loads; applying the aerodynamic loads, gravity, and centrifugal force together to the overall structural dynamics to obtain the vibration response of the tower; determining the contact stress of key parts of the gearbox based on the principle of contact mechanics using the main shaft torque, transmission ratio, and gear geometry; determining the temperature distribution of the generator windings and core and the aerodynamic power under the current operating conditions based on electrical losses and heat conduction laws; and organizing the multiphysics coupling simulation results of each time step in chronological order, outputting the physical field coupling state data.

[0045] Furthermore, the pre-trained wind power digital twin model is jointly trained based on historical operating data and high-fidelity multiphysics simulation data. The training data includes the turbine state vectors and corresponding physical field responses (such as structural displacement, temperature field, and aerodynamic forces) under different operating conditions. The training process adopts a supervised learning approach, using the turbine state vectors as input and the coupled physical field state data as labels, and using mean squared error as the loss function. The training set, validation set, and test set are divided in a 7:2:1 ratio. This 7:2:1 ratio ensures that the wind power digital twin model has sufficient data for parameter learning (70% for the training set) while retaining sufficient independent data for hyperparameter tuning (20% for the validation set) and final performance evaluation (10% for the test set). The 7:2:1 ratio is suitable for wind turbine operating datasets, ensuring that the validation set has sufficient representativeness. To effectively monitor the model's generalization ability and promptly trigger the early stopping mechanism to prevent overfitting, ensuring that each set covers the full range of operating conditions, the wind power digital twin model training adopts an early stopping mechanism. Training is terminated when the validation set loss has not decreased for consecutive rounds (consecutive rounds usually refer to 5 to 10 training rounds, the specific value is determined by parameter tuning on the validation set: if the validation loss does not continuously decrease during the period of 5 to 10 training rounds, it is considered that the wind power digital twin model has begun to overfit or has stalled in convergence. 5 to 10 training rounds need to balance the sufficiency of training and the risk of overfitting - too few rounds will easily lead to early stopping and underfitting, while too many rounds will waste computing power and may lead to overfitting; in the wind turbine data scenario, because the operating conditions change slowly and the data distribution is relatively stable, 5 consecutive rounds are often taken as the default early stopping window) to prevent overfitting. At the same time, an L2 weight regularization term is introduced, and batch normalization and Dropout are combined to further improve the generalization ability.

[0046] Perform time-series state extrapolation analysis on the physical field coupling state data and output spatiotemporal evolution data.

[0047] Specifically, the physical field coupled state data is organized into a multi-dimensional time series according to the time step sequence. Based on the blade deformation displacement, tower vibration response, gearbox contact stress, generator temperature distribution, and aerodynamic power output of the current time step, combined with the preset state evolution rules, the predicted state value of the next time step is statistically calculated. A rolling prediction method is adopted to gradually extrapolate from the initial time step to generate a continuous state sequence of several future time steps (several time steps specifically refer to the next 5 time steps, the values ​​of which are determined based on the wind turbine fault development time scale (usually in the range of several seconds to tens of seconds) and the safety intervention response window, which can capture the early state evolution trend and avoid error accumulation caused by excessively long predictions). The predicted state values ​​of all time steps are bound with the corresponding spatial location information to form a joint data structure containing time and spatial dimensions, and spatiotemporal evolution data is output.

[0048] Furthermore, the preset state evolution rules are a set of dynamic evolution equations derived offline from the multi-physics mechanism model of the wind turbine, specifically including: structural dynamics equations (describing the relationship between displacement and stress over time), heat conduction differential equations (describing the diffusion law of temperature field in components), aerodynamic power recursive formulas (corresponding to the dynamic response of wind speed, rotational speed and output power), and electromagnetic torque balance equations. The state evolution rules are formed by performing long-term simulations of the pre-trained wind power digital twin model under different operating conditions, extracting the temporal dependencies between state variables, and solidifying them in the form of difference or state-space equations for use in temporal state deduction analysis.

[0049] Spatiotemporal evolution data are graphically integrated to form a state evolution map.

[0050] Specifically, blade deformation displacement, tower vibration response, gearbox contact stress, generator temperature distribution, and aerodynamic power output at each time step are extracted from the spatiotemporal evolution data. The distribution of each physical quantity in space is mapped as a state node, with the time step as the time dimension. The changes in physical quantities at the same spatial location in adjacent time steps are used as directed edge weights to construct a spatiotemporal directed graph. Cross-channel connection edges are established for the coupling relationship of different physical quantities at the same spatiotemporal location. All state nodes and attributes are organized into a unified graph structure in chronological order to form a state evolution graph.

[0051] A pattern matching algorithm is used to identify fault modes in the state evolution graph. The pre-stored graph pattern template is used to perform subgraph isomorphic matching on the state evolution graph to output a set of potential fault points.

[0052] Specifically, based on the temporal characteristics of physical quantities and the evolution trend of edge weights of each state node in the state evolution graph, node-level feature vectors are constructed. These node-level feature vectors are then compared with pre-stored graph pattern templates for five typical faults: gearbox tooth breakage, main shaft misalignment, blade cracking, generator winding overheating, and tower resonance. (The graph pattern templates are created by clustering and typifying the state evolution graphs collected from wind turbines under historical fault conditions. Clustering uses a hierarchical clustering method based on graph edit distance to group the historical fault state evolution graphs. Typification selects the subgraph with the smallest average edit distance from each cluster as the standardized graph pattern template for the fault type, extracting the stable graph structure features corresponding to the five faults: gearbox tooth breakage, main shaft misalignment, blade cracking, generator winding overheating, and tower resonance.) Isomorphic matching involves determining whether local subgraphs in the state evolution graph have a one-to-one mapping relationship with pre-stored fault graph pattern templates in terms of structure (node ​​connection relationships) and attributes (physical quantity characteristics). For each graph pattern template, it identifies subgraph instances in the graph that meet the isomorphic condition and assigns a matching score (the matching score is obtained by calculating the structural similarity and attribute deviation between the local subgraphs in the state evolution graph and the pre-stored fault graph pattern templates, combining the quantitative indicators of structural similarity and attribute deviation). If the matching score is greater than or equal to a preset matching threshold, the component corresponding to the state node is determined to be in a potential fault state; if the overall matching score is less than the preset matching threshold, the component corresponding to the state node is determined to be in a normal state. The physical locations and time markers associated with all state nodes determined to be in potential fault states are summarized, and a set of potential fault points is output.

[0053] Furthermore, the preset matching threshold is set by statistically analyzing the distribution of pattern matching scores under historical normal operating conditions and known fault operating conditions. The median value corresponding to the minimum overlap area between the score distributions of historical normal operating conditions and known fault operating conditions is taken as the preset matching threshold. For example, the preset matching threshold is set to 0.75. The reason for setting 0.75 is that the matching score of normal operating conditions in historical data is mostly lower than 0.7, while the score of fault operating conditions is generally higher than 0.8. Choosing 0.75 as the matching threshold can effectively reduce false alarms while ensuring a high detection rate, and avoid misjudging normal state fluctuations as potential faults.

[0054] Perform causal correlation analysis on each fault point in the potential fault point set and output a fault propagation relationship diagram.

[0055] Specifically, a fault node graph is constructed based on a set of potential fault points, with each fault point serving as a fault node. Based on the physical structure and operating mechanism of the wind turbine, graph theory algorithms are used to statistically analyze the transmission probability and influence intensity between each fault node. That is, based on the physical structure and operating mechanism of the wind turbine, the physical connections and causal dependencies between each fault node are determined. The frequency of time lag between the first occurrence of a fault at one node and the subsequent occurrence of an anomaly at another node in historical fault data is statistically analyzed, and the co-occurrence frequency is used as the transmission probability. Simultaneously, the magnitude coupling degree of physical quantity changes between each fault node (e.g., after one node fails, another...) is analyzed. The degree to which the physical quantity of a node deviates from the normal range is quantified to determine the impact intensity. The propagation probability and impact intensity are combined into a single weight value and assigned to the corresponding directed edge. All possible propagation paths from the initial fault node to subsequent affected nodes are identified through path search algorithms (such as Dijkstra's algorithm or A* algorithm), and the impact weight of each edge on each path is quantified. Directed edges are connected to each fault point according to causal relationship and propagation direction to form a fault propagation relationship graph containing directed edges and weights, where the weights represent the probability or impact intensity of fault propagation, thus intuitively showing the causal relationship and propagation path between each fault point.

[0056] Traverse the propagation path of the fault propagation graph, calculate the cumulative impact intensity of each propagation path in the graph, filter the path in the graph according to the cumulative impact intensity, and output the fault propagation path.

[0057] Specifically, all fault nodes with an in-degree of zero (meaning no fault node is pointed to in the fault propagation graph, i.e., the fault node is not a subsequent influence of any other fault node) are identified as initial fault sources from the fault propagation graph. A depth-first search algorithm is used to traverse all reachable paths along the directed edges, recording each complete propagation path and the sequence of fault nodes it contains. The cumulative influence intensity of each propagation path is calculated, which is the product of the weights of each edge on the path. Based on the cumulative influence intensity, all propagation paths are sorted and filtered, and the top three propagation paths with the highest cumulative influence intensity are retained. In practice, wind turbine fault chains mainly develop along a few high-influence paths, and retaining the top three can cover the dominant fault evolution path while avoiding redundant interference. The retained propagation paths are organized in chronological and causal order, and the fault propagation paths are output.

[0058] S3. Spatiotemporal encoding of worker behavior data to generate worker behavior trajectory sequences, and dynamic risk coupling analysis of worker behavior trajectory sequences to output risk behavior warning areas.

[0059] The data on worker behavior is processed by timestamp alignment and spatial coordinate normalization to output standardized spatiotemporal data.

[0060] Specifically, timestamps recorded by each sensor are extracted from the operator's behavior data. Using the wind turbine's main control clock as a reference, linear interpolation is used to resample the asynchronously acquired position coordinates to align the timestamps of all behavior records. The aligned position coordinates are then transformed from the local coordinate system to the wind turbine's unified global coordinate system. Based on the spatial boundaries of key areas of the wind turbine (such as the center of the tower base, the center of the nacelle, and the position of the blade root), the coordinate values ​​are min-max normalized to map all spatial coordinates to the [0, 1] interval. The time series after timestamp alignment is combined with the normalized spatial coordinates to form a unified spatiotemporal record and output standardized spatiotemporal data.

[0061] Sliding window segmentation and motion pattern extraction are performed on standardized spatiotemporal data to output spatiotemporal coded fragments. Sequence recombination and trajectory smoothing are then performed on the spatiotemporal coded fragments to generate a sequence of human behavior trajectories.

[0062] Specifically, the standardized spatiotemporal data is segmented into sliding windows, with each window containing continuous timestamps and normalized spatial coordinates. For each window, the spatiotemporal point sequence is statistically analyzed for displacement vectors, velocity amplitudes, rate of change of direction, and dwell time characteristics, forming a feature vector representing the movement state of personnel within the current window, which serves as a spatiotemporal encoded segment. All spatiotemporal encoded segments are concatenated according to the window's start time sequence to form a preliminary trajectory sequence representation. A Savitzky-Golay filter is used to perform local polynomial smoothing on the spatial coordinate components of the preliminary trajectory sequence representation. Specifically, the Savitzky-Golay filter slides a window along the trajectory, fitting a smooth curve for each coordinate point within the window, and replacing the intermediate values ​​of the spatial coordinate components with the values ​​at the midpoints of the curve, suppressing high-frequency jitter noise and preserving the overall movement trend. The smoothed trajectory data is then reconstructed into a continuous sequence, i.e., a sequence of personnel behavior trajectories.

[0063] Clustering algorithms are used to identify abnormal behavior patterns in personnel behavior trajectory sequences and output a set of risk behavior points.

[0064] Specifically, by querying data, behavioral features are extracted from the personnel behavior trajectory sequence in four dimensions: velocity amplitude, rate of change of direction, minimum distance to key component areas of the wind turbine (including tower inlet, nacelle platform, pitch control cabinet, and gearbox maintenance port), and dwell time, forming a behavioral feature vector. All behavioral feature vectors are input into the DBSCAN clustering algorithm. Based on preset neighborhood radius and minimum neighbor number parameters, points within the neighborhood radius containing more than or equal to the minimum neighbor number are classified as core points, and core points and points within the neighborhood are grouped into the same cluster. Trajectory points not classified into any cluster are marked as outliers. The timestamps and spatial coordinates corresponding to all outliers are summarized to output a risk behavior point set.

[0065] Furthermore, the preset neighborhood radius and minimum neighbor number parameters are determined by analyzing historical normal work behavior data: different neighborhood radii and minimum neighbor numbers are scanned on historical normal samples of personnel behavior trajectory sequences, and the silhouette coefficient and outlier ratio of the clustering results under each combination are statistically analyzed. Typical values ​​are a neighborhood radius of 0.2 and a minimum neighbor number of 5. Typical values ​​can effectively distinguish normal patterns such as routine inspections and short stays from abnormal behaviors such as long-term stays and unauthorized approach, taking into account both sensitivity and robustness.

[0066] The risk behavior point set is correlated and coupled with the wind turbine status data to output risk coupling assessment data.

[0067] Specifically, for each outlier in the risk behavior point set, the corresponding wind turbine operating status parameters, including speed, vibration amplitude, temperature, and active power, are queried from the wind turbine status data. These parameters are then normalized using the Z-score normalization method. The spatial distance between the outlier and the nearest critical component area of ​​the wind turbine is calculated, and a weighted sum is obtained by combining this distance with the corresponding wind turbine operating status parameters. The weights of the wind turbine operating status parameters are determined by analyzing the frequency of occurrence and deviation from normal operating conditions of each parameter (such as speed, vibration amplitude, temperature, and active power) in safety accidents or high-risk events in historical operating data. Statistical methods are used to quantify their contribution to risk, and the weights of each parameter are normalized to a sum of 1 to ensure the consistency and comparability of the risk coupling scores. Finally, the risk coupling scores of all outliers are combined with the timestamps, spatial coordinates, and associated wind turbine status parameters to output risk coupling assessment data.

[0068] The formula for calculating the risk coupling score is as follows: ; in, This represents the risk coupling score of the current outlier. The weight representing the rotational speed, This represents the normalized rotational speed value. The weight representing the amplitude of vibration, This represents the normalized vibration amplitude. The weighting of temperature This represents the normalized temperature value. The weight representing active power. This represents the normalized active power value. Weights representing spatial distance This represents the normalized spatial distance value.

[0069] Furthermore, the exemplary weights for each parameter are: rotational speed 0.20, vibration amplitude 0.30, temperature 0.15, active power 0.10, and spatial distance 0.25. These values ​​are based on statistical analysis of historical high-risk events: vibration amplitude significantly deviates from the normal range in the vast majority of mechanical anomalies and is highly correlated with the severity of the fault, thus receiving the highest weight; spatial distance directly reflects whether personnel are near key hazardous areas such as tower inlets and gearbox maintenance ports, playing a decisive role in most violations or near-miss incidents, hence its second-highest weight; rotational speed significantly increases operational risks at high speeds, such as exposure of rotating parts and difficulties in emergency response, thus its weight is moderate; temperature, as an indirect indicator of equipment overheating, has some early warning value, but its coupling with personnel behavior is weak, resulting in a lower weight; active power has the least correlation with personnel behavior, only slightly increasing risk during full-load operation, thus receiving the lowest weight; the sum of all weights is 1, ensuring the risk coupling score is comparable and meaningful across different scenarios.

[0070] The risk coupling assessment data is spatially divided to form risk behavior early warning zones.

[0071] Specifically, the spatial coordinates of each outlier in the risk coupling assessment data are mapped to the corresponding location of the wind turbine. Centered on the key component area of ​​the wind turbine, based on the risk coupling score of the outliers, contour clustering is used to aggregate outliers with risk coupling scores within the same preset interval and spatial distances not exceeding a set clustering threshold into a group. Convex hull calculation is then performed on the aggregated point clusters (convex hull calculation refers to finding the smallest convex polygon on a plane that can contain all points in the point cluster, with the boundary formed by connecting the outermost extreme points of the point cluster in a clockwise or counterclockwise order), generating a closed polygonal region, i.e., the risk behavior warning region. The risk behavior warning region is then... The domain is classified into three risk levels: high risk, medium risk, and low risk. The classification is based on the risk coupling score range of outliers: a score ≥ 0.75 is high risk, 0.50–0.74 is medium risk, and < 0.50 is low risk. The values ​​are based on historical accident statistics—scores above 0.75 indicate that the sample is accompanied by actual near misses or violations; scores below 0.50 are mostly brief approaches during normal inspections with low accident correlation; the middle range is a transitional state. The classification takes into account both safety sensitivity and operational practicality, ensuring that high-level areas focus on behaviors that truly require intervention, and outputting risk behavior warning areas that include the risk level of the risk area.

[0072] Furthermore, the clustering threshold is typically set at 2 meters. This 2-meter threshold is determined based on the on-site safety requirements of wind turbine operations and the accuracy of personnel positioning: the typical error of personnel locators (such as UWB) is 0.5–1 meter, and 2 meters can effectively cover positioning noise; the safety control area around key wind power components (such as tower inlets and gearbox maintenance ports) is usually set at a warning radius of 1.5–3 meters, and 2 meters can reasonably aggregate spatial points belonging to the same risk behavior, avoiding excessive splitting or merging of outliers in different areas, thereby ensuring the physical meaning and practicality of the warning area.

[0073] S4. Perform virtual and real space risk coupling identification on fault propagation paths and risk behavior warning areas to generate a spatiotemporal risk coupling map.

[0074] By aligning the fault propagation path and the risk behavior warning area with spatiotemporal coordinates and performing data fusion processing, a fused risk dataset is generated.

[0075] Specifically, the timestamps and spatial coordinates of each fault point in the fault propagation path, as well as the timestamps and spatial coordinates of each polygonal region in the risk behavior warning area, are uniformly transformed to the global spatiotemporal coordinate system of the wind turbine. The fault propagation path and the risk behavior warning area are resampled along the same time axis to align the fault event and the personnel risk area at the same time step. Within each aligned time step, the spatial intersection of the fault point in the fault propagation path and the risk behavior warning area is calculated. If there is a spatial intersection between the fault point in the fault propagation path and the risk behavior warning area, it is marked as having a spatiotemporal coupling risk; if there is no spatial intersection, it is marked as not having a spatiotemporal coupling risk. The coupling judgment results in all time steps, the corresponding fault point attributes, risk area level, and spatial location information are combined to generate a fused risk dataset.

[0076] Perform virtual-real space risk coupling identification on the fused risk dataset and output the risk coupling strength distribution.

[0077] Specifically, the fault point attributes, risk area level, and spatial location information for each time step are extracted from the fused risk dataset. Based on the fault type and impact range of the fault point in the fault propagation path, combined with the risk level of the risk behavior warning area, the risk coupling mapping table is queried (the risk coupling mapping table is generated by offline analysis of historical concurrent fault events and personnel operation records: the confirmed fault type, impact range, risk area level of personnel in the same period, and whether a safety accident has occurred are extracted from historical operation logs; the frequency of safety accidents under each "fault type - risk area level" combination is counted, and the frequency is normalized to a value between 0 and 1, which serves as the coupling basis weight for the corresponding combination, thus forming the risk coupling mapping table). The corresponding coupling basis weight is obtained. When there is a spatiotemporal coupling risk between the fault point and the risk behavior warning area, the coupling basis weight is fused with the spatial overlap area ratio to obtain the risk coupling intensity value of the current spatiotemporal location. The risk coupling intensity values ​​corresponding to all time steps and spatial locations are interpolated and filled according to the spatial grid of the wind turbine to form a continuous numerical field covering the entire scenario, and the risk coupling intensity distribution is output.

[0078] The formula for the risk coupling strength value is: ; in, This represents the risk coupling strength value at the current spatiotemporal location. Indicates the basic weights of the coupling. This indicates the proportion of spatial overlap between the fault point and the risk behavior warning area in the fault propagation path.

[0079] The risk coupling intensity distribution is integrated and mapped in a spatiotemporal dimension to generate a spatiotemporal risk coupling map.

[0080] Specifically, the risk coupling strength values ​​at each time step are arranged according to the time sequence to form a continuous change sequence in the time dimension. For the spatial dimension, interpolation processing is performed on the risk coupling strength distribution at different times using the spatial grid data of the wind turbine units to ensure seamless connection of the continuous numerical field across the entire scenario, thereby achieving spatial integration. Based on the integrated time and spatial dimension data, a graph-based coding method is used to transform the time and space information and risk coupling strength values ​​into a unified spatiotemporal risk coupling graph representation. That is, the location of each wind turbine unit is used as a fixed spatial point, combined with the risk coupling strength value obtained at each time step to form a "location-time" pair. Interpolation is performed on the gaps between adjacent units and between adjacent time steps to construct a continuous spatiotemporal grid covering the entire field. Graph-based coding uses each spatiotemporal point in the spatiotemporal grid as a graph node, with risk coupling strength as the node attribute, and establishes edges according to time sequence and spatial adjacency to generate a unified spatiotemporal risk coupling graph. The spatiotemporal risk coupling graph accurately reflects the risk coupling situation of wind turbine units at various time points and spatial locations.

[0081] S5. Map the spatiotemporal risk coupling map to the early warning level, output dynamic early warning instructions, and perform safety intervention on the wind turbine according to the dynamic early warning instructions to complete the monitoring of wind turbine safe operation.

[0082] The spatiotemporal risk coupling map and the preset early warning rule table are matched for early warning levels, and early warning level classification data is output.

[0083] Specifically, the risk coupling intensity value, corresponding fault type, risk behavior warning area level, and timestamp for each spatiotemporal location are extracted from the spatiotemporal risk coupling map. A pre-defined warning rule table is consulted to find rule entries that match the current fault type and risk area level combination. These rule entries define the mapping relationship between the risk coupling intensity value range and the warning level (e.g., Level 1, Level 2, and Level 3). The corresponding warning level is determined based on the range into which the extracted risk coupling intensity value falls. The warning levels for all spatiotemporal locations are then bound to timestamps, spatial coordinates, and associated fault and behavior information to output warning level classification data.

[0084] Furthermore, the pre-defined early warning rule table is constructed by analyzing historical safety events and operational intervention records. It contains multiple rule entries, each consisting of a quadruple of fault type, risk behavior early warning area level, risk coupling strength value range, and corresponding early warning level. The risk coupling strength value range is divided based on the statistical relationship between different strength values ​​in historical data and the frequency of actual safety accidents or human interventions. The risk coupling strength values ​​corresponding to safety accidents or triggering human interventions in all historical records are statistically analyzed, and an empirical cumulative distribution is plotted. The strength value range corresponding to the cumulative probability of 0%–20% in the empirical cumulative distribution is divided into a three-level early warning range (low risk), 20%–60% into a two-level early warning range (medium risk), and 60%–100% into a one-level early warning range (high risk). This division ensures that the high early warning level covers the vast majority of serious events in history, achieving a balance between early warning sensitivity and operability.

[0085] The warning level classification data is mapped and formatted for instruction encapsulation, and dynamic warning instructions are output.

[0086] Specifically, based on the warning level of each spatiotemporal location in the warning level classification data, the system queries the preset warning instruction mapping table (the preset warning instruction mapping table is formulated according to the wind turbine safe operation requirements and emergency response requirements, binding different warning levels with corresponding standardized control action instructions; Level 1 warning corresponds to the highest risk scenario, triggering forced shutdown and personnel entry prohibition measures; Level 2 warning corresponds to medium risk, executing power limitation and area warning; Level 3 warning corresponds to low-risk anomaly, only initiating prompts and monitoring pushes) to obtain the corresponding control action instructions; the obtained control action instructions, along with the corresponding timestamp, spatial coordinates, fault type, and risk behavior information, are structured and encapsulated according to a unified communication protocol format to generate instruction messages that conform to the wind turbine safety control center interface specifications, and output dynamic warning instructions.

[0087] Safety interventions are carried out on wind turbines in accordance with dynamic early warning instructions to complete the monitoring of safe operation of wind turbines.

[0088] Specifically, dynamic early warning commands are transmitted to the wind turbine safety control center. The wind turbine safety control center parses the control action commands, timestamps, spatial coordinates, fault types, and risk behavior information in the command messages. Based on the control action commands, corresponding operations are executed: if it is a Level 1 early warning command, the main power supply of the wind turbine is cut off and the tower entrance access is locked; if it is a Level 2 early warning command, the upper limit of the active power output of the wind turbine is limited and the audible and visual alarms in the risk area are activated; if it is a Level 3 early warning command, the audible and visual prompts are triggered and the risk event message is pushed to the monitoring center. After the safety intervention is executed, the wind turbine safety control center provides feedback on the execution status, completing the monitoring of wind turbine safe operation.

[0089] This embodiment also provides a computer device applicable to the wind turbine safety operation monitoring method based on digital twins, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the wind turbine safety operation monitoring method based on digital twins as proposed in the above embodiment.

[0090] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0091] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the method for monitoring the safe operation of wind turbine units based on digital twins as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0092] In summary, this invention achieves wind turbine fault prediction and personnel risk monitoring through the synergistic effect of coupled state evolution simulation and virtual-real space risk coupling identification. By inputting the turbine state vector into a pre-trained digital twin model for multi-physics coupled simulation and temporal state extrapolation, the generated state evolution map reveals the spatiotemporal propagation patterns of fault chains, enabling in-depth insight and advanced early warning of potential wind turbine faults. By aligning the spatiotemporal coordinates of fault propagation paths with risk behavior warning areas and quantifying the coupling strength, the generated spatiotemporal risk coupling map constructs a human-machine-environment risk linkage perception network. Through warning level mapping, dynamic instructions that can guide on-site intervention are formed.

[0093] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A wind turbine generator safety operation monitoring method based on digital twinning, characterized in that: include, Collect wind turbine status data and operator behavior data, perform data quality assessment and credibility labeling on wind turbine status data, generate quality-labeled data stream, perform multi-source status feature parsing on quality-labeled data stream, and output turbine status vector; The unit state vector is input into a pre-trained wind power digital twin model for coupled state evolution simulation, outputting a state evolution map, and performing fault chain analysis on the state evolution map to output the fault propagation path; Spatiotemporal coding of worker behavior data is performed to generate a sequence of worker behavior trajectories. Dynamic risk coupling analysis is then conducted on the sequence of worker behavior trajectories to output risk behavior warning areas. The fault propagation path and risk behavior warning area are identified through virtual and real space risk coupling to generate a spatiotemporal risk coupling map. The spatiotemporal risk coupling map is mapped to a warning level, and dynamic warning commands are output. Safety intervention is carried out on wind turbines according to the dynamic warning commands to complete the monitoring of safe operation of wind turbines.

2. The digital-twin-based wind turbine safety operation monitoring method of claim 1, wherein: The specific steps for performing data quality assessment and reliability labeling on wind turbine status data to generate a quality-labeled data stream are as follows: Perform multi-dimensional data quality verification on wind turbine status data to generate verified turbine status data; The verification unit status data is quantified and labeled with a reliable measure, and the labeled unit status data is output. The unit status data is serialized and reconstructed according to timestamps to generate a quality-marked data stream.

3. The digital-twin-based wind turbine safety operation monitoring method of claim 1, wherein: The specific steps for performing multi-source state feature parsing on the quality-marked data stream and outputting the unit state vector are as follows: Perform feature association and multi-source data collaborative processing on the quality-labeled data stream to output a collaborative state dataset; Time-frequency domain features are extracted from the cooperative state dataset, and a normalization algorithm is used to unify the scale of the extracted time-frequency domain features, outputting a standard state feature set; Principal component analysis is used to reduce the dimensionality of the standard state feature set and output the unit state vector.

4. The digital-twin-based wind turbine safety operation monitoring method of claim 1, wherein: The specific steps are as follows: inputting the unit state vector into a pre-trained wind power digital twin model for coupled state evolution simulation, and outputting a state evolution map. The unit state vector is input into a pre-trained wind power digital twin model for multi-physics coupling simulation, and the physical field coupling state data is output. Perform temporal state extrapolation analysis on the physical field coupling state data and output spatiotemporal evolution data; Spatiotemporal evolution data are graphically integrated to form a state evolution map.

5. The digital-twin-based wind turbine safety operation monitoring method of claim 1, wherein: The specific steps for performing fault chain analysis on the state evolution graph and outputting the fault propagation path are as follows: A pattern matching algorithm is used to identify fault modes in the state evolution graph. The pre-stored graph pattern template is used to perform subgraph isomorphic matching on the state evolution graph to output a set of potential fault points. Perform causal correlation analysis on each fault point in the potential fault point set and output a fault propagation relationship diagram; Traverse the propagation path of the fault propagation graph, calculate the cumulative impact intensity of each propagation path in the graph, filter the path in the graph according to the cumulative impact intensity, and output the fault propagation path.

6. The digital-twin-based wind turbine safety service monitoring method of claim 5, wherein: The causal correlation analysis refers to using graph theory algorithms to analyze the causal relationships and propagation directions between each fault point, and then connecting each fault point with directed edges and quantifying the influence intensity according to the causal relationships and propagation directions.

7. The digital-twin-based wind turbine safety operation monitoring method of claim 1, wherein: The specific steps for performing spatiotemporal encoding on the worker behavior data to generate a sequence of worker behavior trajectories are as follows: The data on worker behavior is processed by timestamp alignment and spatial coordinate normalization to output standardized spatiotemporal data; Sliding window segmentation and motion pattern extraction are performed on standardized spatiotemporal data to output spatiotemporal coded fragments. Sequence recombination and trajectory smoothing are then performed on the spatiotemporal coded fragments to generate a sequence of human behavior trajectories.

8. The digital-twin-based wind turbine safety operation monitoring method of claim 1, wherein: The specific steps for performing dynamic risk coupling analysis on personnel behavior trajectory sequences and outputting risk behavior early warning areas are as follows: Clustering algorithms are used to identify abnormal behavior patterns in personnel behavior trajectory sequences and output a set of risk behavior points. The risk behavior point set is correlated and coupled with the wind turbine status data to output risk coupling assessment data. The risk coupling assessment data is spatially divided to form risk behavior early warning zones.

9. The digital-twin-based wind turbine safety operation monitoring method of claim 1, wherein: The specific steps for identifying the fault propagation path and risk behavior warning area through virtual and real spatial risk coupling to generate a spatiotemporal risk coupling map are as follows: The fault propagation path and risk behavior warning area are aligned in time and space and fused with data to generate a fused risk dataset. Perform virtual-real space risk coupling identification on the fused risk dataset and output the risk coupling strength distribution; The risk coupling intensity distribution is integrated and mapped in a spatiotemporal dimension to generate a spatiotemporal risk coupling map.

10. The digital-twin-based wind turbine safety operation monitoring method of claim 1, wherein: The specific steps for mapping early warning levels to the spatiotemporal risk coupling map and outputting dynamic early warning commands are as follows: The spatiotemporal risk coupling map and the preset early warning rule table are matched for early warning levels, and early warning level classification data is output. The warning level classification data is mapped and formatted for instruction encapsulation, and dynamic warning instructions are output.

Images