Wind power plant safety risk early warning method and device based on multi-source data fusion

Abstract

The invention discloses a multi-source data fusion wind power plant safety risk early warning method and device, relates to the technical field of electric power operation and maintenance, and mainly aims to solve the problem of low accuracy of existing wind power plant safety risk early warning. The method mainly comprises the following steps: acquiring meteorological data, fan operation data, geological data and historical operation and maintenance data of a target wind power plant; carrying out feature extraction to obtain meteorological key features, operation key features and geological key features, and carrying out feature engineering processing on historical operation and maintenance data to obtain a quantitative influence coefficient of maintenance frequency; performing feature fusion processing based on a self-attention mechanism on the quantitative influence coefficients of the meteorological key features, the operation key features, the geological key features and the maintenance frequency through a multi-source feature fusion model to obtain multi-source data fusion features; and performing prediction processing on the multi-source data fusion features by using the safety risk prediction model, and generating safety risk early warning information of the target wind power plant. The method is mainly used for wind power plant safety risk early warning.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology and can be applied to the field of power operation and maintenance, and in particular to a wind farm safety risk early warning method and device based on multi-source data fusion. Background Art

[0002] With the rapid development of the wind power industry and the continuous expansion of wind farms, their operating environments are becoming increasingly complex and volatile, leading to increasingly prominent safety risks. Wind farm safety risk prediction can identify potential safety hazards in advance, allowing appropriate preventive measures to be taken to ensure the safety of wind farm equipment and personnel. Therefore, timely and accurate wind farm safety monitoring has become a key research topic for power operations and maintenance personnel.

[0003] Traditional early warning methods rely heavily on meteorological monitoring data. For example, weather stations collect data such as wind speed, wind direction, temperature, and air pressure. These data are then used to predict wind turbine output power using physical or statistical models, thereby predicting safety risks. Alternatively, they rely on local device sensor data. For example, operating data is acquired by installing sensors on key wind turbine components (such as gearboxes, bearings, and generators). When this data exceeds a preset threshold, an alarm system is triggered. However, these methods suffer from a single data dimension, poor model adaptability, and delayed response. Consequently, their prediction accuracy is low in complex environments, making them unable to meet the safety risk prediction needs of actual wind farms. Summary of the Invention

[0004] In view of this, the present invention provides a wind farm safety risk warning method and device with multi-source data fusion. The main purpose is to solve the problem that the existing wind farm warning method has a single data dimension, resulting in low accuracy of safety risk warning in complex environments, low prediction accuracy in complex environments, and cannot meet the safety risk prediction needs in actual wind farm environments.

[0005] According to one aspect of the present invention, a wind farm safety risk early warning method based on multi-source data fusion is provided, comprising:

[0006] Obtain meteorological data, wind turbine operation data, geological data, and historical operation and maintenance data of the target wind farm;

[0007] Performing feature extraction on the meteorological data, the wind turbine operation data, and the geological data to obtain meteorological key features, operation key features, and geological key features, and performing feature engineering processing on the historical operation and maintenance data to obtain a quantitative impact coefficient of maintenance frequency;

[0008] The multi-source feature fusion model is used to perform feature fusion processing based on a self-attention mechanism on the meteorological key features, the operation key features, the geological key features, and the quantitative influence coefficient of the maintenance frequency to obtain multi-source data fusion features;

[0009] The multi-source data fusion features are predicted and processed using a safety risk prediction model, and safety risk warning information of the target wind farm is generated based on the prediction processing results.

[0010] Furthermore, the geological data includes ground tilt data, groundwater level data, and soil pressure data. The process of extracting features from the geological data includes:

[0011] performing data cleaning and normalization processing on the ground tilt angle data, the groundwater level data, and the soil pressure data, respectively, to obtain preprocessed ground tilt angle data, preprocessed groundwater level data, and preprocessed soil pressure data;

[0012] performing feature extraction on the preprocessed ground tilt angle data using a long short-term memory neural network, performing feature extraction on the preprocessed groundwater level data using a wavelet transform, and performing feature extraction on the preprocessed soil pressure data using an independent component analysis algorithm;

[0013] The feature selection algorithm is used to select the key features that affect the stability of the wind turbine tower from the extracted features as geological key features;

[0014] Among them, the key geological characteristics include at least one of the ground tilt angle change rate, cumulative tilt, groundwater level change amplitude, fluctuation frequency, soil pressure peak, and pressure gradient.

[0015] Furthermore, the feature engineering processing of the historical operation and maintenance data to obtain the quantitative impact coefficient of the maintenance frequency includes:

[0016] Determine the support and confidence between each field in the historical operation and maintenance data by using the Apriori algorithm, and determine the frequent maintenance items based on the support and confidence;

[0017] Correlation modeling is performed on the maintenance frequency item to obtain a linear regression function. By solving the linear regression function, a quantitative influence coefficient of the maintenance frequency is obtained, wherein the quantitative influence coefficient of the maintenance frequency is used to characterize the influence coefficient of each influencing factor on the maintenance frequency, thereby obtaining the quantitative influence coefficient of the maintenance frequency.

[0018] Furthermore, the process of extracting features from the meteorological data to obtain key meteorological features includes:

[0019] Extracting spatial features of meteorological parameters of the meteorological data through the spatiotemporal feature extraction network, and calculating statistical features of meteorological parameters of the meteorological data using a statistical algorithm;

[0020] Using a mutual information algorithm to screen the spatial characteristics of the meteorological parameters and the statistical characteristics of the meteorological parameters to obtain key meteorological characteristics;

[0021] The process of extracting features from the wind turbine operating data to obtain key operating features includes:

[0022] Extracting time series features from the wind turbine operating data using the time series feature extraction network to obtain time series features of the operating data, and processing vibration data in the wind turbine operating data using fast Fourier transform to obtain frequency features;

[0023] The time series features of the operation data are screened using a recursive feature elimination algorithm, and the screened time series features and the frequency features are used together as key operation features.

[0024] Furthermore, the multi-source feature fusion model is used to perform feature fusion processing based on the self-attention mechanism on the meteorological key features, the operation key features, the geological key features, and the quantitative influence coefficient of the maintenance frequency to obtain multi-source data fusion features, including:

[0025] Performing feature splicing processing on different key features and quantitative influence coefficients through the fully connected layer in the multi-source feature fusion model to obtain combined features;

[0026] Based on the self-attention mechanism, the correlation degree corresponding to different combination features is identified, and corresponding attention weights are assigned according to different correlation degrees to obtain multi-source data fusion features matching different attention weights;

[0027] Among them, the multi-source data fusion features include the fusion features of at least two combinations of meteorological key features, operational key features, geological key features and quantitative influence coefficients of maintenance frequency.

[0028] Furthermore, the security risk prediction model includes a classification recognition network and a multi-layer perception network, and the use of the security risk prediction model to predict the multi-source data fusion features includes:

[0029] Classify and identify the multi-source data fusion features through a classification and recognition network to obtain at least one risk type;

[0030] The multi-layer perception network is used to perform multi-layer superimposed nonlinear transformation on the multi-source data fusion features to obtain the expected probability of occurrence of the risk type.

[0031] Furthermore, the safety risk warning information includes warning level and operation and maintenance strategy information. The generating of the safety risk warning information of the target wind farm based on the prediction processing result includes:

[0032] Identifying, from a warning level mapping relationship set, a warning level that matches the risk type and the expected probability of occurrence of the risk type, wherein the warning type includes blade failure, gearbox failure, lightning strike risk, and tower foundation tilt risk;

[0033] For any risk type, operation and maintenance strategy information is generated based on the risk type, the expected probability of occurrence of the risk type, the meteorological data and the historical operation and maintenance data.

[0034] Furthermore, the generating of operation and maintenance strategy information based on the risk type, the expected probability of occurrence of the risk type, the meteorological data and the historical operation and maintenance data includes:

[0035] If the expected probability of occurrence is greater than or equal to the preset probability threshold, and any parameter in the meteorological data satisfies the corresponding operation and maintenance constraint condition, then a shutdown inspection operation and maintenance strategy matching the risk type is generated;

[0036] If the expected probability of occurrence is less than the preset probability threshold, or any parameter in the meteorological data does not meet the corresponding operation and maintenance constraint condition, then the operation power adjustment coefficient is calculated based on the meteorological data and the meteorological data weight, the historical operation and maintenance data and the operation and maintenance data weight, and a power adjustment operation and maintenance strategy is generated based on the operation power adjustment coefficient;

[0037] Matching target operation and maintenance data of the risk type from the historical operation and maintenance data, and generating a spare parts preparation strategy matching the risk type if the target operation and maintenance data represents a component replacement operation;

[0038] If no spare parts preparation strategy is generated, generating single operation and maintenance strategy information according to the shutdown inspection operation and maintenance strategy or the power adjustment operation and maintenance strategy;

[0039] When the shutdown inspection operation and maintenance strategy and the spare parts preparation strategy are generated, the execution priority corresponding to each strategy is retrieved, and combined operation and maintenance strategy information is generated based on the shutdown inspection operation and maintenance strategy, the spare parts preparation strategy, and the execution priority, wherein the execution priority is used to characterize the priority of execution of each strategy.

[0040] According to another aspect of the present invention, a wind farm safety risk prediction device based on multi-source data fusion is provided, comprising:

[0041] Acquisition module, used to obtain meteorological data, wind turbine operation data, geological data and historical operation and maintenance data of the target wind farm;

[0042] a feature extraction module for performing feature extraction on the meteorological data, the wind turbine operation data, and the geological data, respectively, to obtain meteorological key features, operation key features, and geological key features, and performing feature engineering processing on the historical operation and maintenance data to obtain a quantitative influence coefficient of maintenance frequency;

[0043] A feature fusion module is used to perform feature fusion processing based on a self-attention mechanism on the meteorological key features, the operation key features, the geological key features, and the quantitative influence coefficient of the maintenance frequency through a multi-source feature fusion model to obtain multi-source data fusion features;

[0044] The early warning module is used to use the safety risk prediction model to predict the multi-source data fusion features and generate safety risk early warning information of the target wind farm based on the prediction processing results.

[0045] According to another aspect of the present invention, there is provided a terminal comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0046] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the wind farm safety risk early warning method based on multi-source data fusion.

[0047] By means of the above technical solution, the technical solution provided by the embodiment of the present invention has at least the following advantages:

[0048] The present invention provides a wind farm safety risk early warning method and device based on multi-source data fusion. The embodiment of the present invention obtains meteorological data, wind turbine operation data, geological data and historical operation and maintenance data of the target wind farm; extracts features of the meteorological data, wind turbine operation data and geological data respectively to obtain meteorological key features, operation key features and geological key features, and performs feature engineering on the historical operation and maintenance data to obtain the quantitative influence coefficient of maintenance frequency; and performs feature engineering on the meteorological key features, operation key features, geological key features and the quantitative influence coefficient of maintenance frequency through a multi-source feature fusion model. Based on the feature fusion processing of the self-attention mechanism, the multi-source data fusion feature is obtained; the multi-source data fusion feature is predicted and processed using the safety risk prediction model, and the safety risk warning information of the target wind farm is generated according to the prediction processing result, which greatly improves the data source dimension of the wind farm risk warning. Based on feature extraction and feature fusion, the intrinsic correlation between the impact of different data sources on the safety of the wind farm is fully explored, ensuring the comprehensiveness of the data. At the same time, through the safety risk prediction model, the multi-source data fusion feature can be used to predict risks based on the correlation between different data sources, ensuring the accuracy of prediction and warning.

[0049] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, 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 invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0051] Figure 1 A flow chart of a wind farm safety risk early warning method based on multi-source data fusion provided by an embodiment of the present invention is shown;

[0052] Figure 2 A flow chart of another wind farm safety risk early warning method using multi-source data fusion provided by an embodiment of the present invention is shown;

[0053] Figure 3 The following is a block diagram showing the composition of a wind farm safety risk prediction device based on multi-source data fusion provided by an embodiment of the present invention;

[0054] Figure 4 A schematic structural diagram of a terminal provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0055] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0056] The existing wind farm early warning method has a single data dimension, which leads to low accuracy of safety risk warning in complex environments. The embodiment of the present invention provides a wind farm safety risk early warning method based on multi-source data fusion, such as Figure 1 As shown, the method includes:

[0057] 101. Obtain meteorological data, wind turbine operation data, geological data and historical operation and maintenance data of the target wind farm.

[0058] In this embodiment of the present invention, by collecting multi-dimensional data from a target wind farm, including meteorological data, wind turbine operation data, geological data, and historical operation and maintenance data, the inherent relationships among these data dimensions are mined. Based on these relationships, accurate early warnings of safety risks are then provided. The target wind farm serves as the target for wind farm safety risk warnings. Meteorological data includes three-dimensional wind speed vectors, turbulence intensity, atmospheric thermodynamic parameters, and extreme weather warning data. Wind turbine operation data includes mechanical vibration data, structural stress data, fluid system status data, and electrical performance data. Geological data includes ground tilt data, groundwater level data, soil pressure data, soil pressure on the tower foundation, pressure distribution, and dynamic changes. Historical operation and maintenance data includes maintenance time, maintenance personnel, maintenance details, replaced component models, and service life. It should be noted that the operating status of a wind turbine is influenced by a variety of factors. For example, an increase in the vibration amplitude of a wind turbine blade could indicate a fault such as a crack in the blade itself, but it could also be due to a sudden strong wind impact. If the judgment is based solely on vibration amplitude, a blade failure alarm may be erroneously issued. By collecting data from multiple dimensions, including meteorological, geological, equipment operation, and historical maintenance data, the target wind farm can accurately analyze the combined impact of various factors on wind turbine tower stability in complex environments, providing comprehensive data support for subsequent early warning and forecasting. Multi-dimensional data can also include environmental load data, such as extreme loads and environmental corrosion.

[0059] In the actual application environment of a wind farm, various data collection methods can be implemented as follows. Meteorological data can be collected using multiple high-precision meteorological sensors deployed within the target wind farm, such as wind speed sensors, wind direction sensors, temperature sensors, air pressure sensors, and lightning monitors. These sensors can collect real-time meteorological data from various locations within the wind farm. Wind turbine operating data can be collected using sensors installed in key locations of each wind turbine (such as the gearbox, bearings, generator, and blade roots). For example, vibration sensors and oil temperature sensors can be installed in the gearbox to collect gear operating status and lubrication data; high-precision displacement sensors and temperature sensors can be installed in the bearings to collect bearing wear and heat data; power sensors, current sensors, and voltage sensors can be installed in the generator to collect electrical performance data; and strain sensors can be installed at the blade roots to collect blade stress data. Geological data can be collected using inclinometers, groundwater level sensors, and soil pressure sensors deployed in different areas of the target wind farm, such as around the wind turbine tower foundation and in geologically sensitive areas. Specifically, the ground inclination angle, inclination change rate and other data are collected through the inclinometer to serve as the data basis for judging whether the tower foundation has uneven settlement or tilt. The groundwater level height, water level change amplitude and rate are collected through the groundwater level sensor to serve as the data basis for evaluating the impact of groundwater on the stability of the tower foundation. The soil pressure value, pressure distribution and dynamic changes of the soil on the tower foundation are collected through the soil pressure sensor to serve as the data basis for identifying abnormal fluctuations in the soil bearing capacity. Historical operation and maintenance data can be obtained from the database of the wind farm equipment maintenance management system. After the above data are collected, they can be sent to the data processing center of the wind farm in real time through wired (optical fiber, RS-485 bus, etc.) or wireless (4G, LoRa, etc.) transmission methods to complete data processing processes such as feature extraction, feature fusion and wind direction warning based on the data processing center.

[0060] 102. Perform feature extraction on the meteorological data, the wind turbine operation data, and the geological data to obtain meteorological key features, operation key features, and geological key features, and perform feature engineering processing on the historical operation and maintenance data to obtain a quantitative impact coefficient of maintenance frequency.

[0061] In the implementation of this invention, to explore and analyze the impact of various dimensional data on wind turbine tower stability, feature extraction and feature screening are performed on each dimension of data separately to obtain key features corresponding to the different dimensional data. Meteorological key features are extracted from meteorological data, operational key features are extracted from wind turbine operation data, and geological key features are extracted from geological data. Historical operation and maintenance data is processed through feature engineering to convert it into numerical indicators, namely, the quantified impact coefficient of maintenance frequency. Key features are those that significantly affect tower stability. For example, the rate of change of ground tilt angle can reflect the tower tilt trend, and the cumulative tilt can determine the stability of the tower foundation. The amplitude and frequency of groundwater level fluctuations are key to assessing seasonal or sudden factors. The peak value and pressure gradient of soil pressure are key to identifying local stress concentrations. Features closely related to safety risks are extracted and screened from multi-source data. For example, statistical features of wind speed are extracted from meteorological data, and vibration spectrum features are extracted from wind turbine operation data. By removing redundant features, the data dimension is reduced and the efficiency and accuracy of the subsequent multi-source feature fusion model are improved.

[0062] It should be noted that for different data items within the same dimension, corresponding feature extraction algorithms can also be used based on the distribution characteristics of the different data. For example, for tilt data, a time series analysis algorithm (such as the ARIMA model) can be used to establish a baseline tilt curve and calculate the real-time tilt rate (° / h) through a sliding window. Finite element models can also be combined to invert the foundation settlement distribution and identify areas of uneven settlement (accuracy ±2mm). For groundwater level change data, a hydrogeological coupling model can be constructed, and the periodic (tide / precipitation) and trend (aquifer changes) components can be separated through Fourier transform to achieve feature extraction. For soil stress data, three-dimensional stress data can be processed based on tensor decomposition technology to identify changes in the principal stress direction. Digital image correlation can be used to verify the soil strain field and construct the pressure-strain-displacement transfer function.

[0063] 103. Perform feature fusion processing based on a self-attention mechanism on the meteorological key features, the operational key features, the geological key features, and the quantitative influence coefficient of the maintenance frequency through a multi-source feature fusion model to obtain multi-source data fusion features.

[0064] In an embodiment of the present invention, the multi-source feature fusion model is a neural network embedded with a self-attention mechanism. In the multi-source feature fusion model, feature splicing is performed on features associated between different dimensions, such as the combined feature of high wind speed (key meteorological feature) and blade strain anomaly (key operation feature), and the combined feature of tilt change rate (geological feature) and tower vibration acceleration (key operation feature). And through self-attention learning, the degree of influence of different combined features on the stability of the wind turbine tower or the operating status of the wind turbine is identified, and higher attention weights are assigned to features with greater influence to highlight key risk features. Through feature fusion, the standardization of multi-dimensional features can be achieved to obtain standardized feature vectors, and at the same time, features that are strongly related to the stability of the wind turbine tower and the joint influence of different features on the stability of the wind turbine tower can be effectively mined.

[0065] 104. Use a safety risk prediction model to perform prediction processing on the multi-source data fusion features, and generate safety risk warning information of the target wind farm based on the prediction processing results.

[0066] In an embodiment of the present invention, the safety risk prediction model is a pre-trained model, trained using fused feature samples constructed from a large amount of multi-source historical data. The safety risk prediction model establishes a mapping relationship between geological factors and tower stability based on machine learning algorithms (such as random forests and LSTM neural networks). For example, when the local tilt change rate exceeds a threshold and the groundwater level rises significantly, the tower foundation is predicted to be at risk of tilting. When the soil pressure peak remains consistently above the design value and is accompanied by strong wind loads (meteorological data), fatigue damage to the foundation structure is predicted to be possible. In addition to the risk type, the model output may also include the expected probability of occurrence corresponding to different risk types. Based on the different risk types and their corresponding expected probability of occurrence, a warning level can be further determined based on a preset warning threshold. For example, if the risk type is tower foundation tilt risk and the probability of tilting in the next seven days is 45%, which is greater than the warning threshold of 30% corresponding to the second-level warning, a second-level safety risk warning message is generated. Of course, warnings can also be issued based on individual feature data. For example, when the local tilt change rate is >0.1° / day or the groundwater level rise rate is >5cm / hour, different levels of warnings can be triggered accordingly.

[0067] By integrating multi-source data, a prediction model is constructed that comprehensively considers the interaction of multiple factors. This can accurately predict various types of safety risks in wind farms in different time periods in the future, such as the risk of wind turbine blade breakage, tower tilting, lightning strike risk, etc. It can not only predict the probability and time range of occurrence of a single risk event, but also comprehensively evaluate the complex safety situation under the interaction of multiple risks, thereby improving the accuracy of safety risk warnings.

[0068] In one embodiment of the present invention, for further explanation and limitation, as Figure 2 As shown in Figure 1, the process of feature extraction for geological data includes:

[0069] 201. Perform data cleaning and normalization processing on the ground tilt angle data, the groundwater level data, and the soil pressure data, respectively, to obtain preprocessed ground tilt angle data, preprocessed groundwater level data, and preprocessed soil pressure data;

[0070] 202. Use a long short-term memory neural network to perform feature extraction on the preprocessed ground tilt angle data, use a wavelet transform to perform feature extraction on the preprocessed groundwater level data, and use an independent component analysis algorithm to perform feature extraction on the preprocessed soil pressure data.

[0071] 203. The key features that affect the stability of wind turbine towers are selected from the extracted features using feature selection algorithm as geological key features.

[0072] In an embodiment of the present invention, to ensure the reliability of geological data, preprocessing is required before feature extraction. This preprocessing process includes data cleaning and normalization. Data cleaning can remove outliers, such as sudden changes in data caused by sensor failure, and can also fill missing data using methods such as mean filling and time series interpolation. Furthermore, data of different dimensions, such as tilt angle in degrees, groundwater level in meters, and soil pressure in kilopascals, are normalized and uniformly converted to the range [0, 1] or [-1, 1] to eliminate the impact of dimensional differences on the feature extraction process. After preprocessing, feature extraction is performed on the preprocessed ground tilt angle data using a long short-term memory neural network to extract the temporal variation of tilt angle. Feature extraction is performed on the preprocessed groundwater level data using a wavelet transform. For the non-stationary water level data, the wavelet transform is used to extract time-frequency features that can reflect changes at different time scales. An independent component analysis algorithm is used to extract features from the preprocessed soil pressure data. Eigenvalue decomposition is used to convert the signal covariance matrix into a unit matrix. The separation matrix is ​​obtained by assuming that the mixed signal is a linear combination of multiple independent source signals. The separation matrix is ​​then solved using optimization algorithms such as maximum likelihood estimation and the FastICA algorithm. Taking the FastICA algorithm as an example, the row vectors of the separation matrix are iteratively updated, and nonlinear functions (such as the negentropy maximization criterion) are continuously optimized until convergence conditions are met. Fault-related spectral features, such as frequency content and energy distribution, are screened from the separated independent components. Furthermore, a feature selection algorithm is used to screen for features that significantly impact tower stability, removing redundant features such as long-term stable static pressure values ​​to obtain key geological features, thereby reducing feature dimensionality. Key geological features include one or more of the following: rate of change of ground tilt angle, cumulative tilt, groundwater level change amplitude, fluctuation frequency, soil pressure peak, and pressure gradient. Feature selection algorithms can include mutual information method, recursive feature elimination, and other methods, which are not specifically limited in this embodiment of the present invention. It should be noted that the above data preprocessing process can also be applied to the processing of meteorological data and wind turbine operation data, which will not be elaborated in the embodiment of the present invention.

[0073] In one embodiment of the present invention, for further explanation and limitation, the feature engineering processing of the historical operation and maintenance data to obtain the quantitative influence coefficient of the maintenance frequency includes:

[0074] Determine the support and confidence between each field in the historical operation and maintenance data by using the Apriori algorithm, and determine the frequent maintenance items based on the support and confidence;

[0075] Correlation modeling is performed based on the maintenance frequency item to obtain a linear regression function, and a quantitative influence coefficient of the maintenance frequency is obtained by solving the linear regression function.

[0076] In this embodiment of the present invention, the support and confidence of association rules of fields such as maintenance content, component model, and service life in historical operation and maintenance data are calculated based on the Apriori algorithm. Taking the analysis of the association between "replacing the gearbox" and "gearbox service life < design life" as an example, the support calculation formula is: Among them, S represents support, X represents "replace the gearbox", Y represents "gearbox service life < design life (life is insufficient)", M(X∪Y) represents the number of transactions containing X and Y, M 总 Represents the total number of transactions. The confidence calculation formula is Where C represents the confidence level, M(X∪Y) represents the number of transactions containing both X and Y, and M(X) represents the number of transactions containing "gearbox replacement." The support level represents the frequency of "gearbox replacement with insufficient lifespan," while the confidence level represents the probability of "inadequate lifespan" when "gearbox replacement" is performed. These factors are used to help identify the correlation between component quality and maintenance. After obtaining the support and confidence levels for different fields, fields with both support and confidence levels exceeding the corresponding preset thresholds are identified as frequent maintenance items. Correlation modeling is performed for each frequent maintenance item to analyze the factors influencing maintenance frequency. The quantitative influence coefficient of maintenance frequency is used to characterize the influence of various influencing factors on maintenance frequency, resulting in the quantitative influence coefficient of maintenance frequency. For example, taking the component maintenance frequency y as the dependent variable and the component lifespan x1 and the operating and maintenance environment and meteorological conditions x2 as independent variables, a linear regression function can be constructed as follows: y = β0 + β1 × x1 + β2 × x2 + ε. β0, β1, and β2 are the quantitative influence coefficients of maintenance frequency, and ε is a constant that can be customized based on actual scenario requirements.

[0077] In one embodiment of the present invention, for further explanation and limitation, the process of extracting features from the meteorological data to obtain key meteorological features includes:

[0078] Extracting spatial features of meteorological parameters of the meteorological data through the spatiotemporal feature extraction network, and calculating statistical features of meteorological parameters of the meteorological data using a statistical algorithm;

[0079] Using a mutual information algorithm to screen the spatial characteristics of the meteorological parameters and the statistical characteristics of the meteorological parameters to obtain key meteorological characteristics;

[0080] The process of extracting features from the wind turbine operating data to obtain key operating features includes:

[0081] Extracting time series features from the wind turbine operating data using the time series feature extraction network to obtain time series features of the operating data, and processing vibration data in the wind turbine operating data using fast Fourier transform to obtain frequency features;

[0082] The time series features of the operation data are screened using a recursive feature elimination algorithm, and the screened time series features and the frequency features are used together as key operation features.

[0083] In this embodiment of the present invention, meteorological data processing is primarily based on a convolutional neural network. This uses convolutional kernels to extract spatial features such as wind field distribution and lightning activity patterns from time-series spatial data collected by meteorological sensors, such as lightning activity data and wind speed data. Statistical methods are also used to calculate the statistical characteristics of meteorological parameters such as mean wind speed and lightning frequency within the meteorological data. Furthermore, a mutual information algorithm is used to filter these features, identifying meteorological features that are strongly correlated with wind turbine operating status and tower stability, known as key meteorological features.

[0084] The processing of wind turbine operating data is mainly based on long short-term memory networks and fast Fourier transforms. For time series data such as gearbox vibration spectrum and blade strain, LSTM uses memory units to capture the long-term dependencies of equipment operating status, such as the development trend of vibration anomalies. For time domain signal data such as vibration of wind turbine components (gearbox, blades), in addition to extracting time series features, the time domain signal is converted to the frequency domain based on the fast Fourier transform to extract the implicit frequency components of the signal. Specifically, the obtained discrete signal sequence that changes with time is used to calculate the maximum value of the DFT calculation amount through the discrete Fourier transform, and then the discrete sequence is continuously decomposed into shorter subsequences for calculation based on the time extraction method. Then, the frequency domain sequence is gradually calculated through the butterfly operation unit to obtain the amplitude and phase information of the signal at different frequency points, realizing the conversion of time domain signals to frequency domain signals.

[0085] In one embodiment of the present invention, for further explanation and limitation, the multi-source feature fusion model is used to perform feature fusion processing based on a self-attention mechanism on the meteorological key features, the operational key features, the geological key features, and the quantitative influence coefficient of the maintenance frequency to obtain multi-source data fusion features, including:

[0086] Performing feature splicing processing on different key features and quantitative influence coefficients through the fully connected layer in the multi-source feature fusion model to obtain combined features;

[0087] Based on the self-attention mechanism, the correlation degree corresponding to different combination features is identified, and corresponding attention weights are assigned according to different correlation degrees to obtain multi-source data fusion features matching different attention weights;

[0088] In an embodiment of the present invention, a deep learning network based on an attention mechanism learns the correlation between features from different sources such as meteorology, operation, and geology, as well as the correlation between different combination features and the safety risk of wind turbine operation, and distributes attention weights based on the learned correlation to highlight key risk features. Finally, different types of features are spliced ​​together through a fully connected layer to output a fused feature vector, i.e., a multi-source data fusion feature. Among them, the multi-source data fusion feature includes a fusion feature of at least two combinations of meteorological key features, operational key features, geological key features, and the quantitative influence coefficient of maintenance frequency. For example, the combination feature of "high wind speed + blade strain anomaly", the combination feature of "ground tilt change rate + groundwater level change amplitude + soil pressure peak + wind turbine vibration spectrum feature", etc.

[0089] In one embodiment of the present invention, for further explanation and limitation, the predicting process of the multi-source data fusion feature using the security risk prediction model includes:

[0090] Classify and identify the multi-source data fusion features through a classification and recognition network to obtain at least one risk type;

[0091] The multi-layer perception network is used to perform multi-layer superimposed nonlinear transformation on the multi-source data fusion features to obtain the expected probability of occurrence of the risk type.

[0092] In an embodiment of the present invention, the security risk prediction model includes a classification recognition network and a multi-layer perception network. Among them, the classification recognition model can be a model constructed based on algorithms such as random forest and support vector machine (SVM). By performing classification prediction on the multi-source data fusion features, at least one risk type is obtained. The multi-layer perception network includes an input layer, multiple hidden layers and an output layer. The hidden layer is a nonlinear activation function, and the output layer is Softmax. By performing multi-layer nonlinear transformation on the multi-source data fusion features, high-order features are extracted, and the expected probability of occurrence of different risk types is calculated through the function of the output layer to complete the final risk prediction.

[0093] In one embodiment of the present invention, for further explanation and limitation, generating the safety risk warning information of the target wind farm based on the prediction processing result includes:

[0094] Identifying, from a warning level mapping relationship set, a warning level that matches the risk type and the expected probability of occurrence of the risk type;

[0095] For any risk type, operation and maintenance strategy information is generated based on the risk type, the expected probability of occurrence of the risk type, the meteorological data and the historical operation and maintenance data.

[0096] In this embodiment of the present invention, safety risk warning information includes warning levels and operation and maintenance strategy information. Warning types include blade failure, gearbox failure, lightning strike risk, and tower foundation tilt risk. The warning level mapping relationship set includes pre-established warning levels corresponding to different risk types and different expected probability of occurrence. For example, if the predicted probability of a wind turbine blade failure within the next week exceeds 30%, a medium-level warning is issued; if the predicted probability of a wind turbine blade failure within the next week exceeds 50%, a high-level warning is issued. Warning information is sent to wind farm operation and maintenance personnel through various means, such as SMS notifications, pop-up windows on the monitoring system, and audio and visual alarms, ensuring that operators are promptly aware of potential safety risks. Safety risk warning information also includes recommendations for appropriate actions. For example, shutting down the machine to inspect blades during low wind speeds and preparing parts and tools for blade replacement in advance are recommended. Furthermore, the system can also make adjustments to the wind farm's overall operation strategy based on historical data and current risk conditions, such as appropriately reducing wind turbine operating power during periods of high lightning activity to reduce lightning strike risk. After outputting the O&M policy information, the results of each policy execution can be recorded, such as whether faults were avoided and the improvement in O&M efficiency, to update the policy rule base. For example, if a power reduction policy during a lightning period reduces the fault rate, the priority of this policy in similar scenarios can be increased.

[0097] In one embodiment of the present invention, for further explanation and limitation, generating the operation and maintenance strategy information based on the risk type, the expected probability of occurrence of the risk type, the meteorological data, and the historical operation and maintenance data includes:

[0098] If the expected probability of occurrence is greater than or equal to the preset probability threshold, and any parameter in the meteorological data meets the corresponding operation and maintenance constraint conditions, a shutdown inspection operation and maintenance strategy matching the risk type is generated; if the expected probability of occurrence is less than the preset probability threshold, or any parameter in the meteorological data does not meet the corresponding operation and maintenance constraint conditions, an operating power adjustment coefficient is calculated based on the meteorological data and meteorological data weight, the historical operation and maintenance data and operation and maintenance data weight, and a power adjustment operation and maintenance strategy is generated based on the operating power adjustment coefficient; target operation and maintenance data of the risk type is matched from the historical operation and maintenance data, and if the target operation and maintenance data represents a component replacement operation, a spare parts preparation strategy matching the risk type is generated; if no spare parts preparation strategy is generated, a single operation and maintenance strategy information is generated based on the shutdown inspection operation and maintenance strategy or the power adjustment operation and maintenance strategy; if the shutdown inspection operation and maintenance strategy and the spare parts preparation strategy are generated, the execution priority corresponding to each strategy is retrieved, and combined operation and maintenance strategy information is generated based on the shutdown inspection operation and maintenance strategy, the spare parts preparation strategy, and the execution priority.

[0099] In an embodiment of the present invention, operation and maintenance policy information includes single operation and maintenance policy information and combined operation and maintenance policy information. When only a shutdown inspection operation and maintenance policy or a power adjustment operation and maintenance policy is generated, the corresponding single operation and maintenance policy information is generated based on the shutdown inspection operation and maintenance policy or the power adjustment operation and maintenance policy. This means that both the shutdown inspection operation and maintenance policy are generated, and combined operation and maintenance policy information is generated based on these two policies. The combined operation and maintenance policy information carries an execution priority that indicates the priority for executing each policy. Specifically, when multiple rules are triggered, a conflict resolution strategy is employed, with the rule with the higher priority being executed first. For example, if the "Shutdown Inspection" and "Parts Preparation" rules are triggered simultaneously, the combined operation and maintenance policy information is combined according to the principle of "prepare resources first, then execute shutdown." Operation and maintenance constraints primarily consist of meteorological data, including wind speed and lightning index. Inspection actions are executed only when the operation and maintenance constraints are met; otherwise, the inspection actions are not executed. If operating power adjustment is required, an operating power adjustment coefficient is calculated based on meteorological data, historical operating data, and their corresponding weights. Specifically, the operating power adjustment coefficient can be calculated by multiplying the product of different meteorological data items with their corresponding weights, and the product of different operation and maintenance data items with their corresponding weights. Different data items have different weights, for example, a weight of 0.6 for periods with high lightning incidence, and a weight of 0.4 for the historical frequency of blade failures. These weights can be customized based on actual application requirements. After obtaining the operating power adjustment coefficient, the product of the operating power adjustment coefficient and the operating power can be used as the power adjustment amount. Alternatively, the corresponding power adjustment amount can be matched based on the range of the operating power adjustment coefficient using a pre-established mapping relationship.

[0100] In an application example, the current execution entity obtains the risk type (such as blade failure) and expected probability value in the safety risk warning information in real time, and simultaneously retrieves meteorological data (wind speed, lightning index) and historical operation and maintenance data (similar fault processing records). The data is structured: for example, the "wind speed value" is compared with the "blade can withstand safe wind speed threshold" to mark whether it is in the "low wind speed safety range"; the "high lightning incidence period" is marked as a Boolean value (yes / no). Using rule engine technology (such as Drools), preset policy trigger conditions and corresponding operations. Operations include:

[0101] IF (warning type = blade failure, and wind speed < safety threshold) THEN generate the "stop and check blades" strategy;

[0102] IF (warning type = blade failure, and historical data shows similar failures requiring component replacement) THEN generate the "prepare blade component and tool replacement in advance" strategy.

[0103] In another specific example, meteorological data is used to analyze the aerodynamic loads exerted by extreme wind speeds on blades based on wind speed and direction data, emphasizing the potential for structural fatigue damage caused by overloaded operation. Lightning monitoring data is used to assess the frequency of regional lightning activity and assess the risk of blade damage from lightning strikes. Wind turbine operation data is used to analyze abnormal blade oscillation or resonance using vibration frequency and amplitude data. Blade strain data is used to determine the strain value of the blade material and determine potential structural damage such as deformation or cracking due to long-term stress. Operating time data is used to assess the degree of aging wear by accumulating blade operating time and combining it with material fatigue characteristics. Historical maintenance data is used to analyze historical failure types and repair outcomes using historical maintenance records, such as blade repairs and component replacements. Component lifespan data, such as the difference between the current operating life of a blade and its design life, is used to identify whether a blade is approaching a peak failure period due to aging. Geological data is used to identify key geological features, including soil subsidence and ground tilt data, which influence the overall stability of the wind turbine and indirectly increase blade operating loads. After feature extraction is completed, the features are fused through a multi-source feature fusion model and then input into the security risk prediction model, which ultimately outputs the prediction results of the risk type and the expected probability of occurrence of the risk type.

[0104] The present invention provides a wind farm safety risk warning method based on multi-source data fusion. The embodiment of the present invention obtains meteorological data, wind turbine operation data, geological data and historical operation and maintenance data of a target wind farm; performs feature extraction on the meteorological data, wind turbine operation data and geological data respectively to obtain meteorological key features, operation key features and geological key features, and performs feature engineering processing on the historical operation and maintenance data to obtain a quantitative influence coefficient of maintenance frequency; performs feature fusion processing based on a self-attention mechanism on the meteorological key features, operation key features, geological key features and the quantitative influence coefficient of maintenance frequency through a multi-source feature fusion model to obtain multi-source data fusion features; uses a safety risk prediction model to predict the multi-source data fusion features, and generates safety risk warning information of the target wind farm based on the prediction processing results, thereby greatly improving the data source dimension of the wind farm risk warning. Based on feature extraction and feature fusion, the inherent correlation between the impact of different data sources on the safety of the wind farm is fully explored, thereby ensuring the comprehensiveness of the data. At the same time, the safety risk prediction model is used to fusion features of the multi-source data, so that risks can be predicted based on the correlation between different data sources, thereby ensuring the accuracy of prediction and warning.

[0105] Furthermore, as a response to the above Figure 1 The embodiment of the present invention provides a wind farm safety risk prediction device based on multi-source data fusion, such as Figure 3 As shown, the device includes:

[0106] An acquisition module 31 is used to acquire meteorological data, wind turbine operation data, geological data, and historical operation and maintenance data of a target wind farm;

[0107] A feature extraction module 32 is configured to perform feature extraction on the meteorological data, the wind turbine operation data, and the geological data to obtain meteorological key features, operation key features, and geological key features, and to perform feature engineering processing on the historical operation and maintenance data to obtain a quantitative influence coefficient of maintenance frequency;

[0108] A feature fusion module 33 is configured to perform feature fusion processing based on a self-attention mechanism on the meteorological key features, the operational key features, the geological key features, and the quantitative influence coefficient of the maintenance frequency through a multi-source feature fusion model to obtain multi-source data fusion features;

[0109] The early warning module 34 is configured to perform prediction processing on the multi-source data fusion features using a safety risk prediction model, and generate safety risk early warning information of the target wind farm based on the prediction processing result.

[0110] Furthermore, the feature extraction module includes:

[0111] a preprocessing unit, configured to perform data cleaning and normalization processing on the ground tilt angle data, the groundwater level data, and the soil pressure data, respectively, to obtain preprocessed ground tilt angle data, preprocessed groundwater level data, and preprocessed soil pressure data;

[0112] a first feature extraction unit, configured to extract features from the preprocessed ground tilt angle data using a long short-term memory neural network, extract features from the preprocessed groundwater level data using a wavelet transform, and extract features from the preprocessed soil pressure data using an independent component analysis algorithm;

[0113] The first screening unit is used to screen out key features that affect the stability of the wind turbine tower from the extracted features using a feature selection algorithm as geological key features;

[0114] Among them, the key geological characteristics include at least one of the ground tilt angle change rate, cumulative tilt, groundwater level change amplitude, fluctuation frequency, soil pressure peak, and pressure gradient.

[0115] Furthermore, the feature extraction module includes:

[0116] A determination unit, configured to determine the support and confidence between the various fields in the historical operation and maintenance data using an Apriori algorithm, and determine frequent maintenance items based on the support and confidence;

[0117] A linear regression unit is used to perform correlation modeling based on the maintenance frequency item to obtain a linear regression function, and to obtain a quantitative influence coefficient of the maintenance frequency by solving the linear regression function, wherein the quantitative influence coefficient of the maintenance frequency is used to characterize the influence coefficient of each influencing factor on the maintenance frequency, thereby obtaining the quantitative influence coefficient of the maintenance frequency.

[0118] Furthermore, the feature extraction module includes:

[0119] a computing unit, configured to extract spatial features of meteorological parameters of the meteorological data through the spatiotemporal feature extraction network, and calculate statistical features of meteorological parameters of the meteorological data using a statistical algorithm;

[0120] A second screening unit is configured to screen the meteorological parameter spatial features and the meteorological parameter statistical features using a mutual information algorithm to obtain meteorological key features;

[0121] a second feature extraction unit, configured to extract time series features from the wind turbine operation data using the time series feature extraction network to obtain time series features of the operation data, and process vibration data in the wind turbine operation data using a fast Fourier transform to obtain frequency features;

[0122] The third screening unit is configured to screen the time series features of the operation data using a recursive feature elimination algorithm, and use the screened time series features and the frequency features as key operation features.

[0123] Furthermore, the feature fusion module includes:

[0124] A feature splicing unit, configured to perform feature splicing processing on different key features and quantitative influence coefficients through a fully connected layer in the multi-source feature fusion model to obtain a combined feature;

[0125] The weight allocation unit is used to identify the correlation degree corresponding to different combination features based on the self-attention mechanism, and assign corresponding attention weights according to different correlation degrees to obtain multi-source data fusion features matching different attention weights;

[0126] Among them, the multi-source data fusion features include the fusion features of at least two combinations of meteorological key features, operational key features, geological key features and quantitative influence coefficients of maintenance frequency.

[0127] Furthermore, the early warning module includes:

[0128] an identification unit, configured to identify, from a warning level mapping relationship set, a warning level that matches the risk type and the expected probability of occurrence of the risk type, wherein the warning type includes blade failure, gearbox failure, lightning strike risk, and tower foundation tilt risk;

[0129] The generating unit is configured to generate operation and maintenance strategy information for any risk type according to the risk type, the expected probability of occurrence of the risk type, the meteorological data and the historical operation and maintenance data.

[0130] In a specific application scenario, the generating unit is specifically configured to generate a shutdown inspection and operation and maintenance strategy matching the risk type if the expected probability of occurrence is greater than or equal to a preset probability threshold and any parameter in the meteorological data satisfies the corresponding operation and maintenance constraint condition;

[0131] If the expected probability of occurrence is less than the preset probability threshold, or any parameter in the meteorological data fails to meet the corresponding operation and maintenance constraint condition, the operation power adjustment coefficient is calculated based on the meteorological data and the meteorological data weight, the historical operation and maintenance data and the operation and maintenance data weight, and a power adjustment operation and maintenance strategy is generated based on the operation power adjustment coefficient; the target operation and maintenance data of the risk type is matched from the historical operation and maintenance data, and if the target operation and maintenance data represents a component replacement operation, a spare parts preparation strategy matching the risk type is generated; if no spare parts preparation strategy is generated, a single operation and maintenance strategy information is generated based on the shutdown inspection operation and maintenance strategy or the power adjustment operation and maintenance strategy; if the shutdown inspection operation and maintenance strategy and the spare parts preparation strategy are generated, the execution priority corresponding to each strategy is retrieved, and combined operation and maintenance strategy information is generated based on the shutdown inspection operation and maintenance strategy, the spare parts preparation strategy, and the execution priority, wherein the execution priority is used to represent the priority of execution of each strategy.

[0132] The present invention provides a wind farm safety risk warning device with multi-source data fusion. The embodiment of the present invention obtains meteorological data, wind turbine operation data, geological data and historical operation and maintenance data of a target wind farm; performs feature extraction on the meteorological data, wind turbine operation data and geological data respectively to obtain meteorological key features, operation key features and geological key features, and performs feature engineering processing on the historical operation and maintenance data to obtain a quantitative influence coefficient of maintenance frequency; performs feature fusion processing based on a self-attention mechanism on the meteorological key features, operation key features, geological key features and the quantitative influence coefficient of maintenance frequency through a multi-source feature fusion model to obtain multi-source data fusion features; uses a safety risk prediction model to predict the multi-source data fusion features, and generates safety risk warning information of the target wind farm based on the prediction processing results, thereby greatly improving the data source dimension of the wind farm risk warning. Based on feature extraction and feature fusion, the inherent correlation between the impact of different data sources on the safety of the wind farm is fully explored, thereby ensuring the comprehensiveness of the data. At the same time, the safety risk prediction model is used to fusion features of the multi-source data, so that risks can be predicted based on the correlation between different data sources, thereby ensuring the accuracy of prediction and warning.

[0133] Figure 4 A schematic structural diagram of a terminal provided according to an embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the terminal.

[0134] like Figure 4 As shown, the terminal may include: a processor (processor) 402 , a communications interface (Communications Interface) 404 , a memory (memory) 406 , and a communication bus 408 .

[0135] The processor 402 , the communication interface 404 , and the memory 406 communicate with each other via a communication bus 408 .

[0136] The communication interface 404 is used for network communication with other devices such as clients or other servers.

[0137] The processor 402 is configured to execute the program 410 , and specifically to execute the relevant steps in the embodiment of the wind farm safety risk early warning method based on multi-source data fusion.

[0138] Specifically, the program 410 may include program codes, which include computer operation instructions.

[0139] Processor 402 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included in the terminal may be processors of the same type, such as one or more CPUs, or processors of different types, such as one or more CPUs and one or more ASICs.

[0140] The memory 406 is used to store the program 410. The memory 406 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0141] The program 410 may be specifically configured to cause the processor 402 to perform the following operations:

[0142] Obtain meteorological data, wind turbine operation data, geological data, and historical operation and maintenance data of the target wind farm;

[0143] Performing feature extraction on the meteorological data, the wind turbine operation data, and the geological data to obtain meteorological key features, operation key features, and geological key features, and performing feature engineering processing on the historical operation and maintenance data to obtain a quantitative impact coefficient of maintenance frequency;

[0144] The multi-source feature fusion model is used to perform feature fusion processing based on a self-attention mechanism on the meteorological key features, the operation key features, the geological key features, and the quantitative influence coefficient of the maintenance frequency to obtain multi-source data fusion features;

[0145] The multi-source data fusion features are predicted and processed using a safety risk prediction model, and safety risk warning information of the target wind farm is generated based on the prediction processing results.

[0146] Obviously, those skilled in the art will appreciate that the various modules or steps of the present invention described above can be implemented using a general-purpose computing device, centralized on a single computing device, or distributed across a network of multiple computing devices. Alternatively, they can be implemented using program code executable by a computing device, which can then be stored in a storage device and executed by the computing device. In some cases, the steps shown or described can be performed in a different order than that shown, or can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0147] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A wind farm safety risk early warning method based on multi-source data fusion, characterized in that: include: Obtain meteorological data, wind turbine operation data, geological data, and historical operation and maintenance data of the target wind farm; Performing feature extraction on the meteorological data, the wind turbine operation data, and the geological data to obtain meteorological key features, operation key features, and geological key features, and performing feature engineering processing on the historical operation and maintenance data to obtain a quantitative impact coefficient of maintenance frequency; The multi-source feature fusion model is used to perform feature fusion processing based on a self-attention mechanism on the meteorological key features, the operation key features, the geological key features, and the quantitative influence coefficient of the maintenance frequency to obtain multi-source data fusion features; The multi-source data fusion features are predicted and processed using a safety risk prediction model, and safety risk warning information of the target wind farm is generated based on the prediction processing results.

2. The method according to claim 1, characterized in that The geological data includes ground tilt data, groundwater level data, and soil pressure data. The process of extracting features from the geological data includes: performing data cleaning and normalization processing on the ground tilt angle data, the groundwater level data, and the soil pressure data, respectively, to obtain preprocessed ground tilt angle data, preprocessed groundwater level data, and preprocessed soil pressure data; performing feature extraction on the preprocessed ground tilt angle data using a long short-term memory neural network, performing feature extraction on the preprocessed groundwater level data using a wavelet transform, and performing feature extraction on the preprocessed soil pressure data using an independent component analysis algorithm; The feature selection algorithm is used to select the key features that affect the stability of the wind turbine tower from the extracted features as geological key features; Among them, the key geological characteristics include at least one of the ground tilt angle change rate, cumulative tilt, groundwater level change amplitude, fluctuation frequency, soil pressure peak, and pressure gradient.

3. The method according to claim 2, characterized in that The feature engineering processing of the historical operation and maintenance data to obtain the quantitative impact coefficient of the maintenance frequency includes: Determine the support and confidence between each field in the historical operation and maintenance data by using the Apriori algorithm, and determine the frequent maintenance items based on the support and confidence; Correlation modeling is performed on the maintenance frequency item to obtain a linear regression function. By solving the linear regression function, a quantitative influence coefficient of the maintenance frequency is obtained, wherein the quantitative influence coefficient of the maintenance frequency is used to characterize the influence coefficient of each influencing factor on the maintenance frequency, thereby obtaining the quantitative influence coefficient of the maintenance frequency.

4. The method according to claim 3, characterized in that The process of extracting features from the meteorological data to obtain key meteorological features includes: Extracting spatial features of meteorological parameters of the meteorological data through the spatiotemporal feature extraction network, and calculating statistical features of meteorological parameters of the meteorological data using a statistical algorithm; Using a mutual information algorithm to screen the spatial characteristics of the meteorological parameters and the statistical characteristics of the meteorological parameters to obtain key meteorological characteristics; The process of extracting features from the wind turbine operating data to obtain key operating features includes: Extracting time series features from the wind turbine operating data using the time series feature extraction network to obtain time series features of the operating data, and processing vibration data in the wind turbine operating data using fast Fourier transform to obtain frequency features; The time series features of the operation data are screened using a recursive feature elimination algorithm, and the screened time series features and the frequency features are used together as key operation features.

5. The method according to claim 1, wherein The multi-source feature fusion model is used to perform feature fusion processing based on a self-attention mechanism on the meteorological key features, the operational key features, the geological key features, and the quantitative influence coefficient of the maintenance frequency to obtain multi-source data fusion features, including: Performing feature splicing processing on different key features and quantitative influence coefficients through the fully connected layer in the multi-source feature fusion model to obtain combined features; Based on the self-attention mechanism, the correlation degree corresponding to different combination features is identified, and corresponding attention weights are assigned according to different correlation degrees to obtain multi-source data fusion features matching different attention weights; Among them, the multi-source data fusion features include the fusion features of at least two combinations of meteorological key features, operational key features, geological key features and quantitative influence coefficients of maintenance frequency.

6. The method according to claim 5, characterized in that The security risk prediction model includes a classification recognition network and a multi-layer perception network. The security risk prediction model is used to predict the multi-source data fusion features, including: Classify and identify the multi-source data fusion features through a classification and recognition network to obtain at least one risk type; The multi-layer perception network is used to perform multi-layer superimposed nonlinear transformation on the multi-source data fusion features to obtain the expected probability of occurrence of the risk type.

7. The method according to claim 6, characterized in that The safety risk warning information includes warning level and operation and maintenance strategy information. The safety risk warning information of the target wind farm generated according to the prediction processing result includes: Identifying, from a warning level mapping relationship set, a warning level that matches the risk type and the expected probability of occurrence of the risk type, wherein the warning type includes blade failure, gearbox failure, lightning strike risk, and tower foundation tilt risk; For any risk type, operation and maintenance strategy information is generated based on the risk type, the expected probability of occurrence of the risk type, the meteorological data and the historical operation and maintenance data.

8. The method according to claim 7, characterized in that The generating of operation and maintenance strategy information based on the risk type, the expected probability of occurrence of the risk type, the meteorological data and the historical operation and maintenance data includes: If the expected probability of occurrence is greater than or equal to the preset probability threshold, and any parameter in the meteorological data satisfies the corresponding operation and maintenance constraint condition, then a shutdown inspection operation and maintenance strategy matching the risk type is generated; If the expected probability of occurrence is less than the preset probability threshold, or any parameter in the meteorological data does not meet the corresponding operation and maintenance constraint condition, then the operation power adjustment coefficient is calculated based on the meteorological data and the meteorological data weight, the historical operation and maintenance data and the operation and maintenance data weight, and a power adjustment operation and maintenance strategy is generated based on the operation power adjustment coefficient; Matching target operation and maintenance data of the risk type from the historical operation and maintenance data, and generating a spare parts preparation strategy matching the risk type if the target operation and maintenance data represents a component replacement operation; If no spare parts preparation strategy is generated, generating single operation and maintenance strategy information according to the shutdown inspection operation and maintenance strategy or the power adjustment operation and maintenance strategy; When the shutdown inspection operation and maintenance strategy and the spare parts preparation strategy are generated, the execution priority corresponding to each strategy is retrieved, and combined operation and maintenance strategy information is generated based on the shutdown inspection operation and maintenance strategy, the spare parts preparation strategy, and the execution priority, wherein the execution priority is used to characterize the priority of execution of each strategy.

9. A wind farm safety risk early warning device based on multi-source data fusion, characterized in that: include: Acquisition module, used to obtain meteorological data, wind turbine operation data, geological data and historical operation and maintenance data of the target wind farm; a feature extraction module for performing feature extraction on the meteorological data, the wind turbine operation data, and the geological data, respectively, to obtain meteorological key features, operation key features, and geological key features, and performing feature engineering processing on the historical operation and maintenance data to obtain a quantitative influence coefficient of maintenance frequency; A feature fusion module is used to perform feature fusion processing based on a self-attention mechanism on the meteorological key features, the operation key features, the geological key features, and the quantitative influence coefficient of the maintenance frequency through a multi-source feature fusion model to obtain multi-source data fusion features; The early warning module is used to use the safety risk prediction model to predict the multi-source data fusion features and generate safety risk early warning information of the target wind farm based on the prediction processing results.

10. A terminal comprising: A processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the wind farm safety risk early warning method based on multi-source data fusion according to any one of claims 1 to 8.

Images