Method and device for early warning of failure of main bearing of offshore wind turbine generator

Abstract

The application provides a fault early warning method and device based on a main bearing temperature of an offshore wind turbine generator set, wherein a target historical variable is determined from a plurality of initial historical variables based on a mechanism analysis method; a historical characteristic variable is determined from the target historical variable based on a random forest algorithm; a plurality of historical strong correlation variable value groups are obtained by dividing the working conditions of the historical strong correlation variable values in the historical characteristic variable based on a difference dynamic regression vector of an input-output delay order; a target characteristic variable prediction model is obtained by training an initial characteristic variable prediction model based on the plurality of historical strong correlation variable value groups; a target future variable value is determined based on a target current variable value by using the target characteristic variable prediction model; and the main bearing of the offshore wind turbine generator set is faulted early warned based on the target current variable value and the target future variable value. The above method is used to early warn the fault of the main bearing temperature of the offshore wind turbine generator set, thereby improving the safety of the offshore wind turbine generator set during operation.

Description

Technical Field

[0001] This invention relates to the field of offshore wind turbine management, and more specifically, to a fault early warning method and device based on the temperature of the main bearing of an offshore wind turbine. Background Technology

[0002] Offshore wind power generation is one of the currently feasible renewable energy generation technologies. At present, with continuous technological development, more and more regions are investing in offshore wind power facilities in order to obtain clean and sustainable power generation technology. Offshore wind power technology has made significant progress in recent years, including improvements in power generation efficiency, as well as greater durability and reliability. Furthermore, in terms of design, developers have begun to adopt sustainable designs to reduce the potential for environmental pollution. Currently, offshore wind power generation has become a feasible clean energy generation technology and is beginning to impact the world's energy structure, providing society with clean and renewable energy.

[0003] During the daily operation of wind turbine units, the main bearing of the nacelle may experience localized overheating. Overheating of the wind turbine main bearing is a serious problem that can cause severe malfunctions and significant losses to the market. Overheating of the main bearing can lead to bearing failure, which may result in bearing wear, cooling system failure, main shaft notch stretching, main shaft eccentricity, and bearing damage. Furthermore, overheating of the main bearing can also cause wear and malfunctions in the wind turbine unit, accelerating its aging process and reducing the reliability of the operating system. Additionally, overheating can cause oil film detachment, affecting normal operation and even leading to explosions. Therefore, how to implement fault warnings based on the temperature of the main bearing of offshore wind turbines to improve the safety of offshore wind turbine operation has become an urgent problem to be solved. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a fault early warning method and device based on the temperature of the main bearing of an offshore wind turbine, so as to provide fault early warning based on the temperature of the main bearing of the offshore wind turbine and improve the safety of the offshore wind turbine operation.

[0005] In a first aspect, embodiments of this application provide a fault early warning method based on the temperature of the main bearing of an offshore wind turbine, the method comprising:

[0006] Based on the mechanism analysis method, a target historical variable is determined from multiple initial historical variables. The initial historical variables are historical variables that can affect the temperature of the main bearing of the offshore wind turbine. The initial historical variables contain the values ​​of historical variables at multiple different historical moments. The target historical variable is a historically strongly correlated variable of the temperature of the main bearing of the offshore wind turbine. The target historical variable contains the values ​​of historically strongly correlated variables at multiple different historical moments.

[0007] Based on the random forest algorithm, historical feature variables are determined from the target historical variables, wherein the historical feature variables are at least one historical strongly correlated variable value among multiple historical strongly correlated variable values ​​at different historical moments;

[0008] Based on the differential dynamic regression vector of the input-output delay order, the working condition is divided for at least one historically strongly correlated variable value among the historical feature variables to obtain multiple groups of historically strongly correlated variable values.

[0009] The initial feature variable prediction model is trained based on multiple sets of historical strongly correlated variable values ​​to obtain the target feature variable prediction model.

[0010] Using the target feature variable prediction model, the target future variable value is determined based on the target current variable value, wherein the target current variable value is the variable value of the historical strongly correlated variable at the current moment, and the target future variable value is the variable value of the historical strongly correlated variable at a future moment;

[0011] Based on the current and future values ​​of the target variables, a fault warning is issued for the main bearing of the offshore wind turbine.

[0012] Optionally, before determining the target historical variable from multiple initial historical variables based on mechanistic analysis, the method includes:

[0013] The historical variable values ​​contained in the initial historical variables are preprocessed, wherein the preprocessing includes missing value completion and outlier filtering.

[0014] Optionally, the initial historical variables include one or more of the gearbox temperature of the offshore wind turbine, the rotational speed of the offshore wind turbine, the output power of the offshore wind turbine, the sea wind speed in the area where the offshore wind turbine is located, and the sea temperature in the area where the offshore wind turbine is located.

[0015] Optionally, the step of providing fault warning for the main bearing of the offshore wind turbine based on the current value of the target variable and the future value of the target variable includes:

[0016] Calculate the difference between the current value of the target variable and the future value of the target variable;

[0017] Determine whether the difference between the variables exceeds the standard deviation;

[0018] If the difference between the variables exceeds the standard deviation, then the variable weights of the target future variable values ​​are determined using the analytic hierarchy process (AHP) and the entropy weight method, based on the target future variable values.

[0019] The evaluation score of the target future variable value is determined based on the variable weight of the target future variable value;

[0020] Determine whether the evaluation score of the target future variable value exceeds the preset standard evaluation score;

[0021] If the evaluation score of the target future variable value exceeds the preset standard evaluation score, a fault warning will be issued for the main bearing of the offshore wind turbine.

[0022] Optionally, after determining whether the difference in the variables exceeds the standard deviation, the method further includes:

[0023] If the difference between the variables exceeds the standard deviation, no fault warning will be issued for the main bearing of the offshore wind turbine.

[0024] After determining whether the evaluation score of the target future variable value exceeds the preset standard evaluation score, the method further includes:

[0025] If the evaluation score of the target future variable value does not exceed the preset standard evaluation score, then no fault warning will be issued for the main bearing of the offshore wind turbine.

[0026] Secondly, embodiments of this application provide a fault early warning device based on the temperature of the main bearing of an offshore wind turbine, the device comprising:

[0027] The target historical variable determination module is used to determine the target historical variable from multiple initial historical variables based on the mechanism analysis method. The initial historical variables are historical variables that can affect the temperature of the main bearing of the offshore wind turbine. The initial historical variables contain the values ​​of the historical variables at multiple different historical moments. The target historical variable is a historically strongly correlated variable of the temperature of the main bearing of the offshore wind turbine. The target historical variable contains the values ​​of the historically strongly correlated variable at multiple different historical moments.

[0028] The historical feature variable determination module is used to determine historical feature variables from the target historical variables based on the random forest algorithm, wherein the historical feature variables are at least one historical strongly correlated variable value among multiple historical strongly correlated variable values ​​at different historical moments;

[0029] The historical strongly correlated variable value group determination module is used to classify at least one historical strongly correlated variable value among the historical feature variables based on the difference dynamic regression vector of the input-output delay order, and obtain multiple historical strongly correlated variable value groups.

[0030] The target feature variable prediction model generation module is used to train the initial feature variable prediction model based on multiple sets of historical strongly correlated variable values ​​to obtain the target feature variable prediction model.

[0031] The target future variable value determination module is used to determine the target future variable value based on the target current variable value using the target feature variable prediction model, wherein the target current variable value is the variable value of the historical strongly correlated variable at the current time, and the target future variable value is the variable value of the historical strongly correlated variable at a future time;

[0032] The fault warning module is used to provide fault warnings for the main bearing of the offshore wind turbine based on the current variable value and the future variable value of the target.

[0033] Optionally, the device further includes:

[0034] The data preprocessing module is used to preprocess the historical variable values ​​contained in the initial historical variables before the target historical variable determination module determines the target historical variable from multiple initial historical variables based on the mechanism analysis method. The preprocessing includes missing value completion and outlier filtering.

[0035] Optionally, the initial historical variables include one or more of the gearbox temperature of the offshore wind turbine, the rotational speed of the offshore wind turbine, the output power of the offshore wind turbine, the sea wind speed in the area where the offshore wind turbine is located, and the sea temperature in the area where the offshore wind turbine is located.

[0036] Optionally, when the fault early warning module is used to provide fault early warning for the main bearing of the offshore wind turbine based on the current value of the target variable and the future value of the target variable, it is specifically used for:

[0037] Calculate the difference between the current value of the target variable and the future value of the target variable;

[0038] Determine whether the difference between the variables exceeds the standard deviation;

[0039] If the difference between the variables exceeds the standard deviation, then the variable weights of the target future variable values ​​are determined using the analytic hierarchy process (AHP) and the entropy weight method, based on the target future variable values.

[0040] The evaluation score of the target future variable value is determined based on the variable weight of the target future variable value;

[0041] Determine whether the evaluation score of the target future variable value exceeds the preset standard evaluation score;

[0042] If the evaluation score of the target future variable value exceeds the preset standard evaluation score, a fault warning will be issued for the main bearing of the offshore wind turbine.

[0043] Optionally, the fault early warning module is further used for:

[0044] After determining whether the difference of the variables exceeds the standard deviation, if the difference of the variables exceeds the standard deviation, then no fault warning will be issued for the main bearing of the offshore wind turbine.

[0045] After determining whether the evaluation score of the target future variable value exceeds the preset standard evaluation score, if the evaluation score of the target future variable value does not exceed the preset standard evaluation score, then no fault warning will be issued for the main bearing of the offshore wind turbine.

[0046] The technical solution provided in this application includes, but is not limited to, the following beneficial effects:

[0047] This application employs a method combining mechanism and data to select induced feature input variables that influence the operating characteristics of key temperature measurement points, determines the order of input-output delay, and defines a differential dynamic regression vector that can finely characterize the operating conditions of the unit. Then, it utilizes high-dimensional hybrid clustering and hyperplane parameter estimation to achieve a reasonable division of the differential dynamic operating domain, and performs deep learning time-series dynamic modeling in each operating domain to form a multi-operating domain-multi-model system that can efficiently approximate the complex nonlinear dynamics of the entire operating condition. Based on the established dynamic model, it monitors the operating status of key equipment temperature measurement points and provides fault warnings. This enables fault warnings for the main bearing temperature of offshore wind turbines, improving the safety of offshore wind turbine operation.

[0048] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1A flowchart of a fault early warning method based on the temperature of the main bearing of an offshore wind turbine provided in Embodiment 1 of the present invention is shown.

[0051] Figure 2 The flowchart of a specific fault early warning method provided in Embodiment 1 of the present invention is shown;

[0052] Figure 3 A schematic diagram of a fault early warning device based on the temperature of the main bearing of an offshore wind turbine provided in Embodiment 2 of the present invention is shown.

[0053] Figure 4 The diagram shows a structural schematic of the second type of fault early warning device based on the temperature of the main bearing of an offshore wind turbine provided in Embodiment 2 of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0055] Example 1

[0056] To facilitate understanding of this application, the following is combined with... Figure 1 The flowchart illustrating a file access method provided in Embodiment 1 of the present invention will be described in detail for Embodiment 1 of this application.

[0057] See Figure 1 As shown, Figure 1 The flowchart illustrates a fault early warning method based on the temperature of the main bearing of an offshore wind turbine, as provided in Embodiment 1 of the present invention. The method includes steps S101 to S105:

[0058] S101: Based on the mechanism analysis method, a target historical variable is determined from multiple initial historical variables. The initial historical variables are historical variables that can affect the temperature of the main bearing of the offshore wind turbine. The initial historical variables contain the values ​​of historical variables at multiple different historical moments. The target historical variable is a historically strongly correlated variable of the temperature of the main bearing of the offshore wind turbine. The target historical variable contains the values ​​of historically strongly correlated variables at multiple different historical moments.

[0059] Specifically, initial historical variables include, but are not limited to, the temperature at the main bearing temperature measuring point, the ambient temperature, and the operating status of the wind turbine. Historical variable values ​​are obtained by collecting the values ​​of each initial historical variable at multiple different historical moments. Strongly correlated variables, also known as strongly correlated quantities, are defined as those where a change in variable A leads to a change in variable B. When this change is positive and significant, it is called a "strong correlation."

[0060] The temperature of the main bearing of an offshore wind turbine is affected by multiple factors, and there are interactions between internal component factors and external environmental factors. Therefore, we need to determine the strongly correlated quantities affecting the main bearing temperature through mechanistic analysis. First, we preprocess the historical variable values ​​in the initial historical variables to remove missing values, outliers, and data from abnormal operating periods. This preprocessing ensures the accuracy of the input quantities.

[0061] Based on the actual operating conditions of the wind turbine and information regarding the layout of its internal structural components, suitable temperature measurement points are initially selected. Additionally, the impact of main bearing wear must be considered; for example, the presence of wind turbine blade vibration indicates that the main bearing rotation is obstructed and the temperature is too high. Finally, local meteorological forecasts should be combined with comprehensive consideration of the temperature and humidity differences between the wind turbine and the external environment, as well as potential extreme weather events, to determine the strongly correlated quantities affecting the wind turbine main bearing temperature.

[0062] S102: Based on the random forest algorithm, determine the historical feature variables from the target historical variables, wherein the historical feature variables are at least one historical strongly correlated variable value among multiple historical strongly correlated variable values ​​at different historical moments.

[0063] Specifically, through mechanistic analysis, several strongly correlated quantities affecting the main bearing temperature were identified. However, as the number of input variables increases, the number of redundant variables also increases accordingly. Therefore, before dividing the operating domain, we need to determine suitable initially correlated quantities as input to the model to ensure the accuracy of the model input and better divide the operating domain, thereby ensuring the accuracy of the model prediction. To this end, we chose a relevant feature variable screening method (random forest algorithm) to select multiple strongly correlated quantities, that is, to determine historical feature variables from the target historical variables, and finally use the screened historical feature variables as suitable input for the next step.

[0064] More specifically, after data preprocessing, we select relevant features and use the random forest algorithm to determine the input quantities of strongly correlated features. Feature importance is determined by calculating the information gain or Gini coefficient of each feature in the random forest, thus obtaining the importance contribution of each feature. Based on the importance contribution, we can select to retain important features and filter out redundant features. In the random forest, we can extract important features through feature importance or feature ranking. Based on existing SCADA (Supervisory Control and Data Acquisition) data, we use the random forest method to extract features from the data, thereby selecting the most suitable features as input quantities.

[0065] Feature selection is performed using the Random Forest method, data processing is performed, and the effects of different features are compared to ultimately determine the system input. Random Forest is a very powerful machine learning algorithm that can be used for classification and regression problems. Decision Tree is a widely used classification algorithm. Its principle is to judge based on the conditions of the decision point, and if the condition is met, proceed to the next decision point until the leaf node of the decision tree is reached. The learning process of decision tree mainly includes three steps: feature selection, decision tree generation, and decision tree pruning. The first step, feature selection, is to decide which features to use for judgment. The second step, decision tree generation, is to start from the root node, select the feature with the largest information gain as the current node feature, and create child nodes according to different values ​​of the node. Child nodes are generated iteratively until the information gain is very small or there are no features left. The third step, decision tree pruning, is to reduce overfitting by removing some branches.

[0066] We chose CART (Classification and Regression Tree) and used the Gini coefficient instead of information gain. The Gini coefficient is defined as shown in the following formula. CART divides a node into two child nodes, left and right, by selecting the feature whose Gini coefficient decreases the most before and after splitting, until the tree is built.

[0067]

[0068] Where Gini() is the Gini coefficient of data X, n is the number of classes in the dataset, and x i For all feature categories in this dataset, p(x) i ) represents the probability of each category appearing for this feature.

[0069] To optimize the performance of decision trees as individual classifiers, random forests were proposed. Random forests are based on the Bagging (Bootstrap aggregating) ensemble theory and consist of multiple decision trees, which are independent of each other. The basic idea of ​​random forests is to train multiple decision trees simultaneously, comprehensively considering the classification results of each tree to obtain the final classification result of the random forest. That is, when using random forests for classification tasks, each decision tree in the random forest is first judged and classified separately, and each decision tree will produce a classification result. Then, the classification result with the most frequent classification results from all decision trees is taken as the final result. The random forest algorithm consists of two stages: the first stage is to generate a random forest using the training set; the second stage is to make decisions on the test set.

[0070] S103: Based on the differential dynamic regression vector of the input-output delay order, the working condition is divided for at least one historically strongly correlated variable value among the historical feature variables to obtain multiple groups of historically strongly correlated variable values.

[0071] Specifically, we need to divide the operating domain of the unit's main bearing operating data. After division, the data can be processed efficiently. Finite-difference dynamic regression vectors (FDDRs) based on input-output delay orders are a dynamic regression model that treats the changes in input and output as a time series, where there is a certain time delay between input and output at each order, used to describe the dynamic characteristics of the system. The modeling method involves fitting a set of coefficients within a certain time delay order based on the input and output data to represent the dynamic behavior of the system. Specifically, the division of the finite-difference operating domain first determines the delay order, and then, based on the FDDRs, performs hyperplane parameter estimation and clustering to obtain multiple sets of historically strongly correlated variable values.

[0072] Therefore, we use the Akaike information criterion (AIC) to determine the delay order of the input-output delay. According to the Akaike information criterion, the validity of statistical inference depends on the amount of information in the data it contains. Generally, the more information, the higher the validity of the statistical inference. Hyperplane parameter estimation is a machine learning technique that maps a set of multidimensional data to a low-dimensional space, thereby effectively processing the dataset. Hyperplane parameter estimation is an effective technique that can improve the performance of machine learning models.

[0073] Taking the rotational speed of the wind turbine's main bearing as the input variable, we first define a differential dynamic regression vector that can finely characterize the unit's operating conditions. The retardation order is determined using the AIC criterion, and its specific formula is as follows:

[0074] AIC = -2(L) + 2p

[0075] Where AIC is the delay order, L is the likelihood function of the model parameters, and p is the number of parameters. The AIC criterion considers both the model's fit and its complexity, and selects the model with the fewest parameters under the condition of equal goodness of fit as the estimation model.

[0076] After determining the delay order, clustering is performed to better define the operating range. The K-means clustering algorithm is used, and its iterative process is as follows:

[0077] (a) Randomly select k elements from the data set D to serve as the centers of the k clusters.

[0078] (b) Calculate the dissimilarity of the remaining elements to the k cluster centers respectively, and assign these elements to the clusters with the lowest dissimilarity.

[0079] (c) Based on the clustering results, recalculate the center of each of the k clusters by taking the arithmetic mean of the dimensions of all elements in the cluster.

[0080] (d) Recluster all elements in D according to the new centers.

[0081] (e) Repeat step (d) until the clustering results no longer change.

[0082] (f) Output the result.

[0083] K-means clustering has significant advantages in partitioning operating domains: it is an iterative and convergent algorithm with reliable results; it converges quickly, reaching the optimal solution in a relatively small number of iterations; and it can be applied to large datasets, with classification results unaffected by parameters. In actual wind turbine main bearing temperature measurement, the data volume is large, and K-means clustering can effectively and accurately process the data.

[0084] Hyperplane parameter estimation, performed after clustering, works by representing data as a set of parameters and then estimating these parameters using a least-squares fitting method to obtain the best-fit surface, called the hyperplane. Hyperplane parameter estimation can be used for multivariate regression analysis, cluster analysis, classification analysis, and other machine learning tasks. It helps machine learning models better understand complex datasets, thereby improving their accuracy and reliability. Using this technique, machine learning models can better fit multidimensional data and more accurately predict output values.

[0085] In hyperplane parameter estimation, the first step is to select the optimal hyperplane. This can be done by comparing the parameter estimates of the fitted surfaces. These parameter estimates can be calculated using the least squares fitting method. The least squares fitting method is an optimization algorithm that can effectively find parameter estimates that minimize the error. First, we need to estimate the model's parameters, i.e., estimate the model's parameters based on the data to be fitted. Second, we build the model using the estimated parameters and evaluate the model's quality by calculating the sum of squared errors between the fitted curve and the fitted data. Finally, we can gradually adjust the model parameters until the sum of squared errors reaches its minimum value to obtain the best fit result.

[0086] The specific process is as follows:

[0087] Let these dataset sample points be (x i y i Let i = 1, 2, 3, ..., n, and we set the fitted curve as y = kx + b, and let the fitted value be: y i =kx i +b, then we can get:

[0088]

[0089] Another L = ∑(y i -kx i -b) 2 L is defined as the loss function. The values ​​of k and b are determined to minimize the loss function. To achieve this, partial derivatives of L with respect to k and b are taken and set to zero, resulting in:

[0090]

[0091]

[0092] Where y is the ordinate of the fitted curve, k is the slope of the fitted curve, x is the abscissa of the fitted curve, and b is the intersection of the fitted curve and the y-axis. i x represents the ordinate of the sample points in the dataset. i The x-coordinate of the sample points in the dataset. The value of k that minimizes the loss function. Let b be the value that minimizes the loss function. For y i The corresponding predicted value, where n is the total number of curve values.

[0093] S104: Based on multiple sets of historical strongly correlated variable values, train the initial feature variable prediction model to obtain the target feature variable prediction model.

[0094] Specifically, the training process for the initial feature variable prediction model requires time-series dynamic modeling of the wind turbine. This is a method for simulating the wind turbine control system. It can simulate the time-series dynamic behavior of the wind turbine, simulate the operating characteristics of the wind turbine, optimize the performance of the wind turbine, and enable the wind turbine to make autonomous decisions according to different situations.

[0095] The preferred approach is to use a machine learning algorithm based on Block Recurrent Transformer Augmented Regression Tree (BRT) to dynamically model and predict its operation. The BRT model is a deep learning model that combines Transformer (an attention-based sequence model) and RNN (Recurrent Neural Network) structures to process sequence data. The core idea of ​​BRT is to combine Transformer and RNN structures, thus possessing the advantages of both when processing sequence data. Furthermore, this algorithm can perform multi-input, multi-output variable operations, thereby improving the prediction of main bearing temperature.

[0096] After achieving a reasonable division of the differential dynamic operating domain through high-dimensional hybrid clustering and hyperplane parameter estimation, we adopted the block recurrent transformer algorithm for machine learning prediction. By training on data samples within this time series, we predicted the temperature range changes in the following time period.

[0097] Recurrent Neural Networks (RNNs) are a type of neural network model widely used for processing dynamic time-series data. They can process temporal information in the input sequence by introducing a recurrent structure, making them more advantageous in processing time-series data.

[0098] BRT consists of multiple blocks, each containing a Transformer Encoder and an RNN. Within each block, the input sequence first passes through the Transformer Encoder to obtain a new representation. This new representation is then fed into the RNN. The RNN internally maintains a state vector and combines the new representation with the previous state vector to obtain a new state vector. This new state vector is then passed to the next block to process the next input sequence.

[0099] In BRT, the Transformer Encoder is primarily responsible for capturing global dependencies in the input sequence, while the RNN is mainly responsible for capturing local dependencies. By combining the Transformer and RNN structures, BRT can capture both long-term and local dependencies simultaneously, thus better modeling sequence data. Furthermore, BRT utilizes the Transformer's self-attention mechanism, which can compute the representation of each element in the sequence without traversing the entire sequence, thereby improving the model's computational efficiency.

[0100] BRT is a deep learning model that combines Transformer and RNN structures. It can capture both long-term and local dependencies and utilizes the self-attention mechanism of Transformer, resulting in excellent modeling capabilities and computational efficiency.

[0101] In this context, Block Recurrent Transformer (BRT) is employed for multi-input multi-output (MIMO) modeling. The BRT model can accept multiple input sequences and output multiple sequences. This makes BRT highly flexible and applicable to many sequence modeling tasks. For MIMO modeling, the BRT model needs to adjust the shape of the inputs and outputs accordingly. Specifically, each input sequence needs to be encoded with a different embedding, and these embeddings are then concatenated into a large embedding tensor, which serves as the input to the BRT model. On the output side, BRT can output multiple sequences, each with its own output layer. Therefore, BRT has significant advantages in data imputation tasks, better handling multivariate and long sequences, and capable of processing data with missing values.

[0102] S105: Using the target feature variable prediction model, determine the target future variable value based on the target current variable value, wherein the target current variable value is the variable value of the historical strongly correlated variable at the current moment, and the target future variable value is the variable value of the historical strongly correlated variable at a future moment.

[0103] Specifically, by using the current value of the target variable as the model input to the target feature variable prediction model, the future value of the target variable can be obtained.

[0104] S106: Based on the current value of the target variable and the future value of the target variable, provide a fault warning for the main bearing of the offshore wind turbine.

[0105] Specifically, by comparing the predicted values ​​from the previous step with historical data and considering various factors, an evaluation is conducted to determine whether an overheating situation has occurred. Methods for determining indicator weights can be broadly categorized into two types: subjective weighting methods, represented by the analytic hierarchy process (AHP), and objective weighting methods, represented by the entropy weight method (EWM). Subjective weighting methods rely on expert experience for scoring, exhibiting high professionalism and are widely used in the evaluation of new power systems. However, subjective weighting methods are highly subjective, leading scholars to introduce objective weighting methods. These methods use data as a foundation and combine mathematical algorithms to determine weights, reducing subjective influence. Objective weighting methods are prone to indicator failure due to their high sensitivity; therefore, some scholars have proposed combining AHP and EWM to determine indicator weights. After determining the comprehensive weights of each indicator, selecting a suitable comprehensive evaluation method is crucial. The technique for order preference by similarity to ideal solution (TOPSIS) can objectively and realistically reflect the differences between various schemes, avoiding interference from subjective and objective factors.

[0106] After performing machine learning, we observed that it mainly has two output quantities, namely the front and rear main bearing temperatures. Based on this, we conducted a comprehensive evaluation of the variable prediction model.

[0107] Explanation of AHP (Analog-Philosophical Hierarchy Process): In this main bearing temperature prediction model, we need the assistance of experts to complete the model evaluation. There are many factors in this model that will affect our evaluation. For example: (1) Real-time performance: When the feature input is weather temperature and rotational speed, it is generally sufficient to measure the outside temperature every few minutes, because the temperature fluctuation within a day is not too large. At this time, the data volume is small, and the model response time is fast when making predictions. However, for rotational speed, in some cases we need to monitor the rotational speed per second, the data volume is large, the prediction time is long, and the response will be slow. (2) Economy: Experts need to make corresponding considerations. If the data volume of our prediction input and output is too large, the computation cost will also be high. We need to consider the server cost. (3) In addition, for some common model evaluation error indicators such as accuracy and precision, we need to know as much as possible which indicator is more important for our prediction, so that our prediction network can have a focus when adjusting parameters, and better serve the prediction. The main methods are as follows:

[0108] 1) Establish a hierarchical structure. Divide the decision-making objectives, decision-making criteria, and decision-making objects into the highest, middle, and lowest levels according to their relationships.

[0109] 2) Based on the constructed hierarchical architecture model, in practice, scales 1 to 9 are often used to judge the relative importance of each indicator, and the judgment matrix A of the decision model is constructed.

[0110] A = (a ij ) mk×nk

[0111] Where, m k The planning scheme to be evaluated; n k As evaluation indicators for the criteria layer; a ij Empower experts based on their experience.

[0112] 3) Calculate its one-time index λ based on the judgment matrix. CI And find the corresponding average random consistency index λ. RI And calculate the consistency ratio λ CR The formula for a one-time test is:

[0113]

[0114] Where λ is the eigenvalue of matrix A. It is generally considered that λ... CR If the result is ≤0.1, the judgment matrix satisfies the consistency test; otherwise, the judgment matrix needs to be adjusted to satisfy the consistency test.

[0115] 4) After the judgment matrix passes the first test, find its largest eigenvector, which is the objective weight of each indicator.

[0116] EWM calculates the entropy weight of each indicator based on its variability using information entropy, and then adjusts the weights of each indicator using these entropy weights to obtain a more objective set of indicator weights. Unlike subjective analytic hierarchy process (AHP), this method is more objective. The weight of each metric is obtained based on the amount of information in the entropy. The higher the entropy value, the less information is contained, and the smaller the indicator's impact on the overall system. This method is independent of individual consciousness and is a more accurate way to judge the degree of influence of a specific variable on the whole. Researchers can further optimize the indicator system based on the results of the degree of influence of the indicators. For example, in this predictive model, many factors affect the model's accuracy, but their degree of influence is not necessarily the same, so objective and scientific mathematical analysis is required. Taking external temperature as an example, for instance, by investigating the operating conditions of the same type of wind turbine, the impact of rotational speed on the main bearing temperature is different when the external temperature is 10 degrees Celsius compared to 20 degrees Celsius. In this case, it is necessary to combine the power curves at different temperatures with the relationship between rotational speed and the model to judge the degree of influence of external factors on the model, thereby more objectively evaluating the neural network predictive model. The specifics are as follows.

[0117] 1) Based on the proposed scheme data, the original information matrix X is established as follows:

[0118]

[0119] 2) Normalize the index to obtain the elements r of the standardized matrix R. ij for:

[0120]

[0121] 3) Calculate the information entropy E of the indicator. j for:

[0122]

[0123]

[0124] 4) Calculate the weight ω j for

[0125]

[0126] The index weight matrix is ​​obtained based on AHP and EWM. and Composition substrate Let its elements be:

[0127]

[0128] After normalization, we get:

[0129]

[0130] Among them, xi j Let mk be the data dimension of the standardized matrix, and n be the planning scheme to be evaluated. k As the evaluation metric for the criteria layer, x j e is the column vector of the normalized matrix. j For information entropy, Let ω be the index weight matrix obtained from AHP, and let ω be the index weight matrix obtained from EWM. For indicator j, the weight coefficient is the objective weight. With subjective weight ω j The coupling weight, λ j The overall weight of the indicator.

[0131] TOPSIS is a method for systematically evaluating and analyzing multiple objects in multi-objective decision analysis. It evaluates the method's effectiveness by calculating weighted Euclidean distance and the closeness between the positive and negative ideal solutions. TOPSIS is often used to solve multi-indicator decision problems. Its principle is to rank and select the best alternative by calculating the relative distances between each alternative and the positive and negative ideal solutions. Its main steps are as follows:

[0132] 1) Normalize the original matrix.

[0133] 2) Standardize the normalized matrix.

[0134] 3) Calculate and normalize the score. The normalization process involves first comparing the model output with historical values ​​to evaluate the accuracy of the machine learning model. If the parameters of the TOPSI method are unsuitable, the training model needs to be corrected by modifying some parameters of the subjective and objective evaluation indicators, such as the information entropy weight of the objective evaluation indicators. Once the TOPSI indicators are relatively suitable, the model output value can be directly compared with historical values ​​to perform RMSE calculations, and other criteria can be selected to determine whether temperature anomalies have occurred.

[0135] When comparing model output values ​​with historical values ​​to determine whether abnormal temperature values ​​have occurred, we usually select root mean square error (RMSE) and mean absolute percentage error (MAPE) as important criteria for judging whether temperature anomalies have occurred. By setting relevant upper and lower limits for RMSE and MAPE parameters, if the values ​​of the evaluation indicators exceed or fall below the upper and lower limits of the parameters, it can be determined that the main bearing temperature is abnormal at this moment.

[0136] The calculation formula is as follows:

[0137]

[0138]

[0139] Where RMSE is the root mean square error, and N is the total number of samples. * () represents the predicted value, i represents the time series value, y(i) represents the actual value, and MAPE represents the absolute percentage error.

[0140] By comparing the predicted results with historical data values ​​and combining subjective and objective methods, it can be determined whether the temperature of the main bearing working measurement point meets the expected standard during this period and in the future. If the evaluation index values ​​are abnormal, it can be determined that a fault of local overheating has occurred.

[0141] In a feasible implementation, before determining the target historical variable from multiple initial historical variables based on mechanistic analysis, the method includes:

[0142] The historical variable values ​​contained in the initial historical variables are preprocessed, wherein the preprocessing includes missing value completion and outlier filtering.

[0143] In one feasible implementation, the initial historical variables include one or more of the following: gearbox temperature of the offshore wind turbine, rotational speed of the offshore wind turbine, output power of the offshore wind turbine, sea wind speed in the area where the offshore wind turbine is located, and sea temperature in the area where the offshore wind turbine is located.

[0144] In one feasible implementation plan, see Figure 2 As shown, Figure 2 The flowchart illustrates a specific fault early warning method provided in Embodiment 1 of the present invention, wherein the step of providing fault early warning for the main bearing of the offshore wind turbine based on the current value of the target variable and the future value of the target variable includes steps S201 to S205:

[0145] S201: Calculate the difference between the current value of the target variable and the future value of the target variable.

[0146] Specifically, the difference between the current value of the target variable and the future value of the target variable can reflect the change in the future value of the target variable compared to the current value of the target variable.

[0147] S202: Determine whether the difference in the variables exceeds the standard deviation.

[0148] Specifically, it is determined whether the difference between the variables exceeds the standard deviation, so as to determine the degree of change of the target future variable value relative to the target current variable value based on the determination result.

[0149] S203: If the difference between the variables exceeds the standard deviation, then the variable weights of the target future variable values ​​are determined using the analytic hierarchy process (AHP) and the entropy weight method, based on the target future variable values.

[0150] Specifically, if the difference between the variables exceeds the standard deviation, it indicates that the future value of the target variable has a high degree of change compared to the current value of the target variable. This suggests that the main bearing temperature may be abnormally rising or falling. In this case, it is necessary to use the analytic hierarchy process (AHP) and the entropy weight method to determine the variable weights of the future value of the target variable.

[0151] S204: Determine the evaluation score of the target future variable value based on the variable weights of the target future variable value.

[0152] Specifically, the evaluation score of the target future variable value is obtained by multiplying the variable weight of the target future variable value by the target future variable value.

[0153] S205: Determine whether the evaluation score of the target future variable value exceeds the preset standard evaluation score.

[0154] Specifically, it determines whether the evaluation score of the target future variable value exceeds the preset standard evaluation score, and determines whether the wind turbine parameters indicated by the target future variable value exceed the standard value based on the judgment result.

[0155] S206: If the evaluation score of the target future variable value exceeds the preset standard evaluation score, a fault warning will be issued for the main bearing of the offshore wind turbine.

[0156] Specifically, if the evaluation score of the target future variable value exceeds the preset standard evaluation score, it indicates that the wind turbine parameters indicated by the target future variable value have exceeded the standard value, and the offshore wind turbine may be abnormal. In this case, a fault warning needs to be issued for the main bearing of the offshore wind turbine.

[0157] Fault warning methods include, but are not limited to, sound alarms, text or image displays.

[0158] In a feasible implementation, after determining whether the difference of the variables exceeds the standard deviation, the method further includes: if the difference of the variables exceeds the standard deviation, then no fault warning is issued for the main bearing of the offshore wind turbine.

[0159] Specifically, if the difference between the variables exceeds the standard deviation, it indicates that the future value of the target variable is less affected by the current value of the target variable, and it is unlikely that the main bearing temperature will rise or fall abnormally. Therefore, no fault warning will be issued for the main bearing of the offshore wind turbine.

[0160] After determining whether the evaluation score of the target future variable value exceeds the preset standard evaluation score, the method further includes: if the evaluation score of the target future variable value does not exceed the preset standard evaluation score, then no fault warning is issued for the main bearing of the offshore wind turbine.

[0161] Specifically, if the evaluation score of the target future variable value does not exceed the preset standard evaluation score, it means that the wind turbine parameters indicated by the target future variable value do not exceed the standard value, and the offshore wind turbine may not have any abnormalities. Therefore, it is not necessary to issue a fault warning for the main bearing of the offshore wind turbine.

[0162] In addition to providing fault warnings for the main bearing of the offshore wind turbine based on the current and future values ​​of the target variable, the method further includes: after comprehensively evaluating the system output using a machine learning training model, selecting an edge sensing device for efficient state monitoring and early warning functions. An edge intelligent sensing device is a device capable of processing and analyzing data close to the data source. It can send the processed data to the cloud or other systems for further processing. In offshore wind power projects, applying data to edge sensing devices can improve system efficiency and reliability. The edge sensing device, also capable of processing and analyzing data close to the data source, can send the processed data to the cloud or other systems for further processing. Afterward, by uploading to the cloud system, it automatically compares historical data to determine whether an excessively high temperature has occurred, thereby achieving an automatic early warning function.

[0163] Example 2

[0164] See Figure 3 As shown, Figure 3 The diagram shows a structural schematic of a fault early warning device based on the main bearing temperature of an offshore wind turbine provided in Embodiment 2 of the present invention, wherein the device includes:

[0165] The target historical variable determination module 301 is used to determine the target historical variable from multiple initial historical variables based on the mechanism analysis method. The initial historical variables are historical variables that can affect the temperature of the main bearing of the offshore wind turbine. The initial historical variables contain the values ​​of historical variables at multiple different historical moments. The target historical variable is a historically strongly correlated variable of the temperature of the main bearing of the offshore wind turbine. The target historical variable contains the values ​​of historically strongly correlated variables at multiple different historical moments.

[0166] The historical feature variable determination module 302 is used to determine historical feature variables from the target historical variables based on the random forest algorithm, wherein the historical feature variables are at least one historical strongly correlated variable value among multiple historical strongly correlated variable values ​​at different historical moments;

[0167] The historical strongly correlated variable value group determination module 303 is used to classify at least one historical strongly correlated variable value among the historical feature variables based on the differential dynamic regression vector of the input-output delay order, and obtain multiple historical strongly correlated variable value groups.

[0168] The target feature variable prediction model generation module 304 is used to train the initial feature variable prediction model based on multiple sets of historical strongly correlated variable values ​​to obtain the target feature variable prediction model.

[0169] The target future variable value determination module 305 is used to determine the target future variable value based on the target current variable value using the target feature variable prediction model, wherein the target current variable value is the variable value of the historical strongly correlated variable at the current time, and the target future variable value is the variable value of the historical strongly correlated variable at a future time;

[0170] The fault warning module 306 is used to provide fault warnings for the main bearing of the offshore wind turbine based on the current variable value of the target and the future variable value of the target.

[0171] In one feasible implementation plan, see Figure 4 As shown, Figure 4 The diagram illustrates the structure of a second fault early warning device based on the main bearing temperature of an offshore wind turbine, as provided in Embodiment 2 of the present invention. The device further includes:

[0172] The data preprocessing module 401 is used to preprocess the historical variable values ​​contained in the initial historical variables before the target historical variable determination module determines the target historical variable from multiple initial historical variables based on the mechanism analysis method. The preprocessing includes missing value completion and outlier filtering.

[0173] In one feasible implementation, the initial historical variables include one or more of the following: gearbox temperature of the offshore wind turbine, rotational speed of the offshore wind turbine, output power of the offshore wind turbine, sea wind speed in the area where the offshore wind turbine is located, and sea temperature in the area where the offshore wind turbine is located.

[0174] In one feasible implementation, when the fault early warning module is used to provide fault early warning for the main bearing of the offshore wind turbine based on the current value of the target variable and the future value of the target variable, it is specifically used for:

[0175] Calculate the difference between the current value of the target variable and the future value of the target variable;

[0176] Determine whether the difference between the variables exceeds the standard deviation;

[0177] If the difference between the variables exceeds the standard deviation, then the variable weights of the target future variable values ​​are determined using the analytic hierarchy process (AHP) and the entropy weight method, based on the target future variable values.

[0178] The evaluation score of the target future variable value is determined based on the variable weight of the target future variable value;

[0179] Determine whether the evaluation score of the target future variable value exceeds the preset standard evaluation score;

[0180] If the evaluation score of the target future variable value exceeds the preset standard evaluation score, a fault warning will be issued for the main bearing of the offshore wind turbine.

[0181] In one feasible implementation, the fault warning module is further used for:

[0182] After determining whether the difference of the variables exceeds the standard deviation, if the difference of the variables exceeds the standard deviation, then no fault warning will be issued for the main bearing of the offshore wind turbine.

[0183] After determining whether the evaluation score of the target future variable value exceeds the preset standard evaluation score, if the evaluation score of the target future variable value does not exceed the preset standard evaluation score, then no fault warning will be issued for the main bearing of the offshore wind turbine.

[0184] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0185] The fault early warning device based on the main bearing temperature of offshore wind turbines provided in this embodiment of the invention can be specific hardware on the equipment or software or firmware installed on the equipment. The implementation principle and technical effects of the device provided in this embodiment of the invention are the same as those in the foregoing method embodiments. For the sake of brevity, any parts not mentioned in the device embodiments can be referred to the corresponding content in the foregoing method embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can all be referred to the corresponding processes in the above method embodiments, and will not be repeated here.

[0186] In the embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0187] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0188] In addition, the functional units in the embodiments provided by the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0189] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0190] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0191] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. All should be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A fault early warning method based on the temperature of the main bearing of an offshore wind turbine, characterized in that, The method includes: Based on the mechanism analysis method, a target historical variable is determined from multiple initial historical variables. The initial historical variables are historical variables that can affect the temperature of the main bearing of the offshore wind turbine. The initial historical variables contain the values ​​of historical variables at multiple different historical moments. The target historical variable is a historically strongly correlated variable of the temperature of the main bearing of the offshore wind turbine. The target historical variable contains the values ​​of historically strongly correlated variables at multiple different historical moments. Based on the random forest algorithm, historical feature variables are determined from the target historical variables, wherein the historical feature variables are at least one historical strongly correlated variable value among multiple historical strongly correlated variable values ​​at different historical moments; Based on the differential dynamic regression vector of the input-output delay order, the working condition is divided for at least one historically strongly correlated variable value among the historical feature variables to obtain multiple groups of historically strongly correlated variable values. The initial feature variable prediction model is trained based on multiple sets of historical strongly correlated variable values ​​to obtain the target feature variable prediction model. Using the target feature variable prediction model, the target future variable value is determined based on the target current variable value, wherein the target current variable value is the variable value of the historical strongly correlated variable at the current moment, and the target future variable value is the variable value of the historical strongly correlated variable at a future moment; Based on the current and future values ​​of the target variables, a fault warning is provided for the main bearing of the offshore wind turbine. The method of providing fault warning for the main bearing of the offshore wind turbine based on the current value of the target variable and the future value of the target variable includes: Calculate the difference between the current value of the target variable and the future value of the target variable; Determine whether the difference between the variables exceeds the standard deviation; If the difference between the variables exceeds the standard deviation, then the variable weights of the target future variable values ​​are determined using the analytic hierarchy process (AHP) and the entropy weight method, based on the target future variable values. The evaluation score of the target future variable value is determined based on the variable weight of the target future variable value; Determine whether the evaluation score of the target future variable value exceeds the preset standard evaluation score; If the evaluation score of the target future variable value exceeds the preset standard evaluation score, a fault warning will be issued for the main bearing of the offshore wind turbine.

2. The method according to claim 1, characterized in that, Before determining the target historical variable from multiple initial historical variables based on mechanistic analysis, the method includes: The historical variable values ​​contained in the initial historical variables are preprocessed, wherein the preprocessing includes missing value completion and outlier filtering.

3. The method according to claim 1, characterized in that, The initial historical variables include one or more of the following: gearbox temperature of the offshore wind turbine, rotational speed of the offshore wind turbine, output power of the offshore wind turbine, sea wind speed in the area where the offshore wind turbine is located, and sea temperature in the area where the offshore wind turbine is located.

4. The method according to claim 1, characterized in that, After determining whether the difference in the variables exceeds the standard deviation, the method further includes: If the difference in the variables does not exceed the standard deviation, no fault warning will be issued for the main bearing of the offshore wind turbine. After determining whether the evaluation score of the target future variable value exceeds the preset standard evaluation score, the method further includes: If the evaluation score of the target future variable value does not exceed the preset standard evaluation score, then no fault warning will be issued for the main bearing of the offshore wind turbine.

5. A fault early warning device based on the temperature of the main bearing of an offshore wind turbine, characterized in that, The device includes: The target historical variable determination module is used to determine the target historical variable from multiple initial historical variables based on the mechanism analysis method. The initial historical variables are historical variables that can affect the temperature of the main bearing of the offshore wind turbine. The initial historical variables contain the values ​​of the historical variables at multiple different historical moments. The target historical variable is a historically strongly correlated variable of the temperature of the main bearing of the offshore wind turbine. The target historical variable contains the values ​​of the historically strongly correlated variable at multiple different historical moments. The historical feature variable determination module is used to determine historical feature variables from the target historical variables based on the random forest algorithm, wherein the historical feature variables are at least one historical strongly correlated variable value among multiple historical strongly correlated variable values ​​at different historical moments; The historical strongly correlated variable value group determination module is used to classify at least one historical strongly correlated variable value among the historical feature variables based on the difference dynamic regression vector of the input-output delay order, and obtain multiple historical strongly correlated variable value groups. The target feature variable prediction model generation module is used to train the initial feature variable prediction model based on multiple sets of historical strongly correlated variable values ​​to obtain the target feature variable prediction model. The target future variable value determination module is used to determine the target future variable value based on the target current variable value using the target feature variable prediction model, wherein the target current variable value is the variable value of the historical strongly correlated variable at the current time, and the target future variable value is the variable value of the historical strongly correlated variable at a future time; The fault early warning module is used to provide fault early warning for the main bearing of the offshore wind turbine based on the current variable value and the future variable value of the target. When the fault early warning module is used to provide fault early warning for the main bearing of the offshore wind turbine based on the current and future values ​​of the target variable, it is specifically used for: Calculate the difference between the current value of the target variable and the future value of the target variable; Determine whether the difference between the variables exceeds the standard deviation; If the difference between the variables exceeds the standard deviation, then the variable weights of the target future variable values ​​are determined using the analytic hierarchy process (AHP) and the entropy weight method, based on the target future variable values. The evaluation score of the target future variable value is determined based on the variable weight of the target future variable value; Determine whether the evaluation score of the target future variable value exceeds the preset standard evaluation score; If the evaluation score of the target future variable value exceeds the preset standard evaluation score, a fault warning will be issued for the main bearing of the offshore wind turbine.

6. The apparatus according to claim 5, characterized in that, The device further includes: The data preprocessing module is used to preprocess the historical variable values ​​contained in the initial historical variables before the target historical variable determination module determines the target historical variable from multiple initial historical variables based on the mechanism analysis method. The preprocessing includes missing value completion and outlier filtering.

7. The apparatus according to claim 5, characterized in that, The initial historical variables include one or more of the following: gearbox temperature of the offshore wind turbine, rotational speed of the offshore wind turbine, output power of the offshore wind turbine, sea wind speed in the area where the offshore wind turbine is located, and sea temperature in the area where the offshore wind turbine is located.

8. The apparatus according to claim 5, characterized in that, The fault early warning module is also used for: After determining whether the difference of the variables exceeds the standard deviation, if the difference of the variables does not exceed the standard deviation, then no fault warning will be issued for the main bearing of the offshore wind turbine. After determining whether the evaluation score of the target future variable value exceeds the preset standard evaluation score, if the evaluation score of the target future variable value does not exceed the preset standard evaluation score, then no fault warning will be issued for the main bearing of the offshore wind turbine.

Images