CN116928038A

CN116928038A - Fault early warning method and device based on main bearing temperature of offshore wind turbine generator

Info

Publication number: CN116928038A
Application number: CN202310889273.5A
Authority: CN
Inventors: 何璇; 胡阳; 付道一; 刘洪麟; 王庆华
Original assignee: North China Electric Power University; CSIC Haizhuang Windpower Co Ltd
Current assignee: North China Electric Power University; CSIC Haizhuang Windpower Co Ltd
Priority date: 2023-07-19
Filing date: 2023-07-19
Publication date: 2023-10-24

Abstract

The application provides a fault early warning method and device based on the temperature of a main bearing of an offshore wind turbine, wherein a target historical variable is determined from a plurality of initial historical variables based on a mechanism analysis method; determining a historical characteristic variable from the target historical variable based on a random forest algorithm; carrying out working condition division on the historical strong correlation variable values in the historical characteristic variables based on the differential dynamic regression vector of the input-output delay order to obtain a plurality of historical strong correlation variable value groups; training the initial characteristic variable prediction model based on a plurality of historical strong correlation variable value groups to obtain a target characteristic variable prediction model; determining a target future variable value based on the target current variable value by using a target characteristic variable prediction model; and performing fault early warning on the main bearing of the offshore wind turbine based on the target current variable value and the target future variable value. By adopting the method, the fault early warning is carried out aiming at the temperature condition of the main bearing of the offshore wind turbine, so that the safety of the offshore wind turbine during operation is improved.

Description

Fault early warning method and device based on main bearing temperature of offshore wind turbine generator

Technical Field

The invention relates to the field of offshore wind turbine generator system management, in particular to a fault early warning method and device based on the temperature of a main bearing of an offshore wind turbine generator system.

Background

Offshore wind turbines are one of the currently available renewable energy power generation technologies. At present, with the continuous development of technology, more and more areas start to invest in offshore wind power generation facilities so as to obtain clean and sustainable power generation technology. The development of offshore wind power technology has made great progress in recent years, including improvements to increase the power generation efficiency, and higher durability and reliability. In addition, in design, developers have begun to adopt sustainable designs to reduce the likelihood of environmental pollution. At present, offshore wind power generation has become a viable clean energy power generation technology and begins to influence the world energy structure, and can provide clean and renewable energy for society.

In the daily operation process of the wind turbine generator, the main bearing of the engine room can meet the condition of local overhigh temperature, the overheating of the main bearing of the wind turbine generator is a serious problem, serious faults of the wind turbine generator can be caused, and great loss is brought to the market. The overheating of the main bearing of the wind turbine generator system can lead to bearing failure, and the bearing failure can lead to bearing abrasion, failure of a cooling system, stretching of a main shaft notch, main shaft eccentricity, bearing damage and the like. In addition, the overheating of the main bearing of the wind turbine generator system also can lead to abrasion and faults of the wind turbine generator system, so that the aging speed of the wind turbine generator system is accelerated, and the reliability of an operation system is reduced. In addition, the overheat of the main bearing of the wind turbine generator set can also cause oil film to fall off, thereby affecting the normal operation and even causing explosion accidents. Therefore, how to perform fault early warning on the temperature condition of the main bearing of the offshore wind turbine so as to improve the safety of the offshore wind turbine during operation becomes a problem to be solved urgently.

Disclosure of Invention

In view of the above, the application aims to provide a fault early warning method and device based on the temperature of the main bearing of the offshore wind turbine, so as to perform fault early warning aiming at the temperature condition of the main bearing of the offshore wind turbine, and improve the safety of the offshore wind turbine during operation.

In a first aspect, an embodiment of the present application provides a fault early warning method based on a main bearing temperature of an offshore wind turbine, where the method includes:

determining a target historical variable from a plurality of initial historical variables based on a mechanism analysis method, wherein the initial historical variable is a historical variable which can influence the main bearing temperature of the offshore wind turbine, the initial historical variable comprises a plurality of historical variable values at different historical moments, the target historical variable is a historical strong correlation variable of the main bearing temperature of the offshore wind turbine, and the target historical variable comprises a plurality of historical strong correlation variable values at different historical moments;

determining a history characteristic variable from the target history variable based on a random forest algorithm, wherein the history characteristic variable is at least one history strong correlation variable value in history strong correlation variable values at different history moments;

Based on the differential dynamic regression vector of the input-output delay order, carrying out working condition division on at least one historical strong correlation variable value in the historical characteristic variable to obtain a plurality of historical strong correlation variable value groups;

model training is carried out on the initial characteristic variable prediction model based on a plurality of history strong correlation variable value groups, and a target characteristic variable prediction model is obtained;

determining a target future variable value based on a target current variable value by using the target characteristic variable prediction model, wherein the target current variable value is a variable value of the history strong correlation variable at the current moment, and the target future variable value is a variable value of the history strong correlation variable at the future moment;

and performing fault early warning on the main bearing of the offshore wind turbine based on the target current variable value and the target future variable value.

Optionally, before determining the target historical variable from the plurality of initial historical variables based on the mechanism analysis method, the method comprises:

and carrying out data preprocessing on the historical variable values contained in the initial historical variable, wherein the preprocessing comprises missing value complementation and outlier filtering.

Optionally, the initial historical variable includes one or more of a gearbox temperature of the offshore wind turbine, a rotational speed of the offshore wind turbine, an output power of the offshore wind turbine, an offshore wind speed of an area of the offshore wind turbine, and an offshore temperature of the area of the offshore wind turbine.

Optionally, the performing fault early warning on the main bearing of the offshore wind turbine generator based on the target current variable value and the target future variable value includes:

calculating a variable difference between the target current variable value and the target future variable value;

judging whether the variable difference value exceeds a standard difference value or not;

if the variable difference value exceeds the standard difference value, determining the variable weight of the target future variable value according to the target future variable value by using a hierarchical analysis method and an entropy weight method;

determining an evaluation score of the target future variable value according to the variable weight of the target future variable value;

judging whether the evaluation score of the target future variable value exceeds a preset standard evaluation score or not;

and if the evaluation score of the target future variable value exceeds the preset standard evaluation score, performing fault early warning on the main bearing of the offshore wind turbine.

Optionally, after determining whether the variable difference exceeds the standard deviation, the method further comprises:

if the variable difference value exceeds the standard difference value, performing fault early warning on the main bearing of the offshore wind turbine generator;

After judging whether the evaluation score of the target future variable value exceeds a preset standard evaluation score, the method further comprises the steps of;

and if the evaluation score of the target future variable value does not exceed the preset standard evaluation score, performing fault early warning on the main bearing of the offshore wind turbine.

In a second aspect, an embodiment of the present application provides a fault early warning device based on a temperature of a main bearing of an offshore wind turbine, where the device includes:

the system comprises a target historical variable determining module, a target historical variable determining module and a target historical variable determining module, wherein the target historical variable determining module is used for determining a target historical variable from a plurality of initial historical variables based on a mechanism analysis method, the initial historical variable is a historical variable which can influence the main bearing temperature of the offshore wind turbine, the initial historical variable comprises a plurality of historical variable values at different historical moments, the target historical variable is a strong historical related variable of the main bearing temperature of the offshore wind turbine, and the target historical variable comprises a plurality of strong historical related variable values at different historical moments;

the historical characteristic variable determining module is used for determining a historical characteristic variable from the target historical variable based on a random forest algorithm, wherein the historical characteristic variable is at least one historical strong correlation variable value in the historical strong correlation variable values at a plurality of different historical moments;

The historical strong correlation variable value group determining module is used for dividing the working condition of at least one historical strong correlation variable value in the historical characteristic variable based on the differential dynamic regression vector of the input-output delay order to obtain a plurality of historical strong correlation variable value groups;

the target characteristic variable prediction model generation module is used for carrying out model training on the initial characteristic variable prediction model based on a plurality of history strong correlation variable value groups to obtain a target characteristic variable prediction model;

the target future variable value determining module is used for determining a target future variable value based on a target current variable value by utilizing the target characteristic variable prediction model, wherein the target current variable value is a variable value of the history strong correlation variable at the current moment, and the target future variable value is a variable value of the history strong correlation variable at the future moment;

and the fault early warning module is used for carrying out fault early warning on the main bearing of the offshore wind turbine based on the target current variable value and the target future variable value.

Optionally, the apparatus further comprises:

and the data preprocessing module is used for preprocessing data of the historical variable values contained in the initial historical variables before the target historical variable determining module determines the target historical variable from the plurality of initial historical variables based on a mechanism analysis method, wherein the preprocessing comprises missing value complementation and outlier filtering.

Optionally, the fault early warning module is configured to, when performing fault early warning on the main bearing of the offshore wind turbine based on the target current variable value and the target future variable value, specifically:

Optionally, the fault early warning module is further configured to:

after judging whether the variable difference value exceeds a standard difference value, if the variable difference value exceeds the standard difference value, not performing fault early warning on the main bearing of the offshore wind turbine;

after judging whether the evaluation score of the target future variable value exceeds a preset standard evaluation score, if the evaluation score of the target future variable value does not exceed the preset standard evaluation score, performing fault early warning on the main bearing of the offshore wind turbine.

The technical scheme provided by the application comprises the following beneficial effects:

the method adopts a mechanism and data combination method to select an induction characteristic input variable which influences the operation characteristics of key temperature measuring points, carries out input-output delay order determination, and defines a differential dynamic regression vector which can finely represent the operation working condition of the unit; then, reasonable division of differential dynamic working condition domains is realized by utilizing high-dimensional hybrid clustering and hyperplane parameter estimation, deep learning time sequence dynamic modeling is carried out on each working condition domain, and a multiplexing Kuang Yu-multi-model capable of efficiently approaching full-working condition complex nonlinear dynamic is formed; then, based on the established dynamic model, monitoring the running state of the temperature measuring point of the key equipment and performing fault early warning; the method can perform fault early warning on the temperature condition of the main bearing of the offshore wind turbine, and improves the safety of the offshore wind turbine during operation.

In order to make the above objects, features and advantages of the present invention more comprehensible, preferred embodiments accompanied with figures are described in detail below.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the embodiments will be briefly described below, it being understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and other related drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 shows a flow chart of a fault early warning method based on the temperature of a main bearing of an offshore wind turbine provided by an embodiment of the invention;

FIG. 2 is a flowchart of a specific fault early warning method according to a first embodiment of the present invention;

fig. 3 shows a schematic structural diagram of a fault early warning device based on a main bearing temperature of an offshore wind turbine provided by a second embodiment of the invention;

fig. 4 shows a schematic structural diagram of a second fault early warning device based on the temperature of the main bearing of the offshore wind turbine provided by the second embodiment of the invention.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present application more apparent, the technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application, and it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments. The components of the embodiments of the present application generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the application, as presented in the figures, is not intended to limit the scope of the application, as claimed, but is merely representative of selected embodiments of the application. All other embodiments, which can be made by a person skilled in the art without making any inventive effort, are intended to be within the scope of the present application.

Example 1

For the convenience of understanding the present application, the following describes in detail the first embodiment of the present application with reference to the flowchart of the first embodiment of the present application shown in fig. 1.

Referring to fig. 1, fig. 1 shows a flowchart of a fault early warning method based on a main bearing temperature of an offshore wind turbine provided by an embodiment of the application, where the method includes steps S101 to S105:

S101: and determining a target historical variable from a plurality of initial historical variables based on a mechanism analysis method, wherein the initial historical variable is a historical variable which can influence the main bearing temperature of the offshore wind turbine, the initial historical variable comprises a plurality of historical variable values at different historical moments, the target historical variable is a historical strong correlation variable of the main bearing temperature of the offshore wind turbine, and the target historical variable comprises a plurality of historical strong correlation variable values at different historical moments.

Specifically, the initial historical variables include, but are not limited to, main bearing temperature measuring point temperature, external environment temperature and wind turbine running state, and variable values of each initial historical variable are collected at a plurality of different historical moments to obtain historical variable values. A strong correlation variable is also known as a strong correlation quantity, and "strong correlation" refers to a change in a variable B when the variable a is changed, and when the change is positive and significant, the change is referred to as "strong correlation".

The main bearing temperature of the offshore wind turbine is influenced by a plurality of factors, and the internal component factors and the external environment factors of the main bearing temperature still influence each other, so that a strong correlation quantity influencing the main bearing temperature needs to be determined through mechanism analysis. Firstly, each historical variable value in the initial historical variable is subjected to data preprocessing, and missing values, abnormal values and data during abnormal operation in the data are removed. The accuracy of the input quantity is ensured by carrying out certain preprocessing operation on the data.

And a proper temperature measuring point is preliminarily selected according to the relevant information of the layout of the internal structural components of the fan under the condition of actual operation of the fan, and the influence caused by abrasion of the main bearing is considered, so that if the vibration phenomenon of the fan blade occurs, the situation that the rotation state of the main bearing is blocked and the temperature is too high can be judged. And finally, combining the local weather table prediction data, integrating the temperature difference and the humidity difference of the wind turbine generator and the external environment, and fully considering the extreme weather possibly occurring in the future to determine the strong correlation quantity affecting the temperature of the main bearing of the fan.

S102: and determining a history characteristic variable from the target history variable based on a random forest algorithm, wherein the history characteristic variable is at least one history strong correlation variable value in history strong correlation variable values at a plurality of different history moments.

Specifically, a plurality of strong correlation quantities influencing the temperature of the main bearing are finally determined through mechanism analysis, but as input variables are increased, redundant variables generated by the main bearing are correspondingly increased, so before the working condition domain division work is carried out, the initial proper strong correlation quantity is needed to be determined as the input quantity of the model, the model input precision is ensured, the working condition domain division can be carried out better, and the model prediction precision is ensured. For this purpose, we select a related characteristic variable screening method (random forest algorithm), and select a plurality of strong related quantities, namely, determine a historical characteristic variable from the target historical variable, and finally, use the screened historical characteristic variable as a proper input quantity to perform the next operation.

More specifically, after data preprocessing, we choose the relevant feature quantity to choose a random forest algorithm to determine the input quantity of the strong relevant quantity. The feature importance is obtained by calculating the information gain or GINI coefficient (keni coefficient) or the like of each feature in a random forest. Based on the importance contribution, we can choose to preserve important features and filter redundant features. In random forests, we can extract important features by feature importance or feature ordering. Based on the data export of the existing SCADA (Supervisory Control And Data Acquisition, data acquisition and monitoring control system), the data is subjected to feature extraction by adopting a random forest method, so that the most suitable feature quantity is screened out as an input quantity.

And (3) carrying out feature screening by a random forest method, processing data, and finally determining the input quantity of the system by selecting different feature comparison effects. Random forests are a very powerful machine learning algorithm that can be used for classification and regression problems. The decision tree is a widely applied classification algorithm, and the principle is that the decision tree is judged by the condition of the decision point, and the condition is met to enter the next decision point until the leaf node of the decision tree is searched. The decision tree mainly comprises three steps of feature selection, decision tree generation and decision tree pruning in the learning process: step one, feature selection: deciding which features to use for making the determination. Secondly, generating a decision tree: and starting from the root node, selecting the characteristic with the maximum information gain as the current node characteristic, and establishing the child nodes according to different values of the node. And generating the child nodes circularly until the information gain is small or no characteristic exists. Third, pruning the decision tree: the overfitting phenomenon is reduced by removing part of the branches.

We choose CART (classification regression tree) to replace the information gain with the coefficient of the foundation, the coefficient of the foundation is defined as the following formula, and CART divides the node into left and right sub-nodes by selecting the feature that the coefficient of the foundation drops most before and after splitting until the tree construction is completed.

Wherein Gini () is the coefficient of the data X, n is the number of categories of the data set, X _i For all feature classes of the dataset, p (x _i ) For the probability of each category occurring on the feature.

In order to optimize the performance of the decision tree as a single classifier, a random forest is proposed. Random forests are based on Bagging (Bootstrap aggregating, guided aggregation algorithm) integration theory, and consist of a plurality of decision trees, and no relation exists between different decision trees. The basic idea of the random forest is to train a plurality of decision trees simultaneously, comprehensively consider the classification result of each decision tree and finally obtain the classification result of the random forest. When the random forest is used for classifying tasks, firstly, judging and classifying each decision tree in the random forest respectively, wherein each decision tree can obtain a classification result, and then, taking one classification with the largest classification result of all the decision trees as a final result. The random forest algorithm includes two phases: the first stage is to generate a random forest by using a training set; and the second stage makes decisions on the test set.

S103: and based on the differential dynamic regression vector of the input-output delay order, carrying out working condition division on at least one historical strong correlation variable value in the historical characteristic variables to obtain a plurality of historical strong correlation variable value groups.

Specifically, working condition data of the main bearing of the unit needs to be divided into working areas, and the data can be processed efficiently after the working areas are divided. The differential dynamic regression vector based on input-output delay orders is a dynamic regression model that treats the changes in input and output as a time series, where there is a certain time delay between the input and output of each order to describe the dynamic characteristics of the system. The modeling method is to fit a set of coefficients in a certain time delay order according to input and output data to represent the dynamic behavior of the system. Specifically, the finite difference working domain is divided, the delay order is determined first, and then hyperplane parameter estimation is carried out on the finite difference working domain based on the difference dynamic regression vector to cluster the finite difference working domain to obtain a plurality of historical strong correlation variable value groups.

For this purpose we choose the red-pool information volume criterion (Akaike information criterion, AIC) to determine the delay order of the delay input-output, according to which the validity of the statistical inference depends on the information volume of the data it contains. In general, the greater the amount of information, the greater the effectiveness of statistical inference. Hyperplane parameter estimation is a machine learning technique that can map a set of multidimensional data to a low dimensional space to efficiently process a data set. Hyperplane parameter estimation is an effective technique to improve the performance of machine learning models.

The method comprises the steps of obtaining the rotating speed of a main bearing of a wind turbine generator as an input value variable, defining a differential dynamic regression vector capable of finely characterizing the running condition of the wind turbine generator, and determining a delay order by an AIC criterion, wherein a specific formula is characterized as follows:

AIC＝-2(L)+2p

wherein AIC is delay order, L is likelihood function of model parameter, p is parameter number, AIC criterion considers fitting condition of model and complexity condition, and adopts model with minimum parameters under condition of equal fitting goodness as estimation model.

After determining the delay order, carrying out clustering division on the delay order to better determine the working condition domain, and adopting a K-means clustering algorithm, wherein the iterative process of the K-means clustering algorithm is as follows:

(a) K elements are randomly taken from the dataset D as the respective centers of the k clusters.

(b) The dissimilarity of the remaining elements to the centers of k clusters is calculated separately, and these elements are classified separately into clusters with the lowest dissimilarity.

(c) And (3) recalculating the centers of the k clusters according to the clustering result, wherein the calculation method is to take the arithmetic average of the dimensions of all elements in the clusters.

(d) All elements in D are re-clustered according to the new center.

(e) Repeating the step (d) until the clustering result is not changed.

(f) And outputting the result.

The K-means clustering algorithm has obvious advantages in working condition domain division: the K-means clustering algorithm is an iterative convergence algorithm, and the result is very reliable; the convergence speed of the K-means clustering algorithm is high, and the optimal solution can be achieved in a small number of iteration times; the K-means clustering algorithm can be applied to a large dataset, and the result of data classification is not affected by parameters. In the actual temperature measurement process of the main bearing of the fan, the data volume is large, and the data can be effectively and accurately processed through a K-means clustering algorithm.

The working principle of the technology is that the super-plane parameter estimation is carried out after clustering: the best-fit surface is obtained by representing the data as a set of parameters and then estimating these parameters using a least squares fitting method, this surface being referred to as a hyperplane. The hyperplane parameter estimation may be used for multivariate regression analysis, cluster analysis, classification analysis, and other machine learning tasks. The hyperplane parameter estimation may help the machine learning model better understand complex data sets, thereby improving the accuracy and reliability of the model. Using this technique, the machine learning model can better fit the multidimensional data, predicting the output values more accurately.

In the hyperplane parameter estimation, the best hyperplane first needs to be selected. This may be accomplished by comparing parameter estimates of the fitted surface. These parameter estimates may be calculated by least squares fitting. The least squares fitting method is an optimization algorithm that can effectively find the parameter estimates that minimize the error. First, we need to perform parameter estimation on the model, i.e. estimate the parameters of the model from the fitted data. And secondly, constructing a model by using the estimated parameters, and evaluating the advantages and disadvantages of the model by calculating the sum of squares of errors of the fitting curve and the fitted data. Finally, we can adjust the model parameters step by step until the sum of squares of the errors reaches a minimum to obtain the best fit result.

The specific process is as follows:

let these data set sample points be (x _i ，y _i ) Let i=1, 2,3 …, n, let us set the fitted curve to y=kx+b, let the fitted value be: y is _i ＝kx _i +b then we can get:

and l= Σ (y _i -kx _i -b) ² L is defined as a loss function, the loss function is minimized by determining the values of k and b, and for this purpose, the value of k and b is respectively biased and then zero, so that:

wherein y is the ordinate of the fitting curve, and k is the fitting curve X is the abscissa of the fitted curve, b is the intersection of the fitted curve and the y-axis, y _i Is the ordinate, x, of the sample points of the dataset _i As the abscissa of the sample points of the dataset,for the value of k when the loss function is minimum, < ->For the value of b when the loss function is minimum, < ->Is y _i And the corresponding predicted value, n, is the total number of curve values.

S104: and carrying out model training on the initial characteristic variable prediction model based on a plurality of the historical strong correlation variable value groups to obtain a target characteristic variable prediction model.

Specifically, in the training process of the initial characteristic variable prediction model, time sequence dynamic modeling of the wind turbine is needed, which is a method for simulating a wind turbine control system, and the method can simulate time sequence dynamic behavior of the wind turbine, simulate operation characteristics of the wind turbine, optimize performance of the wind turbine, and enable the wind turbine to independently make decisions according to different conditions.

Dynamic modeling prediction of its operating condition is preferably performed using machine learning based on Block Recurrent Transformer enhanced regression tree (BRT) algorithms. The BRT model is a deep learning model based on a combination of a transducer (a sequence model based on the mechanism of attention) and RNN (Recurrent Neural Network ) structures for processing sequence data. The core idea of BRT is to combine the transducer and RNN structures, so that the advantages of the transducer and the RNN are achieved when sequence data are processed, and the algorithm mode can realize multiple-input-multiple-output variable operation, so that the main bearing temperature is predicted better.

After the reasonable division of the differential dynamic working condition domain is realized through high-dimensional hybrid clustering and hyperplane parameter estimation, a block recurrent transformer algorithm is adopted to conduct machine learning prediction, and the temperature range change in the following time is predicted through training the data samples in the time sequence.

A recurrent neural network (Recurrent Neural Network, RNN) is a neural network model that is widely used for processing dynamic time series data, which can process time series information in an input sequence by introducing a recurrent structure, making it more advantageous in processing time series data.

BRT is made up of a plurality of blocks, each Block containing one Transformer Encoder and one RNN. In each Block, the input sequence is first passed through Transformer Encoder to obtain a new token, then the new token is input into the RNN, a state vector is maintained inside the RNN, and the new token is combined with the previous state vector to obtain a new state vector. This new state vector is passed to the next Block for processing the next input sequence.

In BRT, transformer Encoder is mainly responsible for capturing global dependencies in the input sequence, while RNN is mainly responsible for capturing local dependencies in the input sequence. By combining the transducer and RNN structures, BRT can capture both long-term and local dependencies, thereby better modeling sequence data. In addition, BRT also utilizes self-attribute mechanism of the transducer, and can calculate the characterization of each element in the sequence without traversing the whole sequence, thereby improving the calculation efficiency of the model.

BRT is a deep learning model combining a transducer and an RNN structure, can capture long-term dependency and local dependency simultaneously, and has good modeling capability and calculation efficiency by using a self-attribute mechanism of the transducer.

In this case, a multiple input multiple output modeling was performed using Block Recurrent Transformer (BRT). The BRT model may accept multiple input sequences and output multiple sequences. This allows great flexibility and applicability of BRT in many sequence modeling tasks. For multiple-input multiple-output modeling, the BRT model requires a corresponding adjustment of the shape of the inputs and outputs. Specifically, each input sequence needs to be encoded with a different coding, which is then concatenated into one large coding transducer as input to the BRT model. In terms of output, the BRT may output multiple sequences, each with its own output layer. Therefore, BRT has great advantages in data padding tasks, can better handle cases of multiple variables and long sequences, and can handle data with missing values.

S105: and determining a target future variable value based on a target current variable value by using the target characteristic variable prediction model, wherein the target current variable value is a variable value of the history strong correlation variable at the current moment, and the target future variable value is a variable value of the history strong correlation variable at the future moment.

Specifically, the target current variable value is used as a model input of the target characteristic variable prediction model and is input into the target characteristic variable prediction model, so that the target future variable value can be obtained.

S106: and performing fault early warning on the main bearing of the offshore wind turbine based on the target current variable value and the target future variable value.

Specifically, through the output of the predicted value in the last step and the combination of the comparison of the historical data values, the factors in all aspects are comprehensively evaluated to judge whether the condition of overhigh temperature occurs. Methods of determining the index weight can be broadly classified into 2 types: a subjective weighting method represented by an analytic hierarchy process (analytic hierarchy process, AHP) and an objective weighting method represented by an entropy weighting method (entropy weight method, EWM). The subjective weighting method is scored according to the experience of an expert, has high expertise and is widely applied to the evaluation method of the novel electric power system. However, the subjective weighting method is high in subjectivity, and students introduce an objective weighting method, so that the weights are determined based on data and combined with a mathematical algorithm, and subjective influence is reduced. The objective weight method is easy to cause index failure due to high sensitivity, so that partial scholars propose to determine index weight by combining AHP and EWM; after determining the comprehensive weight of each index, it is important to select a proper comprehensive evaluation method. The approach to ideal sorting method (technique for order preference by similarity to ideal solution, TOPSIS) can objectively and truly reflect the differences of all schemes and avoid the interference of subjective and objective factors.

After machine learning, we observe that there are mainly two outputs, namely front and rear main bearing temperatures, and perform systematic comprehensive evaluation on the variable predictive model.

Description of AHP analytic hierarchy process: in this main bearing temperature prediction model, we need the assistance of an expert to complete the model evaluation. There are many factors in this model that can affect our evaluation. Such as: (1) instantaneity: when the characteristic input quantity is weather temperature and rotating speed, the weather condition is enough to measure the external temperature every few minutes generally, because the temperature amplitude change in one day is not too large, the data quantity is small, the model response time is very fast when the prediction is carried out, but the rotating speed per second needs to be monitored under some conditions, the data quantity is very large, and the prediction time is long and the response is very slow. (2) economy: the expert is required to take the corresponding consideration, and if the predicted input and output data size is too large, the calculation cost is also high, and the server cost is considered. (3) In addition, for some common model evaluation error indexes such as the correlation quantities of accuracy and precision, which index is important to our prediction is known as far as possible, so that when our prediction network carries out parameter adjustment, the important point is present, and the prediction service is better. The main method of (2) is as follows:

1) A hierarchical architecture structure is established. The decision targets, decision criteria and decision objects are divided into a highest layer, a middle layer and a lowest layer according to the relation among the decision targets, the decision criteria and the decision objects.

2) According to the constructed hierarchical architecture model, the relative importance of each index is judged by adopting a scale of 1-9 in practice, and a judgment matrix A of a decision model is constructed.

A＝(a _ij ) _mk×nk

Wherein m is _k A planning scheme to be evaluated is adopted; n is n _k An evaluation index which is a criterion layer; a, a _ij The expert is given the right to the experience.

3) Calculating the disposable index lambda according to the judgment matrix _CI And find the corresponding average random consistency index lambda _RI And calculate the consistency ratio lambda _CR The one-time test formula is:

where λ is the characteristic root of matrix a. Generally considered lambda _CR When the value is less than or equal to 0.1, the judgment matrix meets the consistency test, otherwise, the judgment matrix needs to be adjusted to meet the consistency test.

4) After the judgment matrix passes one-time inspection, the maximum eigenvector, namely the objective weight of each index, is calculated

The EWM calculates the entropy weight of each index by utilizing the information entropy according to the variability degree of each index, and corrects the weight of each index by the entropy weight, so that objective index weight is obtained. Unlike subjective analytic hierarchy process, the method is more objective. The weight of each metric is obtained from the amount of information in the entropy. The larger the entropy value, the smaller the information amount, and the smaller the influence of the index on the whole. The method is independent of individual consciousness, and is a method for judging the overall influence degree of specific variables more accurately. Researchers can further optimize the index system according to the results of the degree of influence of the index. For example, in this predictive model, there are many factors that affect the accuracy of the model, but they do not necessarily affect to the same extent, so objective and scientific mathematical theory analysis is required. Taking the external temperature as an example, for example, by researching the running condition of the wind power generator of the same type before, the influence of the rotating speed on the temperature of the main bearing when the external temperature is 10 ℃ is different from that when the external temperature is 20 ℃. In this time, the influence degree of external factors on the model is judged by combining the power curves at different temperatures and the relation between the power curves and the rotating speeds, so that the evaluation of the neural network prediction model is more objectively carried out. Specifically, the following is described.

1) According to the formulated scheme data, the original information matrix X is established as follows:

2) Index normalization processing is carried out to obtain an element R of a standardized matrix R _ij The method comprises the following steps:

3) Calculating index information entropy E _j The method comprises the following steps:

4) Calculating the weight omega _j Is that

Obtaining an index weight matrix according to the AHP and the EWMAndcomposition base->The elements are as follows:

and (3) after normalization treatment, obtaining:

wherein xi _j For the data dimension of the standardized matrix, mk is the planning scheme to be evaluated; n is n _k As an evaluation index of a criterion layer, x _j For normalizing the column vectors of the matrix, e _j For the purpose of the information entropy,in order to obtain an index weight matrix according to AHP, ω is an index weight matrix according to EWM, </i >>Objective weight for index j weight coefficient +.>And subjective weight omega _j Is lambda _j Is the comprehensive weight of the index.

The TOPSIS method is a method for systematically evaluating and analyzing multiple objects in multi-objective decision analysis, and the advantages and disadvantages of the method are evaluated by weighting the closeness degree of Euclidean distance calculation, positive ideal solution and negative ideal solution. TOPSIS is mostly used to solve multi-index decision problem, and its implementation principle is to sort and select by calculating the relative distance between each alternative scheme and positive and negative ideal solutions. The method mainly comprises the following steps:

1) The original matrix is forward.

2) The forward normalization matrix is normalized.

3) The score is calculated and normalized. The normalization process is to compare the history model output with the history value, perform an accuracy evaluation on the machine learning model, and if the relevant parameters of the topsis method are not suitable, modify some relevant parameters of the subjective and objective evaluation indexes, such as the information entropy weight of the objective evaluation indexes, to perform correction on the training model. Until the index of the topsis method is relatively suitable, the model output value at the moment can be directly used for carrying out RMSE operation related to the historical value and then selecting other criteria to judge whether temperature abnormality occurs.

In comparing the model output value with the historical value to determine whether an abnormal temperature value occurs, we generally choose the root mean square error (root mean square error, RMSE), absolute percentage error (mean absolute percentage error, MAPE) as the important decision of whether it occurs a temperature abnormality: by setting the upper and lower parameter limits of the relevant RMSE and MAPE, if the value of the evaluation index exceeds or falls below the upper and lower parameter limits, it can be determined that an abnormality has occurred in the main bearing temperature at that time.

The calculation formula is as follows:

wherein RMSE is the magnitude of the root mean square error, N is the total number of samples, ^* () For the predicted value, i is the timing value, y (i) is the actual value, MAPE is the absolute percentage error

And comparing the predicted result with the historical data value, and determining whether the temperature of the working measuring point of the main bearing in the period and a period in the future meets the expected standard by combining a subjective and objective comprehensive method, and if the evaluation index value is abnormal, judging that the fault with overhigh local temperature occurs.

In one possible embodiment, before determining the target historical variable from the plurality of initial historical variables based on a mechanism analysis method, the method comprises:

In one possible embodiment, the initial historical variable comprises one or more of a gearbox temperature of the offshore wind turbine, a rotational speed of the offshore wind turbine, an output power of the offshore wind turbine, an offshore wind speed of an area of the offshore wind turbine, and an offshore temperature of the area of the offshore wind turbine.

In a possible implementation manner, referring to fig. 2, fig. 2 shows a flowchart of a specific fault early warning method provided by an embodiment of the present invention, wherein the fault early warning is performed on the main bearing of the offshore wind turbine based on the target current variable value and the target future variable value, and the steps include steps S201 to S205:

S201: and calculating a variable difference value between the target current variable value and the target future variable value.

Specifically, the variable difference between the target current variable value and the target future variable value may represent a change in the target future variable value as compared to the target current variable value.

S202: and judging whether the variable difference value exceeds a standard difference value.

Specifically, whether the variable difference value exceeds a standard difference value is judged, so that the change degree of the target future variable value compared with the target current variable value is determined according to the judging result.

S203: and if the variable difference value exceeds the standard difference value, determining the variable weight of the target future variable value by using a hierarchical analysis method and an entropy weight method according to the target future variable value.

Specifically, if the variable difference value exceeds the standard difference value, it is indicated that the target future variable value has a higher degree of change than the target current variable value, and it is indicated that abnormal increase or decrease of the main bearing temperature may occur, and then it is necessary to determine the variable weight of the target future variable value by using a hierarchical analysis method and an entropy weight method.

S204: and determining the evaluation score of the target future variable value according to the variable weight of the target future variable value.

Specifically, the variable weights of the target future variable values are multiplied by the target future variable values respectively to obtain the evaluation scores of the target future variable values.

S205: and judging whether the evaluation score of the target future variable value exceeds a preset standard evaluation score.

Specifically, whether the evaluation score of the target future variable value exceeds a preset standard evaluation score is judged, so that whether the fan parameter indicated by the target future variable value exceeds a standard value is determined according to the judgment result.

S206: and if the evaluation score of the target future variable value exceeds the preset standard evaluation score, performing fault early warning on the main bearing of the offshore wind turbine.

Specifically, if the evaluation score of the target future variable value exceeds the preset standard evaluation score, it is indicated whether the fan parameter indicated by the target future variable value exceeds a standard value, and the offshore wind turbine is likely to be abnormal, then fault early warning is needed to be performed on the main bearing of the offshore wind turbine.

The fault early warning mode includes but is not limited to an audible alarm prompt, a text or image display prompt.

In one possible embodiment, after determining whether the variable difference exceeds a standard deviation, the method further comprises: and if the variable difference value exceeds the standard difference value, not performing fault early warning on the main bearing of the offshore wind turbine.

Specifically, if the variable difference value exceeds the standard difference value, which indicates that the change degree of the target future variable value is lower than that of the target current variable value, the condition that the temperature of the main bearing is abnormally increased or decreased is unlikely to occur, and fault early warning is not performed on the main bearing of the offshore wind turbine.

After judging whether the evaluation score of the target future variable value exceeds a preset standard evaluation score, the method further comprises the steps of; and if the evaluation score of the target future variable value does not exceed the preset standard evaluation score, performing fault early warning on the main bearing of the offshore wind turbine.

Specifically, if the evaluation score of the target future variable value does not exceed the preset standard evaluation score, which indicates that the fan parameter indicated by the target future variable value does not exceed the standard value, the offshore wind turbine may not be abnormal, and fault early warning is not needed to be performed on the main bearing of the offshore wind turbine.

In addition, after performing fault early warning on the main bearing of the offshore wind turbine based on the target current variable value and the target future variable value, the method further includes: after the system output is comprehensively evaluated by a machine learning training model, an edge perception device is selected for carrying out an efficient state monitoring and early warning function. The edge intelligent sensing device refers to equipment capable of processing and analyzing data at a position close to a data source, and the equipment can send the processed data to a cloud or other systems for further processing. In offshore wind projects, the application of data to the marginalized sensing device can improve the efficiency and reliability of the system. The marginalized sensing device is a device capable of processing and analyzing data at a position close to a data source, and can send the processed data to a cloud or other systems for further processing, and then automatically compares historical value data through an uploading cloud system to judge whether the temperature is too high or not, so that an automatic early warning function is realized.

Example two

Referring to fig. 3, fig. 3 shows a schematic structural diagram of a fault early warning device based on a main bearing temperature of an offshore wind turbine provided by a second embodiment of the present invention, where the device includes:

the target historical variable determining module 301 is configured to determine a target historical variable from a plurality of initial historical variables based on a mechanism analysis method, where the initial historical variable is a historical variable that can affect the main bearing temperature of the offshore wind turbine, the initial historical variable includes a plurality of historical variable values at different historical moments, the target historical variable is a strong historical related variable of the main bearing temperature of the offshore wind turbine, and the target historical variable includes a plurality of strong historical related variable values at different historical moments;

a historical feature variable determining module 302, configured to determine a historical feature variable from the target historical variables based on a random forest algorithm, where the historical feature variable is at least one historical strong correlation variable value of the historical strong correlation variable values at a plurality of different historical moments;

the historical strong correlation variable value group determining module 303 is configured to perform working condition division on at least one historical strong correlation variable value in the historical feature variables based on the differential dynamic regression vector of the input-output delay order, so as to obtain a plurality of historical strong correlation variable value groups;

The target feature variable prediction model generation module 304 is configured to perform model training on an initial feature variable prediction model based on a plurality of the historical strong correlation variable value sets, so as to obtain a target feature variable prediction model;

a target future variable value determining module 305, configured to determine a target future variable value based on a target current variable value by using the target feature variable prediction model, where the target current variable value is a variable value of the historical strong correlation variable at a current time, and the target future variable value is a variable value of the historical strong correlation variable at a future time;

and the fault early warning module 306 is used for carrying out fault early warning on the main bearing of the offshore wind turbine based on the target current variable value and the target future variable value.

In a possible implementation manner, referring to fig. 4, fig. 4 shows a schematic structural diagram of a second fault early warning device based on a main bearing temperature of an offshore wind turbine according to a second embodiment of the present invention, where the device further includes:

the data preprocessing module 401 is configured to perform data preprocessing on a history variable value included in an initial history variable before the target history variable determining module determines the target history variable from a plurality of initial history variables based on a mechanism analysis method, where the preprocessing includes deficiency value completion and outlier filtering.

In a possible implementation manner, the fault early warning module is specifically configured to, when performing fault early warning on the main bearing of the offshore wind turbine based on the target current variable value and the target future variable value:

In one possible embodiment, the fault early warning module is further configured to:

It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described system and apparatus may refer to corresponding procedures in the foregoing method embodiments, which are not described herein again.

The fault early warning device based on the temperature of the main bearing of the offshore wind turbine provided by the embodiment of the invention can be specific hardware on equipment or software or firmware installed on the equipment and the like. The device provided by the embodiment of the present invention has the same implementation principle and technical effects as those of the foregoing method embodiment, and for the sake of brevity, reference may be made to the corresponding content in the foregoing method embodiment where the device embodiment is not mentioned. It will be clear to those skilled in the art that, for convenience and brevity, the specific operation of the system, apparatus and unit described above may refer to the corresponding process in the above method embodiment, which is not described in detail herein.

In the embodiments provided in the present invention, it should be understood that the disclosed apparatus and method may be implemented in other manners. The above-described apparatus embodiments are merely illustrative, for example, the division of the units is merely a logical function division, and there may be other manners of division in actual implementation, and for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be through some communication interface, device or unit indirect coupling or communication connection, which may be in electrical, mechanical or other form.

The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in the embodiments provided in the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.

The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

It should be noted that: like reference numerals and letters in the following figures denote like items, and thus once an item is defined in one figure, no further definition or explanation of it is required in the following figures, and furthermore, the terms "first," "second," "third," etc. are used merely to distinguish one description from another and are not to be construed as indicating or implying relative importance.

Finally, it should be noted that: the above examples are only specific embodiments of the present invention, and are not intended to limit the scope of the present invention, but it should be understood by those skilled in the art that the present invention is not limited thereto, and that the present invention is described in detail with reference to the foregoing examples: any person skilled in the art may modify or easily conceive of the technical solution described in the foregoing embodiments, or perform equivalent substitution of some of the technical features, while remaining within the technical scope of the present disclosure; such modifications, changes or substitutions do not depart from the spirit and scope of the corresponding technical solutions. Are intended to be encompassed within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. The fault early warning method based on the temperature of the main bearing of the offshore wind turbine generator is characterized by comprising the following steps:

2. The method of claim 1, wherein prior to determining the target historical variable from the plurality of initial historical variables based on the mechanism analysis, the method comprises:

3. The method of claim 1, wherein the initial historical variable comprises one or more of a gearbox temperature of the offshore wind turbine, a rotational speed of the offshore wind turbine, an output power of the offshore wind turbine, an offshore wind speed of an area of the offshore wind turbine, and an offshore temperature of the area of the offshore wind turbine.

4. The method according to claim 1, wherein the performing fault pre-warning on the main bearing of the offshore wind turbine based on the target current variable value and the target future variable value comprises:

5. The method of claim 4, wherein after determining whether the variance value exceeds a standard deviation value, the method further comprises:

6. Fault early warning device based on marine wind turbine generator system main bearing temperature, its characterized in that, the device includes:

7. The apparatus of claim 6, wherein the apparatus further comprises:

8. The apparatus of claim 6, wherein the initial historical variable comprises one or more of a gearbox temperature of the offshore wind turbine, a rotational speed of the offshore wind turbine, an output power of the offshore wind turbine, an offshore wind speed of an area of the offshore wind turbine, and an offshore temperature of the area of the offshore wind turbine.

9. The device according to claim 6, wherein the fault pre-warning module is configured to, when performing fault pre-warning on the main bearing of the offshore wind turbine based on the target current variable value and the target future variable value, specifically:

10. The apparatus of claim 9, wherein the fault pre-warning module is further configured to: