A wind turbine fault identification and early warning method and system

Abstract

The application belongs to the technical field of fault identification and early warning, and discloses a wind turbine fault identification and early warning method and system. The method comprises the following steps: based on a cloud data center, an artificial intelligence model is constructed; based on the cloud data center, the artificial intelligence model is deployed to all wind turbine monitoring devices; based on the wind turbine monitoring device, real-time monitoring data of the wind turbine is collected; based on the wind turbine monitoring device, fault identification is performed according to the data-processed real-time monitoring data; based on the wind turbine monitoring device, fault early warning is performed according to the real-time fault identification result; based on the wind turbine monitoring device, data is uploaded to the cloud data center; and based on the cloud data center, a report is generated according to the data-processed real-time monitoring data, the real-time fault identification result and a real-time fault early warning strategy. The application solves the problems of poor real-time performance, low accuracy and low intelligence degree in the prior art.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of fault identification and early warning, and particularly relates to a wind turbine fault identification and early warning method and system. BACKGROUND

[0002] With the development and popularization of clean energy, more and more wind power plants are put into use. Wind turbines are devices that generate electricity using wind energy, which can effectively convert wind energy into electricity for users or transmission to the power grid. The normal operation of wind turbines is the premise of stable and reliable wind power plants. During the operation of wind turbines, faults may occur due to various reasons, affecting the normal operation and service life of wind turbines. Therefore, online monitoring of wind turbines is needed to timely detect faults of wind turbines and issue warning signals to remind workers to repair and handle.

[0003] The existing wind turbine fault identification and early warning technology has the following defects:

[0004] 1) Lack of real-time performance: Traditional fault identification methods often rely on periodic manual inspection or regular data analysis, which cannot achieve real-time monitoring and early warning, leading to delayed fault detection and handling, which may cause greater losses;

[0005] 2) Low accuracy: Existing technologies use simple threshold judgment or rule-based methods for fault identification, which are difficult to cope with complex and variable fault patterns and are prone to false positives or false negatives;

[0006] 3) Low intelligence: Existing technologies only achieve fault identification and early warning, lack intelligent fault report generation function, and require manual sorting of fault reports by workers, which cannot meet the needs. SUMMARY

[0007] In order to solve the problems of lack of real-time performance, low accuracy and low intelligence in the existing technology, the application aims to provide a wind turbine fault identification and early warning method and system.

[0008] The technical solution adopted by the application is:

[0009] A wind turbine fault identification and early warning method, comprising the following steps:

[0010] Based on the cloud data center, an artificial intelligence algorithm is used to build a wind turbine fault identification model, a wind turbine fault early warning model, and a wind turbine fault report generation model;

[0011] Based on the cloud data center, the wind turbine fault identification model and the wind turbine fault early warning model are deployed to all wind turbine monitoring devices connected to the cloud data center;

[0012] Based on the wind turbine monitoring device, real-time monitoring data of the wind turbine is collected, and the real-time monitoring data is processed to obtain data-processed real-time monitoring data;

[0013] Based on the wind turbine monitoring device, according to the data-processed real-time monitoring data, a wind turbine fault identification model is used for fault identification to obtain real-time fault identification results;

[0014] Based on the wind turbine monitoring device, according to the real-time fault identification results, a wind turbine fault warning model is used for fault warning to obtain and respond to real-time fault warning strategies;

[0015] Based on the wind turbine monitoring device, the data-processed real-time monitoring data, real-time fault identification results, and real-time fault warning strategies are uploaded to the cloud data center;

[0016] Based on the cloud data center, according to the data-processed real-time monitoring data, real-time fault identification results, and real-time fault warning strategies, a wind turbine fault report generation model is used for report generation to obtain real-time wind turbine fault reports.

[0017] Further, based on the cloud data center, an artificial intelligence algorithm is used to construct a wind turbine fault identification model, a wind turbine fault warning model, and a wind turbine fault report generation model, including the following steps:

[0018] Based on the cloud data center, historical monitoring data of a plurality of wind turbines is collected, and the historical monitoring data is processed to obtain a plurality of data-processed historical monitoring data;

[0019] According to the plurality of data-processed historical monitoring data, a deep learning and image recognition algorithm is used to construct a wind turbine fault identification model, and a plurality of historical fault identification results are generated;

[0020] According to the plurality of historical fault identification results, a reinforcement learning algorithm is used to construct a wind turbine fault warning model, and a plurality of historical fault warning strategies are generated;

[0021] According to the plurality of data-processed historical monitoring data, the plurality of historical fault identification results, and the plurality of historical fault warning strategies, a deep learning algorithm is used to construct a wind turbine fault report generation model.

[0022] Further, the historical monitoring data includes historical sequence monitoring data and historical image monitoring data; the historical sequence monitoring data includes historical wind deviation monitoring data and historical unit operation monitoring data;

[0023] The real-time monitoring data includes real-time sequence monitoring data and real-time image monitoring data; the real-time sequence monitoring data includes real-time wind deviation monitoring data and real-time unit operation monitoring data.

[0024] Further, the wind turbine fault identification model is constructed based on an FPN-LSTM-Attention-Elman algorithm, and the wind turbine fault identification model comprises an image feature extraction module constructed based on an FPN algorithm, a sequence feature extraction module constructed based on an LSTM algorithm, an attention weight module constructed based on an Attention mechanism, and a wind turbine fault identification module constructed based on an Elman algorithm.

[0025] The wind turbine fault early warning model is constructed based on an MOPPO algorithm, and the wind turbine fault early warning model is provided with a target function set, an experience replay pool, an Actor network, a Critic network, and an agent.

[0026] The wind turbine fault report generation model is constructed based on a cGAN-MLP algorithm, and the wind turbine fault report generation model comprises a generator and a discriminator both constructed based on an RNN algorithm, and a condition embedding module and a condition processing module both constructed based on an MLP algorithm, the generator is connected with the condition embedding module and the discriminator respectively, and the discriminator is connected with the condition processing module.

[0027] Further, according to a plurality of data-processed historical monitoring data, a wind turbine fault identification model is constructed using a deep learning and image recognition algorithm, and a plurality of historical fault identification results are generated, including the following steps:

[0028] An initial wind turbine fault identification model is constructed using an FPN-LSTM-Attention-Elman algorithm, and a plurality of data-processed historical monitoring data are divided into a model training set and a model test set;

[0029] According to the model training set, the initial wind turbine fault identification model is optimized and trained to obtain an optimized wind turbine fault identification model, and a plurality of historical fault identification results and corresponding historical weighted fusion features are generated;

[0030] According to the model test set, the optimized wind turbine fault identification model is tested to obtain a model test accuracy rate, if the model test accuracy rate is greater than an accuracy rate threshold, the final wind turbine fault identification model is output, otherwise, the optimization and training are continued.

[0031] Further, according to a plurality of historical fault identification results, a wind turbine fault early warning model is constructed using a reinforcement learning algorithm, and a plurality of historical fault early warning strategies are generated, including the following steps:

[0032] The fault early warning strategy generation problem is taken as a simulation environment, an initial wind turbine fault early warning model is constructed using the MOPPO algorithm, a set of objective functions, an experience replay pool, an Actor network, a Critic network and an agent are set for the initial wind turbine fault early warning model;

[0033] The historical fault identification results are analyzed to obtain a plurality of historical fault identification states, a state space of the agent is defined according to the plurality of historical fault identification states, and an action space of the agent is defined according to a plurality of preset fault early warning decision actions;

[0034] Based on any objective function in the set of objective functions, the initial wind turbine fault early warning model is pre-trained according to the plurality of historical fault identification results to obtain a pre-trained wind turbine fault early warning model, and a plurality of historical fault early warning strategies and corresponding historical fault early warning experiences are generated;

[0035] The Critic network of the pre-trained wind turbine fault early warning model is used to obtain a plurality of historical wind turbine fault early warning strategies rewards, and an optimized Critic network is obtained;

[0036] The Actor network of the pre-trained wind turbine fault early warning model is optimized according to the plurality of rewards to obtain an optimized Actor network;

[0037] All objective functions in the set of objective functions are traversed, and the above steps are repeated to obtain a final Actor network and a final Critic network;

[0038] The final wind turbine fault early warning model is obtained according to the final Actor network and the final Critic network, and the plurality of historical fault early warning experiences are stored in the experience replay pool.

[0039] Further, according to the plurality of data-processed historical monitoring data, the plurality of historical fault identification results and the plurality of historical fault early warning strategies, a deep learning algorithm is used to construct a wind turbine fault report generation model, including the following steps:

[0040] An initial wind turbine fault report generation model is constructed using the cGAN-MLP algorithm; the initial wind turbine fault report generation model includes an initial generator, an initial discriminator, a condition embedding module and a condition processing module;

[0041] A real wind turbine fault report is set for each data-processed historical monitoring data, corresponding historical fault identification result and corresponding historical fault early warning strategy;

[0042] Combine the first loss function of the initial generator and the second loss function of the initial discriminator of the initial wind turbine fault report generation model to obtain a comprehensive loss function;

[0043] Extract the historical weighted fusion feature of each data processing historical monitoring data, the historical fault identification result feature corresponding to the historical fault identification result, and the historical fault early warning strategy feature corresponding to the historical fault early warning strategy;

[0044] Use the condition information embedder of the initial wind turbine fault report generation model to conditionally embed each historical weighted fusion feature, historical fault identification result feature, historical fault early warning strategy feature, and random noise to obtain a plurality of historical condition information embedding features;

[0045] According to the plurality of historical condition information embedding features, the initial generator of the initial wind turbine fault report generation model is trained to obtain an optimized generator, and a plurality of generated wind turbine fault reports are generated;

[0046] Use the condition information processor to conditionally process the historical condition information embedding features, the corresponding real wind turbine fault report, and the corresponding generated wind turbine fault report to obtain historical condition information;

[0047] According to the real wind turbine fault report, the corresponding generated wind turbine fault report, and the corresponding historical condition information, the initial discriminator is trained to obtain an optimized discriminator, and a plurality of historical data discrimination results are generated;

[0048] According to each generated wind turbine fault report and the corresponding historical data discrimination result, use the comprehensive loss function to obtain the historical comprehensive loss value in the training process;

[0049] If the historical loss value is lower than the loss value threshold, output the final generator and the final discriminator, otherwise, continue the optimization training;

[0050] Integrate the final generator and the final discriminator to obtain the final wind turbine fault report generation model.

[0051] Further, based on the wind turbine monitoring device, according to the data processing real-time monitoring data, using the wind turbine fault identification model, the fault identification is carried out, and the real-time fault identification result is obtained, including the following steps:

[0052] Based on the wind turbine monitoring device, the data processing real-time sequence monitoring data and the data processing real-time image monitoring data are obtained by analyzing the data processing real-time monitoring data;

[0053] An image feature extraction module of the wind turbine fault identification model is used to extract real-time image features of the real-time image monitoring data after data processing;

[0054] A sequence feature extraction module of the wind turbine fault identification model is used to extract real-time sequence features of the real-time sequence monitoring data after data processing;

[0055] According to the preset attention weight, an attention weight module of the wind turbine fault identification model is used to weight and fuse the real-time image data features and the real-time sequence data features to obtain real-time weighted fusion features;

[0056] According to the real-time weighted fusion features, a wind turbine fault identification module of the wind turbine fault identification model is used for fault identification to obtain real-time fault identification results.

[0057] Further, based on the wind turbine monitoring device, according to the real-time fault identification results, a wind turbine fault warning model is used for fault warning to obtain and respond to real-time fault warning strategies, including the following steps:

[0058] According to the real-time fault identification results, the most suitable objective function is selected from the objective function set of the wind turbine fault warning model;

[0059] The real-time fault identification results are analyzed to obtain a plurality of real-time fault identification states, and the state space of the agent is updated according to the plurality of real-time fault identification states to obtain an updated state space;

[0060] A plurality of historical fault warning experiences are extracted in the experience replay pool, and the action space of the agent is updated according to a plurality of preset fault warning decision actions of the plurality of historical fault warning experiences to obtain an updated action space;

[0061] Based on the most suitable objective function, an agent of the wind turbine fault warning model is used to control the Actor network to generate a probability distribution of all possible fault warning decision actions corresponding to each real-time fault identification state in the updated state space in the updated action space;

[0062] The possible fault warning decision action with the highest probability distribution in the updated action space is taken as the execution fault warning decision action of the real-time fault identification state, and the execution fault warning decision actions of all real-time fault identification states are integrated to obtain a real-time fault warning strategy;

[0063] According to the real-time fault warning strategy, a real-time fault warning signal, a first real-time fault control instruction of the wind turbine, and a second real-time fault control instruction of the auxiliary control device are generated;

[0064] In response to the real-time fault early warning signal, the first real-time fault control instruction is sent to the corresponding wind turbine, and the second real-time fault control instruction is sent to the corresponding auxiliary control device.

[0065] A wind turbine fault identification and early warning system for implementing a wind turbine fault identification and early warning method, the system comprising a cloud data center and a plurality of wind turbine monitoring devices, the cloud data center being in communication connection with the plurality of wind turbine monitoring devices, each wind turbine monitoring device being arranged at a wind turbine of a power plant, and each wind turbine monitoring device being in communication connection with a wind turbine and a corresponding auxiliary control device.

[0066] The present application has the following advantages:

[0067] The present application discloses a wind turbine fault identification and early warning method and system, which realizes real-time monitoring and early warning of wind turbines through real-time data acquisition, processing and model analysis, greatly shortens the response time of fault discovery and processing, and effectively avoids greater losses caused by delayed fault discovery; the advanced artificial intelligence algorithm is used to construct a wind turbine fault identification and early warning model, which can more accurately identify complex and variable fault patterns, and develop corresponding fault early warning strategies according to the fault conditions, reducing the phenomenon of false positives and false negatives, improving the accuracy of fault diagnosis, and realizing automatic fault early warning control; not only realizes automatic fault identification and early warning, but also integrates intelligent fault report generation function, automatically organizes and generates personalized fault reports, greatly reduces the burden of staff, and improves the overall work efficiency; realizes the ability of quickly generating and executing early warning strategies, which can quickly respond when a fault occurs, take timely measures, and prevent the fault from expanding.

[0068] Other advantages of the present application will be further described in the specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0069] Figure 1 is a flowchart of the wind turbine fault identification and early warning method in the present application.

[0070] Figure 2 is a structural block diagram of the wind turbine fault identification and early warning system in the present application. DETAILED DESCRIPTION

[0071] The present application will be further explained below in combination with the drawings and specific embodiments.

[0072] Embodiment 1:

[0073] As shown in the drawings, the present embodiment provides a wind turbine fault identification and early warning method, comprising the following steps: Figure 1

[0074] ​S1: Based on the cloud data center, using artificial intelligence algorithm, constructing wind turbine fault identification model, wind turbine fault early warning model and wind turbine fault report generation model, including the following steps:

[0075] S1-1: Based on the cloud data center, collecting historical monitoring data of several wind turbines, and processing the historical monitoring data to obtain several data processed historical monitoring data;

[0076] The historical monitoring data includes historical sequence monitoring data and historical image monitoring data; the historical sequence monitoring data includes historical wind deviation monitoring data and historical unit operation monitoring data;

[0077] The wind deviation monitoring data includes wind deviation parameters, wind deviation mean, wind deviation standard deviation and distribution model fitting parameters;

[0078] The wind deviation monitoring data is obtained by the preset wind deviation calculation program in the wind turbine monitoring device, including the following steps:

[0079] A-1: The wind turbine monitoring device collects the original monitoring parameters of wind direction, wind speed, power and wind angle of the wind turbine, and performs data quality processing such as data time alignment, missing value interpolation and abnormal value filtering on the original parameters to obtain quality processed monitoring parameters;

[0080] A-2: Calculate the wind deviation of each wind turbine, the calculation method is as follows:

[0081] 1) Wind deviation parameter calculation:

[0082] Wind deviation parameter = |wind direction-wind angle|;

[0083] 2) Statistical quantity calculation:

[0084] Wind deviation mean (sliding window) = sum(wind deviation value) / window size;

[0085] Wind deviation standard deviation (sliding window) = sqrt(sum((wind deviation value-mean)^2) / (window size-1));

[0086] 3) Distribution model fitting parameter calculation:

[0087] Normal distribution model:

[0088] Probability density function: f(x|μ,σ) = 1 / sqrt(2×π×σ^2)×exp(-(x-μ)^2 / (2×σ^2));

[0089] In the formula, μ is the mean of wind deviation, σ is the standard deviation of wind deviation, the mean of wind deviation and the standard deviation of wind deviation are obtained by robust estimation;

[0090] t distribution model:

[0091] Probability density function: f(x|μ', σ', v) = Gamma((v+1) / 2) / (sqrt(v x pi) x Gamma(v / 2) x σ' x (1+(x-μ')^2 / (v x σ'^2))^((v+1) / 2));

[0092] In the formula, μ' is a location parameter, σ' is a scale parameter, and v is a degree of freedom, the location parameter, the scale parameter, and the degree of freedom are obtained by maximum likelihood estimation and the like;

[0093] For all units in the same scene, the overall mean and standard deviation of the wind deviation are calculated, the robust estimation method is used to reduce the influence of abnormal values, a normal distribution or t distribution probability model is constructed, and the overall deviation distribution is fitted;

[0094] A-3: The wind deviation monitoring data is obtained by integrating the wind deviation parameters, the mean of wind deviation, the standard deviation of wind deviation, and the distribution model fitting parameters;

[0095] The unit operation monitoring data includes unsaturated operation monitoring data and design defect monitoring data;

[0096] The unsaturated operation monitoring data includes wind speed, power, time, state, and power and other unit operation monitoring parameters of the wind turbine;

[0097] The design defect monitoring data includes gear box oil temperature, power, wind speed, temperature slope, power slope, kurtosis, and zero-crossing rate and other unit operation monitoring parameters of the wind turbine;

[0098] In addition to the above data cleaning, normalization and other preprocessing, principal component analysis algorithm is used for data dimension reduction to improve data quality while reducing data dimension;

[0099] S1-2: According to a plurality of data processing historical monitoring data, a wind turbine fault identification model is constructed using deep learning and image recognition algorithm, and a plurality of historical fault identification results are generated;

[0100] The wind turbine fault identification model is constructed based on a Feature Pyramid Networks (FPN)-Long Short-Term Memory (LSTM)-Attention-Elman algorithm, and the wind turbine fault identification model comprises an image feature extraction module constructed based on the FPN algorithm, a sequence feature extraction module constructed based on the LSTM algorithm, an attention weight module constructed based on the Attention mechanism, and a wind turbine fault identification module constructed based on the Elman algorithm.

[0101] A basic network architecture of the image feature extraction module is constructed using a Convolutional Neural Networks (CNN) algorithm; the basic network architecture comprises alternately connected convolutional layers and pooling layers; through the alternately connected convolutional layers and pooling layers, multi-scale feature maps can be effectively extracted from the image, local features of the image can be captured, and the spatial dimension of the features can be reduced while important feature information is maintained through the pooling layers; the outputs of all the convolutional layers are connected horizontally in a top-down order to obtain a feature pyramid, which is used to extract feature maps of the image data at different scales and perform multi-scale fusion to obtain high-level image features; the LSTM network realizes a memory function through a gate mechanism and can extract deep features of sequence data; the attention weight module is used to weight and fuse the image features and the sequence features according to preset attention weight values, to strengthen the representation of key features affecting the analysis result and improve the analysis accuracy; and the wind turbine fault identification module learns deep information of the weighted and fused features and performs label prediction according to the weighted and fused features.

[0102] According to historical monitoring data after data processing, a wind turbine fault identification model is constructed using a deep learning and image recognition algorithm, and a plurality of historical fault identification results are generated, comprising the following steps:

[0103] S1-2-1: An initial wind turbine fault identification model is constructed using the FPN-LSTM-Attention-Elman algorithm, and the plurality of data-processed historical monitoring data are divided into a model training set and a model test set;

[0104] S1-2-2: The initial wind turbine fault identification model is optimized and trained according to the model training set to obtain an optimized wind turbine fault identification model, and a plurality of historical fault identification results and corresponding historical weighted and fused features are generated;

[0105] S1-2-3: According to the model test set, the optimized wind turbine fault identification model is tested, and the model test accuracy is obtained, if the model test accuracy is greater than the accuracy threshold, the final wind turbine fault identification model is output, otherwise, the optimization training is continued;

[0106] S1-3: According to the historical fault identification results, a wind turbine fault warning model is constructed using a reinforcement learning algorithm, and a plurality of historical fault warning strategies are generated;

[0107] The wind turbine fault warning model is constructed based on a multi-objective proximal policy optimization (MOPPO) algorithm, and the wind turbine fault warning model is provided with a target function set, an experience replay pool, an actor network, a critic network and an agent;

[0108] According to the historical fault identification results, a wind turbine fault warning model is constructed using a reinforcement learning algorithm, and a plurality of historical fault warning strategies are generated, including the following steps:

[0109] S1-3-1: The fault warning strategy generation problem is regarded as a simulation environment, and an initial wind turbine fault warning model is constructed using the MOPPO algorithm, and the initial wind turbine fault warning model is provided with a target function set, an experience replay pool, an actor network, a critic network and an agent;

[0110] The actor network is used to output the probability distribution of the action to be taken under the given state, and the target is to learn an optimal strategy, that is, to maximize the long-term cumulative reward, in the continuous action space, the actor network usually outputs a mean value and an optional variance parameter to describe the probability distribution of the action, the critic network is used to evaluate the value of the given state, that is, to predict the expected return obtained by following the current strategy from the state, usually outputting a scalar value representing the value of the state or the state-action value, the experience replay pool is used to store historical experience for reuse in the training process, and the target function set includes a plurality of functions defined for the wind turbine fault warning targets, including the minimum fault warning cost target, the maximum wind turbine fault warning efficiency target, the minimum fault response time target, the minimum fault response loss target and the maximum wind turbine operation reliability target, etc. ;

[0111] S1-3-2: The historical fault identification results are analyzed to obtain a plurality of historical fault identification states, the state space of the agent is defined according to the plurality of historical fault identification states, and the action space of the agent is defined according to a plurality of preset fault warning decision actions;

[0112] S1-3-3: Based on any objective function in the objective function set, pre-training the initial wind turbine fault early warning model according to a plurality of historical fault identification results to obtain a pre-trained wind turbine fault early warning model, and generating a plurality of historical fault early warning strategies and corresponding historical fault early warning experiences;

[0113] S1-3-4: Using the Critic network of the pre-trained wind turbine fault early warning model, obtaining the rewards of a plurality of historical wind turbine fault early warning strategies, and obtaining an optimized Critic network;

[0114] S1-3-5: According to a plurality of rewards, optimizing the Actor network of the pre-trained wind turbine fault early warning model to obtain an optimized Actor network;

[0115] S1-3-6: Iterating through all objective functions in the objective function set, repeating the above steps to obtain a final Actor network and a final Critic network;

[0116] S1-3-7: According to the final Actor network and the final Critic network, obtaining a final wind turbine fault early warning model, and storing a plurality of historical fault early warning experiences in an experience replay pool;

[0117] S1-4: According to a plurality of historical monitoring data processed by a plurality of data, a plurality of historical fault identification results and a plurality of historical fault early warning strategies, using a deep learning algorithm to construct a wind turbine fault report generation model;

[0118] The wind turbine fault report generation model is constructed based on a conditional generative adversarial network (cGAN)-multilayer perceptron (MLP) algorithm, and the wind turbine fault report generation model includes a generator and a discriminator both constructed based on a recurrent neural network (RNN) algorithm, and a condition embedding module and a condition processing module both constructed based on an MLP algorithm. The generator is connected with the condition embedding module and the discriminator, and the discriminator is connected with the condition processing module.

[0119] The conditional embedding module receives the weighted fusion features, the fault identification result features, the fault early warning strategy features and the random noise, converts the features into embedding vectors that can be understood by the generator using a multilayer perceptron (MLP), and the embedding vectors serve as conditional information to guide the generation process of the generator. The MLP algorithm can effectively handle nonlinear relationships and convert the features into useful embedding representations. The conditional embedding module ensures that the generator can consider specific decision contexts when generating outputs, thereby improving the relevance and accuracy of the generation process. The generator receives the embedding vectors from the conditional embedding module and other relevant information, and generates outputs that match the input conditions using a recurrent neural network (RNN) algorithm, i.e., generates reports. The generator attempts to generate sufficiently realistic data to deceive the discriminator. The generator can generate customized outputs based on decision information, improving flexibility and adaptability. The discriminator receives the outputs from the generator and the outputs from the conditional processing module, and determines whether the data generated by the generator is sufficiently realistic, i.e., whether it meets the actual situation of the real report, using an RNN algorithm. The discriminator guides the training process of the generator through feedback signals. The discriminator improves the quality and authenticity of the data generated by the generator through adversarial training. The adversarial process helps to improve the overall performance and consistency of the execution model. The conditional processing module receives part of the output from the generator and the output from the conditional embedding module, processes this information using an MLP algorithm, and provides additional conditional information to the discriminator to help the discriminator better understand the context of the generator's output. The conditional processing module enhances the model's ability to evaluate the generator's output and improves the reliability of the entire model. By incorporating conditional information, the generator's output is better matched to the actual application scenario.

[0120] According to the historical monitoring data after data processing, the historical fault identification results and the historical fault early warning strategies, a deep learning algorithm is used to construct a wind turbine fault report generation model, including the following steps:

[0121] S1-4-1: An initial wind turbine fault report generation model is constructed using a cGAN-MLP algorithm. The initial wind turbine fault report generation model includes an initial generator, an initial discriminator, a conditional embedding module and a conditional processing module.

[0122] S1-4-2: A real wind turbine fault report is set for each data processing historical monitoring data, corresponding historical fault identification result and corresponding historical fault early warning strategy.

[0123] S1-4-3: The first loss function of the initial generator of the initial wind turbine fault report generation model and the second loss function of the initial discriminator are combined to obtain a comprehensive loss function.

[0124] S1-4-4: extract the historical weighted fusion feature of each data processing historical monitoring data, the historical fault identification result feature corresponding to the historical fault identification result, and the historical fault early warning strategy feature corresponding to the historical fault early warning strategy;

[0125] S1-4-5: using the initial wind turbine fault report generation model, the condition information embedder is used to conditionally embed each historical weighted fusion feature, historical fault identification result feature, historical fault early warning strategy feature and random noise to obtain a plurality of historical condition information embedding features;

[0126] S1-4-6: according to a plurality of historical condition information embedding features, the initial generator of the initial wind turbine fault report generation model is trained to obtain an optimized generator, and a plurality of generated wind turbine fault reports are generated;

[0127] S1-4-7: using the condition information processor, the historical condition information embedding feature, the corresponding real wind turbine fault report and the corresponding generated wind turbine fault report are processed to obtain the historical condition information;

[0128] S1-4-8: according to the real wind turbine fault report, the corresponding generated wind turbine fault report and the corresponding historical condition information, the initial discriminator is trained to obtain an optimized discriminator, and a plurality of historical data discrimination results are generated;

[0129] S1-4-9: according to each generated wind turbine fault report and the corresponding historical data discrimination result, using the comprehensive loss function, the historical comprehensive loss value in the training process is obtained;

[0130] S1-4-10: if the historical loss value is lower than the loss value threshold, the final generator and the final discriminator are output, otherwise, the optimization training is continued;

[0131] S1-4-11: integrating the final generator and the final discriminator, a final wind turbine fault report generation model is obtained;

[0132] S2: based on the cloud data center, the wind turbine fault identification model and the wind turbine fault early warning model are deployed to all wind turbine monitoring devices connected to the cloud data center;

[0133] S3: based on the wind turbine monitoring device, the real-time monitoring data of the wind turbine is collected, and the real-time monitoring data is processed to obtain the data processing real-time monitoring data;

[0134] The real-time monitoring data includes real-time sequence monitoring data and real-time image monitoring data; the real-time sequence monitoring data includes real-time wind deviation monitoring data and real-time unit operation monitoring data;

[0135] S4: Based on the wind turbine monitoring device, real-time monitoring data after data processing is used to perform fault identification using a wind turbine fault identification model to obtain real-time fault identification results, including the following steps:

[0136] S4-1: Based on the wind turbine monitoring device, real-time sequence monitoring data after data processing and real-time image monitoring data after data processing are obtained by analyzing the real-time monitoring data after data processing;

[0137] S4-2: Real-time image features of the real-time image monitoring data after data processing are extracted using an image feature extraction module of the wind turbine fault identification model;

[0138] S4-3: Real-time sequence features of the real-time sequence monitoring data after data processing are extracted using a sequence feature extraction module of the wind turbine fault identification model;

[0139] S4-4: Real-time weighted fusion features are obtained by weighting and fusing the real-time image data features and the real-time sequence data features using an attention weight module of the wind turbine fault identification model according to a preset attention weight;

[0140] S4-5: Real-time fault identification results are obtained by performing fault identification using a wind turbine fault identification module of the wind turbine fault identification model according to the real-time weighted fusion features;

[0141] The fault identification results include wind deviation fault identification results, unsaturated operation fault identification results, and design defect fault identification results, etc.

[0142] S5: Based on the wind turbine monitoring device, real-time fault identification results are used to perform fault warning using a wind turbine fault warning model to obtain and respond to real-time fault warning strategies, including the following steps:

[0143] S5-1: The most suitable objective function is selected from a set of objective functions of the wind turbine fault warning model according to the real-time fault identification results;

[0144] S5-2: A number of real-time fault identification states are obtained by analyzing the real-time fault identification results, and the state space of the agent is updated according to the number of real-time fault identification states to obtain an updated state space;

[0145] S5-3: A number of historical fault warning experiences are extracted from an experience replay pool, and the action space of the agent is updated according to a number of preset fault warning decision actions of the number of historical fault warning experiences to obtain an updated action space;

[0146] S5-4: Based on the most suitable objective function, using the agent of the wind turbine fault early warning model, controlling the Actor network, generating the probability distribution of all possible fault early warning decision actions in the updated action space corresponding to each real-time fault identification state in the updated state space;

[0147] S5-5: Taking the possible fault early warning decision action with the highest probability distribution in the updated action space as the execution fault early warning decision action of the real-time fault identification state, integrating the execution fault early warning decision actions of all real-time fault identification states to obtain a real-time fault early warning strategy;

[0148] S5-6: According to the real-time fault early warning strategy, generating a real-time fault early warning signal, a first real-time fault control instruction of the wind turbine and a second real-time fault control instruction of the auxiliary control device;

[0149] The fault early warning signal includes a wind deviation fault early warning signal, an unsaturated operation fault early warning signal and a design defect fault early warning signal, etc.

[0150] S5-7: In response to the real-time fault early warning signal, sending the first real-time fault control instruction to the corresponding wind turbine and the second real-time fault control instruction to the corresponding auxiliary control device;

[0151] S6: Based on the wind turbine monitoring device, uploading the data processing real-time monitoring data, real-time fault identification result and real-time fault early warning strategy to the cloud data center;

[0152] S7: Based on the cloud data center, using the wind turbine fault report generation model to generate a report according to the data processing real-time monitoring data, real-time fault identification result and real-time fault early warning strategy, to obtain a real-time wind turbine fault report, including the following steps:

[0153] S7-1: Based on the cloud data center, extracting the real-time weighted fusion features of the data processing real-time monitoring data, the real-time fault identification result features of the corresponding real-time fault identification result and the real-time fault early warning strategy features of the corresponding real-time fault early warning strategy;

[0154] S7-2: Using the condition information embedder of the wind turbine fault report generation model to conditionally embed the real-time weighted fusion features, real-time fault identification result features and real-time fault early warning strategy features to obtain real-time condition information embedded features;

[0155] S7-3: Using the generator of the wind turbine fault report generation model to generate a report according to the real-time condition information embedded features to obtain a real-time wind turbine fault report.

[0156] Embodiment 2:

[0157] As Figure 2 shown, the embodiment provides a wind turbine fault identification and early warning system for realizing a wind turbine fault identification and early warning method, the system comprising a cloud data center and a plurality of wind turbine monitoring devices, the cloud data center being in communication connection with the plurality of wind turbine monitoring devices respectively, each wind turbine monitoring device being arranged at a wind turbine of a power plant and being in communication connection with a wind turbine and a corresponding auxiliary control device respectively;

[0158] The cloud data center is configured to use an artificial intelligence algorithm to construct a wind turbine fault identification model, a wind turbine fault early warning model, and a wind turbine fault report generation model, and deploy the wind turbine fault identification model and the wind turbine fault early warning model to all wind turbine monitoring devices connected to the cloud data center.

[0159] The wind turbine monitoring device is configured to collect real-time monitoring data of the wind turbine, perform data processing on the real-time monitoring data to obtain data-processed real-time monitoring data, use the wind turbine fault identification model to perform fault identification according to the data-processed real-time monitoring data to obtain a real-time fault identification result, use the wind turbine fault early warning model to perform fault early warning according to the real-time fault identification result to obtain and respond to a real-time fault early warning strategy, upload the data-processed real-time monitoring data, the real-time fault identification result, and the real-time fault early warning strategy to the cloud data center, and use the wind turbine fault report generation model to perform report generation according to the data-processed real-time monitoring data, the real-time fault identification result, and the real-time fault early warning strategy to obtain a real-time wind turbine fault report.

[0160] The present application discloses a wind turbine fault identification and early warning method and system, which realizes real-time monitoring and early warning of wind turbines through real-time data collection, processing, and model analysis, greatly shortens the response time of fault discovery and processing, and effectively avoids greater losses caused by delayed fault discovery; advanced artificial intelligence algorithms are used to construct wind turbine fault identification and wind turbine early warning models, which can more accurately identify complex and variable fault patterns and develop corresponding fault early warning strategies according to fault conditions, reducing the phenomenon of false positives and false negatives, improving the accuracy of fault diagnosis, and realizing automatic fault early warning control; not only realizing automatic fault identification and early warning, but also integrating intelligent fault report generation functions, automatically organizing and generating personalized fault reports, greatly reducing the burden on staff, and improving overall work efficiency; realizing the ability to quickly generate and execute early warning strategies, which can quickly respond when a fault occurs and take timely measures to prevent the fault from expanding.

[0161] The above detailed description of the specific embodiments of the present application has been given to understand the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A wind turbine generator fault identification and early warning method, characterized in that: It comprises the following steps: Based on the cloud data center, using artificial intelligence algorithm, the wind turbine fault identification model, wind turbine fault warning model and wind turbine fault report generation model are constructed; The wind turbine fault identification model is constructed based on FPN-LSTM-Attention-Elman algorithm, and the wind turbine fault identification model comprises an image feature extraction module constructed based on FPN algorithm, a sequence feature extraction module constructed based on LSTM algorithm, an attention weight module constructed based on Attention mechanism and a motor set fault identification module constructed based on Elman algorithm; The construction method of the wind turbine fault identification model comprises the following steps: Using FPN-LSTM-Attention-Elman algorithm, an initial wind turbine fault identification model is constructed, and a plurality of data processing historical monitoring data are divided into model training set and model test set; According to the model training set, the initial wind turbine fault identification model is optimized and trained to obtain the optimized wind turbine fault identification model, and a plurality of historical fault identification results and corresponding historical weighted fusion features are generated; According to the model test set, the optimized wind turbine fault identification model is tested to obtain the model test accuracy, if the model test accuracy is greater than the accuracy threshold, the final wind turbine fault identification model is output, otherwise, the optimization training is continued; The wind turbine fault warning model is constructed based on MOPPO algorithm, and the wind turbine fault warning model is provided with a target function set, an experience replay pool, an actor network, a critic network and an agent; The wind turbine fault report generation model is constructed based on cGAN-MLP algorithm, and the wind turbine fault report generation model comprises a generator and a discriminator both constructed based on RNN algorithm, and a condition embedding module and a condition processing module both constructed based on MLP algorithm, the generator is connected with the condition embedding module and the discriminator respectively, and the discriminator is connected with the condition processing module; Based on the cloud data center, the wind turbine fault identification model and the wind turbine fault warning model are deployed to all wind turbine monitoring devices connected with the cloud data center; Based on the wind turbine monitoring device, real-time monitoring data of the wind turbine is collected, and the real-time monitoring data is processed to obtain data processing real-time monitoring data; Based on the wind turbine monitoring device, according to the data processing real-time monitoring data, the wind turbine fault identification model is used for fault identification to obtain real-time fault identification result; Based on the wind turbine monitoring device, according to the real-time fault identification result, the wind turbine fault warning model is used for fault warning to obtain and respond to real-time fault warning strategy; Based on the wind turbine monitoring device, the data processing real-time monitoring data, the real-time fault identification result and the real-time fault warning strategy are uploaded to the cloud data center; Based on the cloud data center, according to the real-time monitoring data, the real-time fault identification result and the real-time fault early warning strategy after data processing, using the wind turbine fault report generation model, the report generation is carried out, and the real-time wind turbine fault report is obtained.

2. The wind turbine fault identification and early warning method of claim 1, wherein: Based on the cloud data center, using artificial intelligence algorithm, a wind turbine fault identification model, a wind turbine fault early warning model and a wind turbine fault report generation model are constructed, including the following steps: Based on the cloud data center, a plurality of historical monitoring data of wind turbines are collected, and a plurality of historical monitoring data are processed to obtain a plurality of data processing historical monitoring data; According to the plurality of data processing historical monitoring data, using deep learning and image recognition algorithm, the wind turbine fault identification model is constructed, and a plurality of historical fault identification results are generated; According to the plurality of historical fault identification results, using reinforcement learning algorithm, the wind turbine fault early warning model is constructed, and a plurality of historical fault early warning strategies are generated; According to the plurality of data processing historical monitoring data, the plurality of historical fault identification results and the plurality of historical fault early warning strategies, using deep learning algorithm, the wind turbine fault report generation model is constructed.

3. The wind turbine fault identification and early warning method of claim 2, wherein: The historical monitoring data includes historical sequence monitoring data and historical image monitoring data; the historical sequence monitoring data includes historical wind deviation monitoring data and historical unit operation monitoring data; The real-time monitoring data includes real-time sequence monitoring data and real-time image monitoring data; the real-time sequence monitoring data includes real-time wind deviation monitoring data and real-time unit operation monitoring data.

4. The wind turbine fault identification and early warning method of claim 3, wherein: According to the plurality of historical fault identification results, using reinforcement learning algorithm, the wind turbine fault early warning model is constructed, and a plurality of historical fault early warning strategies are generated, including the following steps: The fault early warning strategy generation problem is regarded as a simulation environment, MOPPO algorithm is used to construct an initial wind turbine fault early warning model, and a target function set, an experience replay pool, an Actor network, a Critic network and an agent are set for the initial wind turbine fault early warning model; The historical fault identification results are analyzed to obtain a plurality of historical fault identification states, the state space of the agent is defined according to the plurality of historical fault identification states, and the action space of the agent is defined according to a plurality of preset fault early warning decision actions; Based on any target function in the target function set, the initial wind turbine fault early warning model is pre-trained according to the plurality of historical fault identification results, a pre-trained wind turbine fault early warning model is obtained, and a plurality of historical fault early warning strategies and corresponding historical fault early warning experiences are generated; Using the Critic network of the pre-trained wind turbine fault early warning model, the rewards of the plurality of historical wind turbine fault early warning strategies are obtained, and the optimized Critic network is obtained; According to the plurality of rewards, the Actor network of the pre-trained wind turbine fault early warning model is optimized to obtain the optimized Actor network; All target functions in the target function set are traversed, and the above steps are repeated to obtain the final Actor network and the final Critic network; According to the final Actor network and the final Critic network, a final wind turbine fault early warning model is obtained, and a plurality of historical fault early warning experiences are stored in an experience replay pool.

5. The wind turbine fault identification and early warning method of claim 4, wherein: According to a plurality of data-processed historical monitoring data, a plurality of historical fault identification results, and a plurality of historical fault early warning strategies, a wind turbine fault report generation model is constructed using a deep learning algorithm, including the following steps: An initial wind turbine fault report generation model is constructed using a cGAN-MLP algorithm; the initial wind turbine fault report generation model includes an initial generator, an initial discriminator, a conditional embedding module, and a conditional processing module; A real wind turbine fault report is set for each data-processed historical monitoring data, corresponding historical fault identification result, and corresponding historical fault early warning strategy; The first loss function of the initial generator of the initial wind turbine fault report generation model and the second loss function of the initial discriminator are combined to obtain a comprehensive loss function; The historical weighted fusion features of each data-processed historical monitoring data, the historical fault identification result features of the corresponding historical fault identification result, and the historical fault early warning strategy features of the corresponding historical fault early warning strategy are extracted; The conditional information embedding of the initial wind turbine fault report generation model is used to conditionally embed each historical weighted fusion feature, historical fault identification result feature, historical fault early warning strategy feature, and random noise to obtain a plurality of historical conditional information embedding features; According to the plurality of historical conditional information embedding features, the initial generator of the initial wind turbine fault report generation model is trained to obtain an optimized generator, and a plurality of generated wind turbine fault reports are generated; The historical conditional information is obtained by using the conditional information processor to conditionally process the historical conditional information embedding features, the corresponding real wind turbine fault report, and the corresponding generated wind turbine fault report. According to the real wind turbine fault report, the corresponding generated wind turbine fault report, and the corresponding historical conditional information, the initial discriminator is trained to obtain an optimized discriminator, and a plurality of historical data discrimination results are generated. According to each generated wind turbine fault report and the corresponding historical data discrimination result, the comprehensive loss function is used to obtain a historical comprehensive loss value during the training process. If the historical loss value is lower than the loss value threshold, the final generator and the final discriminator are output, otherwise, the optimization training is continued. The final generator and the final discriminator are integrated to obtain the final wind turbine fault report generation model.

6. The wind turbine fault identification and early warning method of claim 5, wherein: Based on the wind turbine monitoring device, according to the data-processed real-time monitoring data, the wind turbine fault identification model is used for fault identification to obtain a real-time fault identification result, including the following steps: Based on the wind turbine monitoring device, the data-processed real-time monitoring data is parsed to obtain data-processed real-time sequence monitoring data and data-processed real-time image monitoring data; The image feature extraction module of the wind turbine fault identification model is used to extract real-time image features from the data-processed real-time image monitoring data; The sequence feature extraction module of the wind turbine fault identification model is used to extract real-time sequence features of the real-time sequence monitoring data after data processing; According to the preset attention weight, the attention weight module of the wind turbine fault identification model is used to weight and fuse the real-time image data features and the real-time sequence data features, and real-time weighted fusion features are obtained. According to the real-time weighted fusion features, the wind turbine fault identification model is used to identify the fault, and the real-time fault identification result is obtained.

7. The wind turbine fault identification and early warning method of claim 6, wherein: Based on the wind turbine monitoring device, according to the real-time fault identification result, the wind turbine fault warning model is used for fault warning, and the real-time fault warning strategy is obtained and responded, including the following steps: According to the real-time fault identification result, the most suitable objective function is selected from the objective function set of the wind turbine fault warning model; The real-time fault identification result is analyzed to obtain a plurality of real-time fault identification states, and the state space of the agent is updated according to the plurality of real-time fault identification states to obtain an updated state space; A plurality of historical fault warning experiences are extracted in the experience replay pool, and the action space of the agent is updated according to a plurality of preset fault warning decision actions of the plurality of historical fault warning experiences to obtain an updated action space; Based on the most suitable objective function, the agent of the wind turbine fault warning model is used to control the Actor network to generate the probability distribution of all possible fault warning decision actions corresponding to each real-time fault identification state in the updated state space in the updated action space; The possible fault warning decision action with the highest probability distribution in the updated action space is taken as the execution fault warning decision action of the real-time fault identification state, and the execution fault warning decision actions of all real-time fault identification states are integrated to obtain the real-time fault warning strategy; According to the real-time fault warning strategy, a real-time fault warning signal, a first real-time fault control instruction of the wind turbine and a second real-time fault control instruction of the auxiliary control device are generated; In response to the real-time fault warning signal, the first real-time fault control instruction is sent to the corresponding wind turbine, and the second real-time fault control instruction is sent to the corresponding auxiliary control device.

8. A wind turbine fault identification and early warning system for implementing the wind turbine fault identification and early warning method according to any one of claims 1-7, characterized in that: The system includes a cloud data center and a plurality of wind turbine monitoring devices, the cloud data center is respectively connected with a plurality of wind turbine monitoring devices, each wind turbine monitoring device is arranged at the wind turbine of the power plant, and each wind turbine monitoring device is respectively connected with a wind turbine and a corresponding auxiliary control device.

Images