| The use of a large amount of fossil energy has brought about environmental pollution.Wind energy has attracted more and more attention as a renewable energy source,and the penetration rate of wind power in the power generation field is increasing.Considering that the wind turbine runs in a bad environment all the year round,each component will be confronted with fault in a certain probability.Therefore,it is necessary to diagnose the fault accurately and establish an efficient diagnosis and isolation system for wind terbine system so as to improving the stability and reliability of wind power technology.The data-driven diagnosis method can reduce the dependence on the physical model and realize the effective fault diagnosis through the feature mining of historical data.Based on the benchmark model of wind turbine,this paper proposes a fault diagnosis system based on deep learning,which can effectively diagnose single and multiple faults of wind turbine.This article first briefly analyzes the main components of the wind turbine and summarizes the utility of each system in the wind turbine as well as their common faults,and establishes a mathematical model of each component.Then we establish a benchmark model of the wind turbine in the Matlab / Simulink simulation platform.By setting related faults,the fault data can be obtained to provide our subsequent work.Second,because the data-driven algorithm is a complex and tedious project.To this end,this article establishes an end-to-end deep learning framework,including data preprocessing,model selection,hyperparameter automatic optimization and other functions,which greatly reduces the workload.Since this article is based on data driving,the original simulation data needs to be processed accordingly.In the data preprocessing part,the data quality is improved by methods such as noise reduction,missing value interpolation,and imbalanced data processing.Through hyperparameter automatic tuning,the network can be effectively optimized.Then,on the basis of building the wind turbine benchmark model and deep learning framework.This paper proposes a single fault diagnosis model for wind turbines through onedimensional convolutional network and denoised auto encoder decoder.One-dimensional convolutional network can extract sequence data features in parallel,which achieves excellent result on the diagnosis of single faults.At the same time,a denoised auto encoder decoder is introduced to effectively reduce the noise of original sequence data,which greatly improves the robustness of the modelFinally,consider the conditions that multiple fault may occur in the wind turbine system.A sequence generation model based on attention mechanism is proposed.The sequence generation model realizes the diagnosis of multiple faults by performing feature mining on the data of multiple fan sensors and generating a sequence containing possible faults.Comparison with other multi-label algorithms proves the effectiveness of the model proposed in this paper. |
