Wind Turbine Fault Detection And Diagnosis Based On LSTM Neural Network

Nowadays,the world’s energy supply is more and more dependent on renewable energy,and wind energy is widely used because of its rich resources and low cost.In the development of all kinds of renewable energy,wind power generation technology has become one of the most potential power generation methods because of its relatively mature technology,high performance-to-price ratio and large-scale development.With the increasing expansion of the scale of wind turbines and the continuous increase of maintenance costs,the maintenance needs of wind turbines have been widely concerned.It is necessary to diagnose and isolate the faults of wind turbines in time to find the hidden troubles in the operation of wind turbines.This is of great practical significance to improve the operation reliability of wind turbines,reduce the waste of resources and the cost of manpower and material resources caused by wind turbines outage,and promote the healthy development of wind power industry.The fault detection of wind turbine systems can be scientifically classified as the model-based methods and the data-based method.The model-based fault detection methods first need to have a full understanding of the wind turbine,and then build an accurate mathematical model.Their accuracy is high,but at the high cost of manpower and material resources,and poor robustness.Data-based methods obtain data from the monitoring and data acquisition systems,which do not need additional sensors.They are relatively economical.With the help of big data accumulated by wind turbines,we can adopt artificial intelligence method to analyze wind turbine faults,which is more suitable for real-world applications.This thesis studies the fault detection method based on artificial intelligence,mainly considering the robustness of the method when deployed on different types of wind turbines.We design a fault detection algorithm based on artificial intelligence to improve the detection accuracy.The ultimate goal of this thesis is to design a real-time and high-precision wind turbine fault detection method.In order to achieve the above goals,the research work is divided into the following parts:Firstly,in order to train and verify the fault detection method based on wind turbines data,it is necessary to construct a wind turbine reference model under the Matlab/Simulink environment to simulate the operation of the wind turbine under normal conditions and various faults.The model divides the wind turbine into pitch control system,transmission system,generator&converter system,and controller.At the same time,the sensor faults and system faults that may occur in the wind turbine operation process are comprehensively considered.In order to judge whether the fault occurs and locate the fault in time,it is necessary to adopt an effective fault diagnosis scheme to quickly take countermeasures to make the wind turbine operate stably and save the maintenance cost.In addition,to get the characteristic signal of the wind turbine failure,the researchers can modify the key parameters in the model to simulate the fault location,time period,fault type and degree of the wind turbine.After obtaining all kinds of data of the wind turbines benchmark model,two wind turbines fault detection methods based on LSTM are designed based on their data characteristics.The first method is to use LSTM to predict the future wind turbines operation condition and to predict the future wind turbines data by learning the non-fault wind turbines data.If there is some specific error between the future data and the expected data,then it can be considered that a fault has occurred.At the same time,it can also judge the type of wind turbines faults.The second method is to emulate the traditional data-based fault detection mode,by using LSTM to train the non-fault data and fault data,and apply the trained model to fault classification.Two methods both have a good effect in the simulation experiment of wind turbine fault prediction.In addition,as one of the research achievements in the field of wind turbines fault detection,the fault detection method based on support vector machine will be compared with the proposed method in this paper.After finishing the relevant experiments in this paper,the following conclusions can be drawn:Firstly,the common characteristic of many fault detection algorithms based on classification is that they need fault data and fault-free data for model training,but in practical application,the limited fault data samples will lead to the lack of data information.In fact,the prediction method based on LSTM is to predict the future wind turbines operation by training normal data,which can be detected in time when faults that are different from the actual operation occur.Secondly,the classification method based on LSTM designed in this paper lies in the fault detection method based on SVM.The wind turbines fault detection method based on LSTM classification designed in this paper has higher accuracy than that method based on SVM.