Research On Wind Turbine Load Forecasting Method Based On Combined Model

In recent years,with the development of the wind power industry,the focus of wind power construction has shifted from land to sea,and the design capacity of single units has continued to increase.The safety issues during the operation of wind turbines have become more serious.One of the relatively prominent problem at present is that it is difficult to analyze the load conditions of various parts of the wind turbine during operation,which may cause damage to the wind turbine.In engineering,The load of wind turbines will be calculated to check if it meets the requirements before it is put into use.There is no process of recalculating the load in subsequent operations.In view of the large number of data parameters required for calculation by software,and the rapid change in operating conditions in actual operation,the simulation method is not suitable for real-time monitoring during the operation of the unit.So it is proposed to use measured data and neural network,and take multiple parts of the wind turbine as the research object,the following research has been done in this thesis:Firstly,the thesis uses the method of actual measurement,such as install a lidar in front of the wind turbine,to collect information such as wind conditions,and collect the data of the wind turbine itself through the SCADA system to obtain the initial parameters at the same time.This thesis collects the load conditions of important parts of the wind turbine,and selects the wheel hub load,foundation load and tower bottom load as objects for the prediction and research.After obtaining the parameters,the wind speed during the operation of the wind turbine is carried out,and the single-step time series model is used to predict the ultra-short-term wind speed,and the error is reduced to 0.0225.Secondly,in view of the difficulty of calculating the short-term load of wind turbines,this paper proposes to use two types of single network models: BP neural network and Extreme Learning Machine(ELM),to predict the short-term load status of wind turbines.The predicted data of established predictive model will be compared with the data of actual measured to determine the feasibility of a single neural network model for prediction.Taking the force in the x direction of the wheel hub load as an example,the average relative error of the BP neural network prediction results is 0.0512,and the average relative error of the ELM prediction results is0.0509.In this thesis,this two models are improved on the basis of single neural network prediction.The genetic algorithm is used to optimize the BP neural network,the particle swarm algorithm is used to optimize the Extreme Learning Machine,and the data is substituted into the optimized model for prediction.The average relative error of the GA-BP prediction results is 0.0490,and the average relative error of the PSO-ELM prediction results is 0.0491.The optimized models have improved the prediction accuracy.Finally,this article uses Fixed-Weight model and Convolutional Neural network to combine the optimized neural network,which is equivalent to adding a convolutional layer and substituting the data into a new model for prediction,and using the Fixed-Weight model to predict the average relative error of the result is0.0482,and the average relative error of the prediction result of the BP-ELM-GRNN model is 0.0483,and the error of the prediction result of the combined model is reduced.Analyzing the research content can be obtained,using the optimization model and the combination model for predicting can improve the accuracy of the prediction,and at last,put forward ideas for the future development prospects of the combination algorithm in other aspects.