Wind Power Forecasting Based On Deep Belief Network And Multiple Linear Regression

Wind power as a high quality of renewable resources,the specific gravity of the resources system in China has augmented gradually for the past few years.The larger volatility in the wind to the safe and stable operation of power grids bring great difficulty,accurate prediction of wind turbine power output reduces the disadvantage of wind power,better suited to gradually augment the specific gravity of wind power grid,power system can advance prevention,timely scheduling resources.In view of the poor power curve fitting and the low accuracy of wind power prediction,a deep learning model combining deep confidence network(DBN)and multiple linear regression(MLR)is proposed in this paper to forecast wind power.Firstly,aiming at the problem of missing and deviation of daily statistical data of wind turbines,the bad data were deleted considering the existing interval of theoretical wind speed,wind direction and power data,and then the missing data were supplemented and completed by relation coefficient matrix method.In view of the error increase caused by the dimensionless inconsistency of a variety of data,z-score standardization is used to process the data of different types and input the prediction model after unified processing.Then,for the power curve does not complete reflect the various factors influencing the power,the power to predict the limitations,in this paper,at the same time considering the influence of wind speed,wind direction and other data on power,using multivariate linear regression algorithm to construct including wind speed,wind direction and power of the best relation equation,the search for the best fitting solution,solves the limitations of a single wind speed prediction power.Finally,the best multiple linear regression equation as the top of a deep belief network building wind power prediction model,this model to training and supervision of fine-tuning process of hierarchical search and update the best values of model parameters,the preliminary training of deep belief networks and at the top of the join multiple linear regression,the best solution,build the prediction model under supervision and fine tuning the parameters best,to predict a certain wind speed,wind direction and power,through 30 seconds and two groups of sampling time,10 minutes forecast wind field data examples,the results confirmed that the model has better fitting effect and prediction accuracy.