Research On Application Of Wind Power Prediction Using Gray BP Neural Network

Wind power forecasting is very important for power scheduling scheme and maintenanceplan in wind farm development. The present domestic wind farm numerical weather forecastsystem is not perfect, and the wind power prediction accuracy is not high. Therefore, it isdifficult to make electric power dispatching plan and maintenance plan. Based on animproved gray GM(1,1) model and Back Propagation (BP) neural network this dissertationforecasts of the output power of wind farm.Firstly, the differences of wind power prediction method between a new wind farm andbuilt wind farms are analyzed. Then two different schemes for power prediction are studied.Scheme one is about the prediction of the power plant which have been built and put into use.The improved gray GM(1,1) model will be used to predict historical data of the built powerplant. As the input of BP neural network, the predicted data used to predict power value.Scheme two is about the prediction of new power plant. The improved gray GM(1,1) modelwill be applied to optimize the wake effect of wind speed and wind direction. The optimizedparameters as the input of BP neural network will be used for power prediction value.Secondly, the numerical approximation algorithm of gray GM(1,1) model is improved.The purpose is to solve the excessive error problem that the results of nonlinear numericalprediction use traditional gray GM(1,1) model. The optimization of GM(1,1) model is toimprove the prediction accuracy of power. By deduction and improvement,using theoptimized GM(1,1) model to predict the weather data of wind speed, wind direction and airtemperature, the result error decrease. The wind speed relative prediction error reduces34.3%,wind direction predicted less than1%relative error accounted for98.6%of the total numberof samples, the environment temperature of less than10%of the predicted relative error of thetotal samples82.5%. The prediction effect is relatively good.Thirdly, the two schemes of BP neural network power prediction results are analyzed. Aroot mean square error of prediction results of single unit is16.84%and the wind farm is18.22%. Two forecasting results root mean square error of single units of19.74%and thewhole wind farm is24.08%. Calculation shows that except for the entire wind powerprediction error is more than20%in scheme two, other predictive effect is satisfied. It provedthe feasibility of the proposed scheme.Finally, the prediction error is analyzed. Aiming at the causes of errors, put forwardsome requirements and recommendations for data collectionabout wind farm.