Short-Term Forecasting of Wind Speed and Wind Power Based on BP and Adaboost_BP

Abstract

Due to wind is intermittent and less dispatchable, wind power fluctuates as the wind fluctuating and is uncontrollable. Therefore, when wind power accounts for a higher proportion of total electricity generation of the system, power generation plan needs to arrange according to the variation of wind power output. The way to solve the problem is forecasting the wind power. In this paper, we focus on the wind speed and wind power forecasting in the time scales of 10min, 1h and 3h in the future. BP Neural Network and Adaboost_BP Neural Network are selected as the forecasting model for wind speed forecasting. And for wind power forecasting model, we use BP Neural Network as the method. The wind farm which the data used in this paper comes from has 31 wind turbines in the same type. As the geographical distribution of this wind farm is unknown, we pick one No.17 wind turbine as the optimal one by analyzing the data of 31 wind turbines and making the curve fitting of wind speed and wind power. Then we analyze the influencing factor of wind power, and find out wind speed as the most influential factor. For the wind speed, we deal with the raw data which selected from SCADA (Supervisory Control and Data Acquisition) system before using it. For the wind speed forecasting model, we find the optimal number of the training data for each training sets for the BP Neural Network in each time scale. Then we make a contrast of the accuracy of the single-step forecasting accuracy between BP Neural Network model and Adaboost_BP model in 10min, 1h and 3h at their respectively optimal number of the training data. And there is a comparison between the accuracy of the single-step and iterative multi-step wind speed forecasting model in 1h and 3h time scales at the number of the training data of 10 for the two models. For the wind power forecasting model, we use the forecasting wind speed and its corresponding wind power to build the input matrix for the network training. And we find that not only the number of the training data for each training sets but also the range of wind power affects the forecasting accuracy. Then make a contrast of the the wind power forecasting result with forecasting wind speed and actual wind speed as the input.