An improved combination approach based on Adaboost algorithm for wind speed time series forecasting

As one of promising renewable energy, wind energy has an important position in the field of energy market and it has been always used to generate electricity, thus, the high precision wind speed forecasting is important but challenging for power generation system. Many researchers have devoted their attention to establish efficient wind speed forecasting models, limited by the structure of original models and database, these models could not solve overfitting problem very well and it is rare that a single wind speed forecasting model is always best in all cases since each model has its own particular strengths and weaknesses. The combining method, composed of multiple forecasting models, is regarded as a type of outstanding approach to take advantage of strengths of each model. However, the properties of the individual forecasting models may vary over time, which leads to poorly performance of the combining method using fixed weights, thus, it is more appropriate to allow the combining weights to change according to the time-varying underlying process. This study develops a reliable combination model for wind speed forecasting based on an improved Adaboost algorithm named time-vary-forecasting-effectiveness (TW-FE-Adaboost) algorithm in order to improve the overall forecasting accuracy. In the proposed model, concept drift is firstly used to deal with wind speed time series mainly because the contributions of samples which are varying with time, the second order forecasting effectiveness is used to measure the performance of weak learners. Then, the multi-step ahead forecasting for each site is conducted using TW-FE-Adaboost model in which the input-output sample pairs are determined in a reasonable way. Finally, the ultimate forecast result of wind speed is obtained by aggregating the forecast result of each weak learner. The proposed model is tested using four sites wind speed series collected in Hexi corridor from wind farms located in northwest of China. The experimental results show that the proposed model outperforms all other comparison models in this paper, which demonstrates that the proposed model has superior performances for wind speed forecasting.