A wind speed interval prediction system based on multi-objective optimization for machine learning method

• Published in: Applied energy Vol. 228; pp. 2207 - 2220
• Main Authors: Li, Ranran, Jin, Yu
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 15.10.2018
• Subjects: Feature selection; Least squares support vector machines; Multi-objective optimization; Prediction intervals; Wind speed forecasting; Prediction intervals; Feature selection; Multi-objective optimization; Least squares support vector machines; Wind speed forecasting
• ISSN: 0306-2619; 1872-9118
• DOI: 10.1016/j.apenergy.2018.07.032

•Novel wind speed interval forecasting approach in multi-objective formulation introduced.•Hybrid framework building on data feature selection method.•Simultaneously the lower and upper bounds of the prediction intervals of future wind speed time series constructed.•The best compromise solution selected by the smallest coverage width criterion method. Accurate forecast of wind speed is the first prerequisite to supply high quality power energy to customer in a secure and economic manner. However, traditional point forecast may not be sufficiently reliable and accurate for decision-makers to perform operational strategies purely when the uncertainty level increases. For the sake of quantifying the uncertainty associated with point predictions, it is necessary to conduct interval prediction to provide reliable and accurate wind speed information. In this work, a hybrid model framework based on combinatorial modules was proposed and successfully adopted to construct the prediction intervals of the future wind speed. Feature selection methods are developed to determine the most suitable modes of original time series and the optimal input form of the model, while the optimization forecasting module is applied to model the wind speed series based on the machine learning method and the multi-objective optimization algorithm, then the compromise solution of Pareto front is chosen by “Min-max” method. Finally, the proposed combined model was investigated via the hourly wind speed data from two different periods in Penglai, China. Besides, the study’s experimental results indicated that the prediction intervals generated perform well and are satisfactory in both criterion functions of high coverage and small width through discussion among single-objective models and other multi-objective models (signal pre-processing method comparison included).