K-nearest neighbors and a kernel density estimator for GEFCom2014 probabilistic wind power forecasting

Probabilistic forecasts provide quantitative information in relation to energy uncertainty, which is essential for making better decisions on the operation of power systems with an increasing penetration of wind power. On the basis of the k-nearest neighbors algorithm and a kernel density estimator method, this paper presents a general framework for the probabilistic forecasting of renewable energy generation, especially for wind power generation. It is a direct and non-parametric approach. Firstly, the k-nearest neighbors algorithm is used to find the k closest historical examples with characteristics similar to the future weather condition of wind power generation. Secondly, a novel kernel density estimator based on a logarithmic transformation and a boundary kernel is used to construct wind power predictive density based on the k closest historical examples. The effectiveness of this approach has been confirmed on the real data provided for GEFCom2014. The evaluation results show that the proposed approach can provide good quality, reliable probabilistic wind power forecasts.