Short-term power load probability density forecasting based on Yeo-Johnson transformation quantile regression and Gaussian kernel function

Penetration of renewable resources into power systems, such as wind and solar power, has significantly grown the complexity and level of uncertainty in both power generation and demand sides, which are highly desirable to exploit more advanced methods to address the uncertainties. The probability density forecasting method using quantile regression can describe probability distributions of future power load. However, existing quantile regression probability density forecasting methods may encounter embarrassing cross phenomenon, affecting the effectiveness of probability density forecasting. To avoid the crossing issue, this study proposes a probability density forecasting method based on Yeo-Johnson transformation quantile regression using Gaussian kernel function. Gaussian kernel density estimation using a rule of thumb bandwidth is innovatively hybridized with Yeo-Johnson transformation quantile regression for short-term power load probability density forecasting. The evaluation metrics for forecasting errors and prediction interval are adopted to carry out a comprehensive study on load uncertainty handling. One-hourly historical load data of August 2014 in summer and December 2014 in winter from Ottawa, Canada are used to evaluate the performance of proposed model. The results show that the proposed method not only efficiently avoids the quantile crossing problem but also obtains smooth probability density curves and more accurate forecasting results.