Probabilistic Load Prediction with Risk-Severity Score

Emerging interest in probabilistic load forecasting (PLF) has increased because it can address load uncertainty better than point forecasting. This paper proposed a new strategy to treat the input of PLF as a probability map, which enables an actual pictorial to depict the prediction outcome risk of a load demand profile according to the preferred length of the prediction horizon. The proposed probabilistic method consists of optimal signal decomposition, prediction model construction and an evaluation strategy to obtain the prediction interval risk outcome. Whale optimization algorithm-discrete wavelet transformation (WOA-DWT) is employed to detect optimal input decomposition. The quantile regression random forest (QRRF) is utilized to build a prediction interval from various quantiles. A novel scheme to calculate the prediction outcome risk due to current input, which provides a degree of confidence, is also presented. The efficacy of prediction interval is verified using the Independent System Operator - New England (ISO-NE) set with the Winkler score and proposed risk-severity score.