Statistical Load Forecasting Using Optimal Quantile Regression Random Forest and Risk Assessment Index

To support daily operation of smart grid, the stochastic load behavior is analyzed by a day-ahead prediction interval (PI) which is built from predictor's probability density function, computed in statistical mean-variance, and achieves a symmetrical PI. However, this approach lacks for intended risk information on the predictors' uncertainty, e.g., weather condition and load variation. This article proposes a novel statistical load forecasting (SLF) using quantile regression random forest (QRRF), probability map, and risk assessment index (RAI) to obtain the actual pictorial of the outcome risk of load demand profile. To know the actual load condition, the proposed SLF is built considering accurate point forecasting results, and the QRRF establishes the PI from various quantiles. To correlate the uncertainty of external factors to the actual load, the probability map computes the most probable quantile happening in the training horizon. Based on the current inputs, the RAI calculates the PI's intended risk. The proposed SLF is verified by Independent System Operator-New England data, compared to benchmark algorithms and Winkler score. The results show that the proposed method can model a more precise load PI along with the risk evaluation, as compared to results of the existing benchmark models.