A New Calendar Effect and Weather Conditions based Day-ahead Load Forecasting Model

Abstract

Energy demand forecasting performs a key role in the planning and operation of day-ahead electricity market and resource adequacy assessment, both of which heavily depend on weather and calendar conditions. We have developed a regional short-term hourly load model which considers both weather and calendar effects on electricity consumption in a given day. Instead of using nature calendar months as factors in the load model, consecutive days with similar regional weather conditions and temporal electricity consumption patterns have been grouped in the same segment by a two-stage clustering process upon k-means clustering algorithm. Component-wise gradient boosting framework with penalized regression smooth splines is then applied to estimate hourly load model within each segment. In an instance derived from an Independent System Operator in the U.S., the preliminary numerical results of up to 24-hour ahead forecasting through a year present more accurate hourly load forecasts.