Abstract
This paper presents a daily load curve forecasting method using hourly regressions. Electric load varies significantly during the day. Many factors, such as weather conditions, hours and day-types, relate to the load behavior. We formulate a daily load curve in set of independent 24-h regression equations that split hourly load into four parts: weather related load (WE), weekday's day-types (DTW), non-weekday's day-types (DTH), and yearly trends (TR). We incorporate temperature, humidity and insolation into the WE part. Non-linear relationships of weather factors and loads are formulated by polynomial functions. Another approach, based on Gaussian functions, is also applied to the modeling for the weather and load relationships. In order to estimate regression coefficients properly, we should consider seasonal load change and estimate the coefficients with statistically enough amounts of data. The proposed method estimates the equations with data from both forecasting year and past years. TR compensates yearly load difference among the data. As TR representations, we propose ‘additive trends model’ and ‘multiple trends model’. Experimental studies on the next day load forecasting are carried out with TEPCO system load. The results indicate effectiveness of (1) combination of ‘polynomial function’ and ‘multiple trends model’ and (2) ‘Day-types’ and ‘Insolation’; in the next day load curve forecasting. Performance of less than 1% MAPE is also observed on the next day weekdays' peak load forecasting.
