پیشبینی نیازهای غیرمستقیم الکتریکی در ساختمانهای تجاری پیشبینی روزانه: بر اساس رگرسیون بردار حمایتهای ترکیبی

Accurate and highly-generalized forecasting models of hourly electric power demand are in urgent need for buildings, as to be the basis of operation management and bottom-up regional energy forecasting. Combined with multi-resolution wavelet decomposition (MWD), a hybrid support vector regression model was applied in a non-stationary operated hotel to predict the hourly electric power. With 15-dimensional parameters of 29 clustered days as the training sample, a nonlinear SVR model was carried out. Relative errors (RE) with and without MWD were compared at different ɛ-non-insensitive values. Results show that the MWD processing can reduce the deviations slightly only when ɛ is higher than 0.1, and the optimal daily mean RE of a typical day is around 5.6%. This paper aims to offer engineers and planners a feasible method for energy prediction based on the historical meter readings.