Daily Load Gurve Forecasting with Corrective Mechanism by Multi-Layered Neural Network

Abstract

In electric power systems, to improve economy and reliability, it is necessary to forecast the following day's load curve accurately.
Load forecasting models based on statistical techniques require a great deal of effort to analyse the complex relationship between load and its determining factors. Furthermore, because such forecasting models are based on averaged values, a large weather variation may result in a large error.
In this paper, to overcome these issues, a load forecasting model combining an artificial neural network (ANN) with multiple regression analysis is proposed. Therefore, this model is able to reflect nonlinear relationships between inputs (temperature, cloudiness and general weather conditions) and outputs (forecast-ed load values) without analytical effort. To improve the learning characteristics, “dead bands” are introduced into the error values and the computed outputs. Furthermore, various techniques are adopted to overcome problems associated with the fundamental back-propagation method.
From the simulation results, the effectiveness of the proposed correcting model is confirmed; the dead bands performed stable correction and considerable improvement in accuracy was achieved for the summer period which is normally inferior to other periods.