A Hybrid Model Approach for Forecasting Electricity Demand

This paper presents the forecast of electricity consumption based on mathematical models by using the available historical consumption, with daily resolution, including upscaling, applying a hybrid model that incorporates multiple linear regression with artificial neural networks. Therefore, the hybrid model exploits both the unique features of the regression model and of the artificial neural network to determine different patterns. Thus, it is advantageous to model linear and nonlinear patterns separately using different models and then combine the forecasts to improve overall performance modelling and forecasting. The applied methodology is very reliable given that the forecast errors are close to zero, and the observed differences are justified essentially by temperature effect.