Short-Term Load Forecast in Power System Using Neural Network

Modern power systems are expanding and becoming more sustainable, but the increased use of renewables, electric vehicles, and smart grids unavoidably adds another layer of system complexity and uncertainty. Due to the lack of efficient energy storage in the power distribution and transmission system, an optimal power system requires electrical generation to match electric load demand. There is the need to implement methods/strategies for estimating electrical demand to properly plan for an effective power distribution, which fulfils operating requirements i.e. meeting demand at minimal cost of operation. Load forecasting methods are categorized as follows, very short-term (one minute to one hour), short-term (one hour to one week), medium-term (two weeks to a few months), and long-term (a few months to a couple of years). This study focuses on short-term forecasting, where the prediction time allows for timely scheduling, system operations evaluation, and system control. Short-term forecasting can be challenging in general because load demand is dynamic and depends on non-linear factors such as weather and appliance type, which generate considerable fluctuations in aggregated load. This work employs a feedforward neural network (FNN) based-model with seven input variables for short-term load forecasting. In comparison with the linear regression method, FNN-based model produced good prediction with a minimum absolute percentage error (MAPE) of 3.91%.