Neural Network based short-term electricity demand forecast for Australian states

Load forecasting is an essential part of power system planning. This helps in proper and economic operation of power system plant. The problem of accurate load forecasting has become more baffling in current scenario since deregulation came into picture and industrial growth is positive. This brought highly non-linear electric load profile that is difficult to forecast. Artificial Intelligence (AI) based forecasting models are always preferred over conventional models, because it is heuristic and adaptive approach to learn. In this paper, AI based Feed Forward Neural Network (FFNN) is used to evolve half-hourly load forecasting model for a period of three days. This model is tested and validated on five Australian states' load data. Out of these five cases, forecasting accuracy in four cases is confined to 2.7233%; amongst these four cases two show the forecasting error less than 2%. In one case forecasting error is 6.4593%, which is acceptable because the state consumes lesser energy. This shows the appropriateness of the FFNN forecasting model for multi-regional load forecasting.