Stochastic hourly load forecasting for smart grids in Korea using NARX model

Smart Grids - intelligent electricity grids enable two way communication between utility and its customers. Smart Grid technologies are being deployed throughout the world, to make the traditional grids more reliable, secure and environmentally sustainable. Consumers are the important domain in a Smart Grid, participating in real-time demand response programs. To ensure the qualities of a Smart Grid and to affirm a smooth operation of the grids we need an accurate forecasting model. Accurate load forecasting enables a service provider to plan and balance the electricity demand and supply of customers. South Korea, is one of the countries, deploying Smart grids in an active phase. They have made commendable progress by establishing test beds in Jeju Island, and implementing Demand Response (DR) programs. In this paper we have analyzed and proposed a Short Term Load Forecasting (STLF) model pertaining to Smart grids in Korea. We have developed a STLF model using Non-Linear Auto Regressive with exogenous inputs (NARX) method in Artificial Neural Networks (ANN) and forecasted the day-ahead electricity need for an year by hourly-granularity. This model forecasts the day-ahead electricity need, taking in the temperature and seasonal variables as exogenous inputs. The STLF model is simulated with various training algorithms and the least error rate of ~2% is achieved, thereby ensuring a better forecasting solution.