An approach for Short-Term Load Forecasting Using RBFNN in the Smart Grid

Accurate demand forecasts are important for managing energy efficiently in electric grids. However, building models for demand forecasting is a challenging task as it depends on numerous factors that are both intrinsic and external to the grid. Furthermore, these factors are time-varying and non-linear as well. This makes demand forecasting a cumbersome task. This investigation proposes a simple model for short-term load forecasting in smart grids that uses the radial basis function neural networks and time-series data to provide short-term demand forecasts. The proposed method is simple as it just uses the time-series to data in forecasting. The use of RBFNN is motivated by its ability to model non-linear and time-varying entities. The modelling approach consists of four phases: data-collection, pre-processing, modelling and validation. The time-series data of the demand is collected during the data-collection phase. As the data obtained has numerous outliers, missing information and bad-data, the pre-processing approaches are used to eliminate them. The modelling step consists of tuning the RBFNN and finally in the validation step, the forecasting model is validated using test data-set. The proposed forecasting approach is illustrated using data obtained from Delhi load dispatch centre and an Australian energy grid data. Our results show that the proposed approach provides reasonably accurate forecasts and is simple compared to existing methods in literature.