Short-term load forecasting using recurrent neutral networks

This paper has proposed the prediction of hourly load in power systems through two recurrent neutral networks, one based on multilayer perception models with a second order learning rule and the second on radial basis function network. Two different formulations of load forecasting have been simulated, one to predict hourly load based on historical data and the second to predict peak and valley loads of a particular day type and then forecasting the hourly load using a normalized load curve of that day type. The results obtained on a practical Indian system data demonstrate that the second approach based on prediction of peak and valley load for a day type along with radial basis function network model provides more accurate forecast of the load.