An artificial neural network application to weekly electrical peak load forecasting

Abstract EN

Artificial neural network (ANN) techniques have been applied to various subjects in the electrical power system area.

In this paper, a neural network approach is applied to the weekly load forecast problem.

Two neural network (NN) models for the prediction of the seven-day ahead peak load of an electric utility in the region have been developed.

The development was based on a nonlinear autoregressive (AR) time series modeling using simple Backpropagation NNs architecture.

These NN models were trained using two weeks data window for the most recent daily peak loads.

The first model treated the peak load at weekdays and weekends separately.

The second model treated the peak load at weekdays and weekends together with the introduction of indicator variables for identifying weekdays from weekends and holidays.

The generated forecasts from the two developed models showed that the second model gives more accurate forecasts for the tested utility data.

The average absolute percent error (AAPE) for the generated forecasts using this model was 2%.