An ensemble approach for forecasting net interchange schedule

The net interchange schedule (NIS) is the sum of the transactions (MW) between an Independent System Operator/Regional Transmission Organization (ISO/RTO) and its neighbors. Effective forecasting of the submitted NIS can improve grid operation efficiency. This paper applies a Bayesian model averaging (BMA) technique to forecast submitted NIS. As an ensemble approach, the BMA method aggregates different forecasting models in order to improve forecasting accuracy and consistency. In this study, the BMA method is compared to two alternative approaches: a stepwise regression method and an artificial neural network (ANN) trained for NIS forecasting. In our comparative analysis, we use field measurement data from the PJM Interconnection RTO to train and test each method. Our preliminary results indicate that ensemble-based methods can provide more accurate and consistent NIS forecasts in comparison to non-ensemble alternate methods.