Forecasting real-time net interchange of electric power

Regional Transmission Organizations (RTOs) or Independent System Operators (ISOs) exchange electric power with their neighboring RTOs or ISOs to improve reliability and efficiency of grid operations. For an RTO/ISO, actual net interchange could deviate from the scheduled one and the difference is called inadvertent interchange. Accurate forecasts of actual net interchange will enable RTOs or ISOs to minimize the amount of inadvertent interchange and consequently reduce operational costs. The task compared with load forecasting, is difficult because actual net interchange is volatile and there are a lot of relevant factors to be considered such as demand uncertainties and price differentials. In this paper, features of actual net interchange and relevant factors are explicitly examined to identify the key factors through spectral analysis and correlation analysis. Wavelet decomposition is applied to decompose the data into components at different frequencies. Neural networks are used to forecast each frequency component individually, and the results are then combined as the final forecasts. This method is compared with persistence and linear regression models to evaluate the performance. Numerical testing based on ISO New England and PJM´s data shows that our method provides the best predictions.