Multi-Day-ahead Net Interchange Schedule Forecast based on LSTM Recurrent Neural Network

Maintaining an accurate net interchange forecast is an important task for independent system operators (ISOs) because it helps reduce the real-time operational cost. The task is challenging due to the high volatility of the interchange. In this paper, a deep long-short-term-memory (LSTM) recurrent neural network is applied as an efficient net interchange forecast tool, which can extract temporal features from various internal and external factors and generate forecasted value with considerable accuracy. In addition, a rolling forecast framework is designed for the LSTM to conduct a multiple day-ahead net interchange forecast to better prepare the ISOs for potential future uncertainties. The feasibility of the proposed deep learning method for practical applications is verified with simulation results based on actual interchange data from New England area and comparison with the current industry method.