A neural network-based method for forecasting zonal locational marginal prices

Abstract

To direct market participants to make efficient use of both transmission and generation resources, energy for both day-ahead and real-time markets is priced by locational marginal price (LMP). Good LMP forecasting will help market participants make effective decisions when preparing offers and bids and making bilateral contracts. LMP forecasting, however, is difficult since LMPs are closely affected by complicated market behaviors and depend heavily on transmission congestions. This paper presents a neural network-based method for forecasting zonal LMPs by using the decoupled extend Kalman filter (DEKF) with UD factorization and sequential updating. Congestion components are emphasized by forecasting differences between zonal LMPs and hub LMPs, and the hub LMPs themselves are forecasted by their logarithm to better reflect market behaviors. A method is also developed to capture system transmission outages. To overcome the unavailability of zonal loads, a method is developed to predict zonal loads based on the total load. Two neural networks are developed to forecast real-time LMPs before and after the day-ahead market is cleared. Testing results with data from the PJM and New England markets show that the method is computationally efficient and provides accurate LMP predictions and confidence intervals. Testing at New England market shows the method requires a small set of training data.