Locational Marginal Price Forecasting Using Convolutional Long-Short Term Memory-Based Generative Adversarial Network

In wholesale electricity markets, locational marginal prices (LMPs) are strongly spatio-temporal correlated. Most previous data-driven studies on LMP forecasting only leveraged temporal correlations among historical LMPs, very few of them learned the spatial correlations to improve forecasting accuracy. In this paper, a convolutional long-short term memory (CLSTM)-based generative adversarial network (GAN) is proposed to forecast LMPs from market participants' perspective. Historical LMPs of different price nodes are organized into a 3-dimensional (3D) tensor which stores the spatio-temporal correlations among LMPs. The LMP forecasting problem is formulated as a spatiotemporal sequence-to-sequence forecasting problem. The proposed approach is verified through case studies using public historical LMPs from Midcontinent Independent System Operator (MISO) and ISO-New England (ISO-NE), in comparison with other state-of-the-art LMP prediction approaches.