Weather-Independent Forecasting for State-Wide Energy Markets Using Hybrid GPI-DSSM Model

This paper introduces a novel hybrid forecasting model for state-wide energy markets, combining a Graph-based Patch Informer (GPI) and Deep State Sequential State Memory (DSSM) architecture to predict electricity prices. Integrating spatial-temporally variable weather data can introduce noise and inconsistencies, as different regions within a state experience varying weather conditions. The GPI-DSSM model avoids these issues by relying solely on demand and price data, focusing on intrinsic and hidden relationships between aggregate demand and regional reference price (RRP). By using aggregate demand as a hidden state within the Gated Recurrent Unit (GRU) based DSSM, the model captures dependencies driving price fluctuations, while the GPI transformer handles long-range temporal patterns. This hybrid architecture provides robust, accurate forecasts, demonstrating that the geo-spatial-temporal limitations of weather data on aggregate demand forecasting can be effectively managed by focusing on core market variables like demand and price.