Multi-Location and Multi-Feature LMP Forecasting: A 2D Spatiotemporal LSTM-CNN Approach

Accurate Locational Marginal Price (LMP) forecasting is crucial for effective energy procurement in electricity markets, as it helps utilities make informed, cost-efficient decisions. This study addresses the limitations of traditional models by introducing a 2D spatiotemporal LSTM-CNN model that utilizes multi-location and multi-feature inputs to improve forecast accuracy specifically for Batangas I Electric Cooperative, Inc. (BATELEC I). Using the Dynamic Interval Price Collection (DIPC) dataset from the Independent Electricity Market Operator of the Philippines (IEMOP), which provides 5-minute interval data from June 2021 to August 2024, the model captures short-term price fluctuations across resource locations. Results indicate that the 2D Multi-Features and Multi-Locations LSTM-CNN model outperforms 1D and 3D configurations, achieving the lowest RMSE of 5.0976E-03 and effectively capturing spatial and temporal dependencies. These findings support improved decision-making in electricity markets, with future applications in real-time adaptive forecasting for enhanced market responsiveness.