Analysis of aggregated load consumption forecasting in short, medium and long term horizons using Dynamic Mode Decomposition

Abstract

Listen

Translate article

In response to the growing need for precise energy demand forecasting in distribution grids, this study examines the effectiveness of the Dynamic Mode Decomposition (DMD) algorithm for short, medium, and long-term demand forecasting in distribution grids, leveraging diverse time series energy consumption data. The DMD algorithm, a data-driven approach renowned for its capacity to discern frequencies and trends, will be compared against established forecasting techniques, including XGBoosting Regressor, Long Short-Term Memory, and PROPHET. The research is conducted through a comprehensive case study comprising three distinct phases: study on pattern capturing of DMD, comparison in Long term horizon with all models studied and robustness study of DMD. It recalls upon real aggregated hourly energy consumption data sourced from a Spanish distribution grid. The practical applications and benefits of DMD in time series forecasting are examined carefully in order to show its comparative advantages over alternative forecasting models. To ensure a rigorous assessment of the findings, the analysis will employ four key statistical metrics: Root Mean Square Error (RMSE), Mean Squared Error (MSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). The results obtained with the case studies showed a 75% reduction on computational waste compared to other models and the results are closer to classical models, obtaining an improvement in the best case study of 44% comparing RMSE with Prophet. The insights gained can help future applications and research in the field.