Multivariate Peak Demand Forecasting Using Fractal-Fuzzy Feature Selection

Optimizing feature selection to capture temporal dynamics and seasonal variations enhances the accuracy of multivariate load forecasting models. Careful consideration of factors like temperature fluctuations effectively models load behavior and improves forecasting robustness. The paper uses a multivariate forecasting model with fractal-fuzzy feature selection to enhance predictive accuracy by focusing on key and influential features. Performance is evaluated through statistical techniques, including paired t-tests and one-way ANOVA, to test null hypotheses regarding model performance differences before and after feature selection. We applied our method using multivariate data from CEBECO and PAGASA to forecast peak energy demand based on meteorological factors. The temperature-related variables achieved an <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$F F C C \approx 0.99$</tex>, indicating their strong influence on demand variability. As the parameter <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\alpha$</tex> increases, the importance of temperature features grows, while the relevance of rainfall and wind speed decreases. The paper shows that using selected features improves computational efficiency while maintaining accuracy. Feature selection reduced average training times from 25.58 seconds to about 5 to 10 seconds, significantly enhancing processing speed compared to models with the full feature set. Experimental results demonstrate that the fractal-fuzzy feature selection approach improves the predictive accuracy of forecasting models. Key environmental variables, particularly temperature, significantly enhance peak electricity demand predictions. The paper explores how weather affects energy consumption and suggests that better forecasting models can improve energy management by integrating fractal and fuzzy logic to select external factors to include in the forecasts.