Short-term electricity consumption forecasting for low-altitude economy enterprises: A metaheuristic-optimized efficient robust neural network approach

Résumé fourni par la source

Accurate short-term electricity consumption forecasting for low-altitude economy enterprises remains challenging because of nonlinear dynamics, heterogeneous operating patterns, and noise-contaminated measurements. To address these issues, an efficient and robust forecasting framework, termed IMPA-RTBELM, is proposed by integrating an Improved Marine Predators Algorithm (IMPA) with a Regularized Twin-Boundary Extreme Learning Machine (RTBELM). Specifically, IMPA enhances the original Marine Predators Algorithm through Sobol sequence initialization to improve population diversity, dynamic opposition-based learning to alleviate premature stagnation, and pattern-search refinement to strengthen late-stage exploitation. RTBELM retains the closed-form training advantage of ELM while improving robustness via ridge-stabilized output learning and a twin-boundary clipping mechanism that suppresses impulsive prediction excursions. Benchmark-function experiments show that IMPA achieves faster convergence and better final solutions than the original MPA and several representative metaheuristics. Under four mixed-noise SinC regression settings, IMPA-RTBELM consistently improves generalization performance and substantially reduces forecasting errors, lowering the average test MAPE from 0.783% to 0.292% relative to RTBELM, while also reducing test MAE and RMSE by 12.98% and 12.50%, respectively. On real enterprise electricity consumption data, IMPA-RTBELM outperforms a broad range of statistical, machine-learning, and deep-learning baselines, achieving an RMSE of 127.0983, an MAE of 93.3518, and a MAPE of 0.78%. These results demonstrate that the proposed framework provides a favorable trade-off among accuracy, robustness, and efficiency for large-scale short-term electricity consumption forecasting.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.