Self-adaptive evolutionary neural networks for high-precision short-term electric load forecasting

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Computational Science

Background:

• Short-term electric load forecasting (STLF) is critical for power system stability and efficiency.
• Existing models like Multi-Layer Perceptrons (MLP) face challenges in capturing nonlinear load dynamics and maintaining computational efficiency.

Purpose of the Study:

• To introduce an optimized forecasting framework, the Self-Adaptive Differential Evolution-Kolmogorov-Arnold Network (SADE-KAN), enhancing both predictive accuracy and computational efficiency for STLF.
• To address the limitations of conventional models in handling complex, nonlinear load variations.

Main Methods:

• Developed SADE-KAN by integrating Kolmogorov-Arnold Networks (KAN) with Self-Adaptive Differential Evolution (SADE).
• KAN utilizes spline-based learnable activation functions for greater flexibility in capturing temporal dependencies.
• SADE dynamically optimizes KAN's hyperparameters for a balance between accuracy, complexity, and training efficiency.
• Validated the model on ISO-NE hourly load data (2019-2023) across multiple forecasting horizons (24-168 hours).

Main Results:

• SADE-KAN demonstrated significant improvements, reducing Mean Absolute Percentage Error (MAPE) by up to 35% and Root Mean Squared Error (RMSE) by 38% compared to MLP models.
• The model required 35% fewer learnable parameters than MLP.
• SADE-KAN showed enhanced generalization and robustness, effectively capturing rapid load fluctuations.
• Achieved superior performance over MLP, conventional KAN, and other advanced models.

Conclusions:

• SADE-KAN provides a computationally efficient and highly accurate forecasting framework for STLF.
• The model offers a robust solution for real-time power system applications, demand response, and energy market operations.
• The integration of SADE with KAN presents a promising advancement in forecasting technology.