Kolmogorov–Arnold recurrent network for short term load forecasting across diverse consumers

2025-01-12

Energy Reports

AI Summary

To address weak nonlinear modeling capability and poor cross-user generalizability in short-term electricity load forecasting, this paper proposes the first integration of the Kolmogorov–Arnold Representation Theorem into a recurrent neural network architecture. The resulting model—Kolmogorov–Arnold Network-GRU (KAN-GRU)—uniquely combines interpretable Kolmogorov–Arnold network layers, gated recurrent units (GRUs), multi-scale temporal feature extraction, and consumer-heterogeneity-aware adaptive normalization. Evaluated on eight real-world residential and commercial load datasets, KAN-GRU achieves a 23.6% reduction in mean absolute percentage error (MAPE), a 41% decrease in cross-user transfer prediction error, and a 3.2× speedup in inference latency compared to state-of-the-art baselines. By unifying theoretical interpretability with empirical generalizability, this work establishes a novel paradigm for high-accuracy, low-latency, and interpretable short-term load forecasting across heterogeneous power consumers.

Technology Category

• Machine Learning: Mixture of Experts (MoE)
• Knowledge Representation and Reasoning: Knowledge Acquisition
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Problem

• Electric Load Forecasting
• Energy Management
• Renewable Energy Integration

Innovation

• Kolmogorov-Arnold Recurrent Network
• Electricity Consumption Prediction
• Enhanced Accuracy