CC-ELM: a compression-coefficient-based extreme learning machine for enhanced load forecasting in electric vehicle charging stations

Accurate load forecasting for Electric Vehicle (EV) charging stations is essential for maintaining grid stability and enabling efficient demand-side management. However, existing approaches, ranging from traditional statistical models to deep learning architectures, often suffer from limitations such as heavy reliance on manual feature engineering, high computational complexity, and poor generalization when data are limited. To overcome these challenges, this paper proposes the Compression-Coefficient-based Extreme Learning Machine (CC-ELM). The core innovation lies in introducing a dynamic compression coefficient that selectively compresses and enhances the input feature space before training, effectively mitigating the redundancy caused by random hidden node initialization in standard ELM. This process improves feature representation while retaining ELM’s hallmark training efficiency. Comprehensive simulations using real-world EV charging data demonstrate that CC-ELM significantly outperforms seven benchmark models, including I-ELM, HELM, SVM, CNN, LSTM, Attention-LSTM, and Transformer. These results validate CC-ELM as a robust, efficient, and highly accurate solution for EV charging load forecasting, offering a practical tool for real-world energy information systems.