Synergistic Optimization and Risk Control: Integrated Price Forecasting Models

This paper presents a hybrid intraday electricity price forecasting model—Info-VMD-iTransformer-CNN-LSTM—tailored for high-dimensional, non-stationary price series. First, variational mode decomposition (VMD) adaptively separates price signals into intrinsic modes, mitigating mode mixing and noise. Next, an improved Transformer (iTransformer) with enhanced positional encoding captures long-range dependencies, while CNN layers extract local spatio-temporal features and LSTM units model sequential dynamics. Finally, the INFO algorithm automates hyperparameter optimization, ensuring both high accuracy and robustness. Empirical evaluations demonstrate that our approach consistently outperforms existing benchmarks under volatile market conditions, making it well suited for real-time forecasting in modern power systems.