Electricity Price Prediction Framework Based on Two-Stage Time Series Decomposition

This research tackles the issue of insufficient accuracy in short-term electricity price forecasting. The novel approach combines a dual-stage signal decomposition technique using CEEMDAN and VMD, alongside a bidirectional gated cyclic unit network. The historical electricity price dataset is initially subjected to decomposition and entropy analysis. Subsequently, a KNN-driven clustering process partitions the data into distinct frequency-based signals. These separated signals then undergo further decomposition through VMD, enhancing the capture of intricate patterns. The enriched features are then channeled into a bidirectional gated cyclic unit network to facilitate comprehensive pattern learning. Rigorously evaluated using real-world US electricity data, the model exhibits a notable enhancement in predictive accuracy, showcasing its potential for practical application.