Transfer learning based CEEMDAN-VMD secondary decomposition and multi-scale modeling for short-term electricity market price trend forecasting.

This paper proposes a short-term electricity market price trend forecasting framework based on transfer learning, which centers on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN)- Variational Mode Decomposition (VMD) secondary decomposition and multi-scale modeling. The framework integrates Attention mechanism, multi-scale Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM), and is named as CEEMDAN-VMD-MultiScale-Att-CNN-BiLSTM (Attention-Convolutional Neural Network-Bidirectional Long Short-Term Memory) to highlight its core innovative characteristics. Firstly, the CEEMDAN and VMD methods are employed to perform quadratic decomposition on the original electricity market price sequence, reducing the intricacy of forecasting; Then, a basic forecasting model is constructed that includes Attention layers, multi-scale CNN layers, and BiLSTM layers, and the model parameters are optimized through hierarchical transfer learning; Finally, the optimal parameter model serves to forecast the electricity market price sequences of each frequency, and the final price trend prediction value is reconstructed. The experimental observation results show that the multi-scale Att-CNN-BiLSTM algorithm achieves consistent performance improvements supported by statistical tests in the test dataset and actual scenario experiments, especially in predicting the jump points and peak points of electricity market prices with higher accuracy. The core innovation of this framework lies in the construction of a CEEMDAN-VMD quadratic decomposition strategy for non-stationary electricity price sequences and a multi-scale feature modeling mechanism, effectively overcoming the limitations of traditional models in reflecting the unique features of electricity price data.