<p>Precise carbon price prediction is significant in providing early warning information on abnormal fluctuations in the carbon emissions trading market. However, it is difficult to predict accurately since its nonlinear and fluctuation. Therefore, many scholars hope to decompose volatile series into smoother subseries before prediction. Actually, many studies on carbon price prediction have proven the excellence of decomposition and reconstruction models. However, some studies use a one decomposition framework that cannot completely decompose and some use a quadratic decomposition framework which lead to over-decomposition. Thus, we propose a new decomposition and reconstruction framework. In this framework, we introduce Empirical Wavelet Transform (EWT) to decompose the original sequence into high-frequency, low-frequency and residuals. Then, the high-frequency sequence is decomposed twice using variational modal decomposition (VMD). Additionally, we introduce fuzzy entropy (FE) to help recombine the subsequences. Finally, we use Gated recurrent unit (GRU) to predict the reconstructed sequences and receive the result. Moreover, a case study and two comparative analyses are constructed using carbon price data collected from Guangdong, Beijing and Shanghai. In Guangdong datasets, the MAPE, R2 and RMSE of the model are 0.0152, 99.50% and 1.2633. In the other datasets, the model also performs superiorly, which shows its robustness.</p>