[Nonlinear Integrated Short-term Carbon Emission Projections Based on Bimodal Decomposition]

Due to the high volatility of short-term carbon emission data， a prediction model based on two-mode decomposition and nonlinear integration is proposed to improve the prediction effect of short-term carbon emission. Firstly， CEEMDAN and RIME-VMD decompositions are used to obtain two different multi-scale components， then three different prediction models， namely， LSTM， GRU， and KELM， are used to predict the two components， and finally the final prediction results are obtained by using the XGBoost ensemble of six models. The results show that bimodal decomposition can better reflect the data information and realize the complementary advantages between different decomposition methods and that nonlinear integration can effectively synthesize the advantages of different prediction models and improve prediction accuracy. The short-term carbon emission prediction model based on dual-mode decomposition and nonlinear integration can effectively extract and utilize the information in data to improve prediction accuracy.