基于二次分解策略和BiLSTM的短期碳排放预测模型设计

Design of short⁃term carbon emission prediction model based on secondary decomposition strategy and BiLSTM

针对现有短期碳排放预测模型残余噪声大、忽略全局信息的特性导致预测精度不高的问题,提出一种基于二次分解策略和双向长短期记忆神经网络(BiLSTM)的新的短期碳排放预测模型。利用改进的自适应噪声完全集成经验模态分解(ICEEMDAN)方法和二次分解思想,将原始时间序列分解为多个本征模态函数(imfs);利用鲸鱼优化算法(WOA)优化的双向长短期记忆神经网络(BiLSTM)对所有函数序列进行预测,并将每个函数序列的预测值累加得到最终结果。实验结果显示,该文提出模型的R2达到0.999,MAPE和RMSE分别为1.3×10-3和97.4,优于其他对比模型,有效降低了预测误差。

Aiming at the problem that the existing short⁃term carbon emission prediction models have the characteristics of large residual noise and ignoring global information,which leads to low prediction accuracy,a new short⁃term carbon emission prediction model based on secondary decomposition strategy and Bi⁃directional Long Short⁃Term Memory(BiLSTM)was proposed.The Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(ICEEMDAN)and the idea of secondary decomposition were used to decompose the original time series into multiple intrinsic mode functions(imfs).BiLSTM optimized by the Whale Optimization Algorithm(WOA)was used to predict all function sequences,and the predicted value of each function sequence was accumulated to obtain the final result.The experimental results show that the R2 of the proposed model reaches 0.999,the MAPE and RMSE are 1.3×10-3 and 97.4,respectively,which are better than other comparison models and effectively reduce the prediction error.