“碳达峰”目标下中国碳排放强度预测——基于LSTM和ARIMA-BP模型的分析

Prediction of China’s Carbon Emission Intensity Under the Goal of Carbon Peak——Analysis Based on LSTM and ARIMA-BP Model

本文基于LSTM神经网络模型并在一定的经济增长预期下推导预测出我国碳排放强度变化趋势,同时,建立ARIMA-BP神经网络模型作为验证模型对碳排放强度进行直接预测。研究结论为:(1)LSTM神经网络模型在验证集上的均方误差(MSE)为0.00001,平均绝对百分比误差(MAPE)为0.33%,表明模型泛化能力十分优秀,在LSTM神经网络模型预测框架下,中国碳排放强度将在2030年达到0.9237吨/万元,相较于2005年的碳排放强度2.9755吨/万元下降68.96%;(2)在ARIMA-BP神经网络模型的预测分析中,预估中国2030年碳排放强度能够下降至0.9840吨/万元,相较于2005年2.9755吨/万元的碳排放强度下降66.93%;(3)将ARIMA-BP神经网络模型得到的碳排放强度预测值与LSTM神经网络模型进行对比,LSTM模型在预测精度上的表现更佳,两个模型对于2030年碳排放强度值的预测相差0.0603吨/万元,对于碳排放强度较2005年降幅预测相差2.03个百分点,验证了本文预测模型的稳健性。

Based on LSTM neural network model,the change trend of China’s carbon emission intensity is indirectly predicted by setting a certain economic growth expectation.At the same time,ARIMA-BP neural network model is established as a verification model to directly predict the carbon emission intensity.The research conclusions are as follows:(1)the mean square error(MSE)of LSTM neural network model on the validation set is 0.00001 and the average absolute percentage error(MAPE)is 0.33%,indicating that the generalization ability of the model is very excellent.Under the prediction framework of LSTM neural network model,China’s carbon emission intensity will reach0.9237 tons per 10000 yuan in 2030,Compared with the carbon emission intensity of 2.9755 tons in 2005,the carbon emission intensity will be reduced by 68.96%;(2)In the prediction and analysis of ARIMA-BP model,it is estimated that China’s carbon emission intensity can be reduced to 0.9840 tons per 10000 yuan in 2030,which is 66.93%lower than that of 2.9755 tons per 10000 yuan in 2005;(3)Comparing the predicted value of carbon emission intensity obtained by ARIMA-BP model with LSTM neural network model,LSTM neural network model performs better in prediction accuracy,and the difference between the two models for the prediction of carbon emission intensity in 2030 is0.0603 tons per 10000 yuan,and the difference between the prediction of carbon emission intensity and that in 2005 is2.03 percentage points,with small error,which proves that the prediction model in this paper is robust.