基于神经网络的碳排放预测及影响因素分析

基于新疆1995—2014年农业生产碳排放源,建立碳排放关系数据库。应用广义神经网络(generalized regression neural network,GRNN)构建了排放量预测模型,结合平均影响值(mean impact value,MIV)方法对碳排放影响因素进行量化。结果表明:1)GRNN模型预测碳排放的平均绝对百分误差和拟合优度分别为2.7860%和0.8720;2)新疆人口、人均GDP、农业贡献值、农机总动力和农户固定资产投资等因素对农业生产碳排放的影响程度分别为0.6210、0.2377、0.3698、0.8500和0.1000。该成果可为新疆碳排放总量分析和影响因素量化方面提供参考。

PREDICTION AND IMPACT FACTOR ANALYSIS OF AGRICULTURAL CARBON EMISSION BASED ON NEURAL NETWORK

The suitable index was proposed for creating carbon emission database of Xinjiang based on statistical panel data from 1995-2014.The prediction model for Xinjiang carbon emission on agricultural production was proposed based on generalized regression neural network (GRNN) and combined with average impact value (M1V) method to quantify the influence degree of impact factors on carbon emissions.The results showed that:1) The mean absolute percentage error and the goodness of fit of the GRNN model were 2.7860％ and 0.8720,respectively.2) Influence degree of population,per capita GDP,agricultural contribution value,the total power of agricultural machinery and household investment on Xinjiang carbon emission were 0.6210,0.2377,0.3698,0.8500 and 0.1000,respectively.It can provide references for the analysis of the total amount of carbon emission in Xinjiang and the influence factors.