摘要
在碳排放预测模型的定量研究中,由于影响碳排放的因素(变量)多而复杂,且变量之间信息彼此重叠,从而出现碳排放预测模型稳定性差、预测精度不高等问题,因此探寻一种有效的变量筛选方法成为急需。本研究基于PLS-VIP算法,以1980—2017年的中国碳排放量为因变量,10个与碳排放量相关的影响因素为自变量进行PLS建模;通过计算原始自变量的投影重要性指标(VIP值)和交叉验证均方根误差(RMSECV)实施变量筛选,然后利用筛选出的变量通过PLS回归对测试数据进行检验。结果发现利用PLS-VIP方法能有效识别出与因变量相关性较强的变量,并从根本上减少进入模型的变量个数;同时基于筛选出的变量建模,其预测的精度和稳定性明显优于传统PLS方法,表明PLS-VIP方法在变量筛选和提高预测性能上是有效的。在模型的可解释性上,与传统的PLS方法相比,PLS-VIP方法筛选出的变量对因变量具有更科学、更合理和更强的解释能力,表明PLS-VIP方法能有效处理多变量的复杂性问题,是一种可行的变量筛选方法。
In the quantitative research of carbon emission prediction model,there are many and complex factors(variables)affecting carbon emissions,and the overlapping information among variables,which leads to the poor stability and low prediction accuracy of carbon emission prediction model.Therefore,it is urgent to find an effective variable selection method.Based on the PLS-VIP algorithm,the author built PLS model with 10 factors related to carbon emissions as independent variables and China’s carbon emissions as dependent variables from 1980 to 2017.Variable screening was carried out by calculating the projection importance index(VIP value)of the original independent variables and cross-validation root mean square error(RMSECV),and then the selected variables are used to validate the test data by PLS regression.The results show that the PLS-VIP method can effectively identify variables with strong correlation with dependent variables,and fundamentally reduce the number of variables entering the model.At the same time,based on the selected variables,the prediction accuracy and stability of PLS-VIP method are significantly better than the traditional PLS method,which shows that PLS-VIP method is superior to the traditional PLS method in variable selection and improving prediction performance.Compared with the traditional PLS method,the variables selected by PLS-VIP method have more scientific,more reasonable and stronger explanatory ability to dependent variables,which shows that PLS-VIP method can effectively deal with the complexity of multivariate problems,and is a feasible variable selection method.
作者
刘贤赵
杨旭
LIU Xian-zhao;YANG Xu(College of Resources,Environment and Safety Engineering,Hunan University of Science and Technology,Xiangtan 411201,China)
出处
《环境生态学》
2019年第8期60-65,共6页
Environmental Ecology
基金
国家社会科学基金项目(17BGL138)资助.
