School of Civil and Transportation Engineering, Guangdong University of Technology, Guangzhou, 510006, China.
School of Management, Guangzhou University, Guangzhou, 510006, China.
Environ Sci Pollut Res Int. 2020 Jun;27(17):21762-21776. doi: 10.1007/s11356-020-08627-1. Epub 2020 Apr 11.
De-carbonization of the transport sector is an important pathway to climate-change mitigation and presents the potential for future lower emissions. To assess the potential quantitatively under different optimization measures, this paper presents a hybrid model combining an integrated machine learning model with the scenario analysis. We compare the training accuracy of the back-propagation neural networks (BPNN), Gaussian process regression (GPR), and support vector machine (SVM) fitting model with different training datasets. The results indicate that the performance of the SVM model is superior to other methods. And the particle swarm optimization (PSO) algorithm is then used to optimize hyper-parameters of the SVM model. Two scenarios including business as usual (BAU) and best case (BC) are set according to the current trends and target trends of driving factors identified by the extended stochastic impacts by regression on population, affluence, and technology (STIRPAT) model. Finally, to find the de-carbonization potentials in the transport sector, the PSO-SVM model is applied to predict transport emissions from 2015 to 2030 under two scenarios. Results show that transport emissions reduce by about 131.36 million tons during 2015-2020 and 372.86 million tons during 2021-2025 in the BC scenario. The findings can effectively track, test, and predict the achievement of policy goals and provide practical guidance for de-carbonization development.
交通部门的脱碳是气候变化缓解的一个重要途径，并有潜力实现未来更低的排放。为了在不同的优化措施下定量评估潜力，本文提出了一种将集成机器学习模型与情景分析相结合的混合模型。我们比较了反向传播神经网络（BPNN）、高斯过程回归（GPR）和支持向量机（SVM）拟合模型在不同训练数据集下的训练精度。结果表明，SVM 模型的性能优于其他方法。然后，粒子群优化（PSO）算法被用于优化 SVM 模型的超参数。根据扩展的人口、富裕程度和技术对冲击回归（STIRPAT）模型确定的当前趋势和目标趋势，设置了两种情景，即照常情景（BAU）和最佳情景（BC）。最后，为了在交通部门找到脱碳潜力，将 PSO-SVM 模型应用于在两种情景下预测 2015 年至 2030 年的交通排放量。结果表明，在 BC 情景下，2015-2020 年期间交通排放量减少了约 1.3136 亿吨，2021-2025 年期间减少了 3.7286 亿吨。研究结果可以有效地跟踪、测试和预测政策目标的实现情况，并为脱碳发展提供实际指导。
