Modeling Study Of Air Quality Using Advanced Machine Learning-based Response Surface Technique And Cost-efficient Control Strategy Optimization

Ambient fine particulate matters(PM_2.5)and ozone（O₃）have been regarded as major air quality evaluation indicators around the world because of their significant effects on human health and eco-environment.For both developing and developed countries like China and USA,the attainment of stringent ambient PM_2.5 and O₃ standards still requires different levels of reductions in precursors emissions.Quickly quantifying the PM_2.5 or O₃ response to their precursor emission changes is a key point for developing effective control policies.The polynomial function-based response surface model（pf-RSM）can rapidly predict the nonlinear response of PM_2.5 and O₃ to precursors,but has drawbacks of overload computation and marginal effects（relatively larger prediction errors under strict control scenarios）.To improve the performance of pf-RSM,a novel self-adaptive RSM（SA-RSM）was proposed by integrating the machine learning-based stepwise regression for establishing robust models to increase the computational efficiency and the collinearity diagnosis for reducing marginal effects caused by overfitting.The pilot study case demonstrated that compared with pf-RSM,SA-RSM can effectively reduce the training number by 70%and40%and the fitting time by 40%and 52%,and decrease the prediction error by 49%and 74%for PM_2.5 and O₃ predictions respectively;moreover,the isopleths of PM_2.5 or O₃ as a function of their precursors generated by SA-RSM were more similar to those derived by chemical transport model（CTM）,after successfully addressing the marginal effect issue.Developing optimal pollutant reduction program is another key point for developing effective control policies.Although the genetic algorithm（GA-LECO）used in the current air benefit and cost and attainment assessment system optimized edition（ABa CAS-OE）is able to solve the complex cost optimization problem of attaining emission reductions in multiple regions,species,and sectors,it has the disadvantages of low optimization efficiency and random non-reproducible results.This study innovatively investigated the cost-minimizing scenario search algorithm that combined SA-RSM with nonlinear programming to meet the emission reduction targets.The case study showed that the new algorithm not only converges faster and was easier to approach and obtain the global optimal solution than GA-LECO,but also provided a pollutant reduction solution that can reduce the emission cost of precursor pollutants by 47.35%while achieving the same ambient PM_2.5 and O₃ concentrations.In particular,the new algorithm also solved the problem of randomness in the results of the genetic algorithm,and improved the stability and scientificity of the results of the emission reduction optimization algorithm.SA-RSM and the nonlinear programming are expected as a better scientific tool for decision-makers to make sound PM_2.5 and O₃ control policies,and it is expected to have good application and promotion prospects.