基于随机森林-投影寻踪法的生物滞留系统多目标评价方法

生物滞留系统性能受植物和介质土的影响显著，且各性能指标具有多样性和不相容性，而传统评价方法具有较强的主观性易导致评价结果出错。采用随机森林模型对原始数据进行特征筛选以降低数据维度，构建投影寻踪模型对不同植物和土壤的水力渗透性能和污染物去除性能进行多目标评价，并利用高鲁棒性的遗传算法（GA）和粒子群算法（PSO）进行模型求解。植物评价结果表明，风车草为生物滞留系统的最佳植物，且评价结果与层次分析模型和BP神经网络模型方法相似；介质土评价结果表明，RST2（9.8%壤砂土+88.2%细砂+2%蛭石）为生物滞留系统的最佳介质土配置方案，且评价结果与传统投影寻踪法相似。研究结果证实了随机森林-投影寻踪（RF-PP）模型适用于生物滞留系统多目标评价。

Multi-objective evaluation method of bioretention system based on random forest-projection pursuit method

The performance of the bioretention system is significantly affected by plants and soil media, and its various diversified performance indexes are incompatible. However, the conventional evaluation methods are flawed by their strong subjectivity, which could easily lead to the errors of evaluation results. Here, the random forest (RF) model was used to screen the original data so as to reduce the dimensionality of the data. Then, the projection pursuit (PP) model was constructed to evaluate the hydraulic permeability and pollutant removal performance of different plants and soil media for multi-objective evaluation, and the model was solved by genetic algorithm (GA) and particle swarm optimization (PSO). The results of plant evaluation showed that Cyperus alternifolius L. was the optimal plant for the bioretention system, which was similar to the results of the analytic hierarchy process model and the back propagation (BP) neural network model. The evaluation results of soil media showed that RST2 (9.8% loamy sand+88.2% fine sand+2.0% vermiculite) was the optimal soil media configuration for the bioretention system, which was similar to the conclusion of the conventional projection pursuit method. These findings indicate that RF-PP model is suitable for multi-objective evaluation of the bioretention system.