基于单分类机器学习的生活垃圾焚烧发电设施选址可行性与影响因素识别：来自全国1058个项目的证据

One-Class Machine Learning-Based Feasibility Assessment and Factor Identification for Siting Waste-to-Energy Incineration Facilities: Evidence from 1058 Projects in China

中文关键词:  垃圾焚烧发电设施  单分类机器学习  选址优化  环境社会风险

英文关键词:waste-to-energy incineration facilities  one-class machine learning  site selection optimization  environmental and social risk

中文摘要:

      如何实现生活垃圾焚烧发电设施的科学高效选址，是其环境社会风险防范的关键问题。本文以全国1058个已建设施为正类样本，整合环境、人口、社会与经济四维度的16项指标，提出一套基于单分类机器学习的选址可行性评估流程。研究构建树模型算法（IF、OCC Tree、RFDE、LOF Tree）用于解释特征变量贡献，并与OC-SVM、LOF、EE、GMM、AE等算法进行跨范式对比，最终将表现最优的OCC Tree、LOF与AE算法集成为综合模型以提升稳健性。结果显示，人口密度与环境投诉为主导因子，植被指数、学校密度和地区碳排放情况具有中等影响，运输效率对选址具有成本约束。综合模型表现出了较好预测性能，在验证集上的Accuracy=0.987、F1=0.994。敏感性分析表明，在±10%～± 20%扰动情景下，植被指数和环境投诉的微小变化即可引起选址适宜度得分的较大波动。研究为“邻避”设施在厂址预筛选和冲突预防阶段提供了可解释、可复用的技术路径。

英文摘要:

      The scientific and efficient siting of waste-to-energy (WtE) incineration facilities has become a key issue for preventing related environmental and social risks. Using 1058 existing projects nationwide as positive samples, this study integrates 16 indicators across four dimensions―environment, population, society, and economy―and proposes a siting feasibility assessment workflow based on oneclass machine learning. Tree-based algorithms (IF, OCC Tree, RFDE, LOF Tree) are constructed to interpret feature contributions and are compared across paradigms with OC-SVM, LOF, EE, GMM, and AE. The best-performing models, OCC Tree, LOF, and AE, are ultimately ensembled into a composite model to enhance robustness. The results show that population density and environmental complaints are the dominant factors, while vegetation index, school density, and regional carbon emissions have moderate impacts; transportation efficiency imposes cost constraints on siting. The ensemble model exhibits strong predictive performance, achieving an accuracy of 0.987 and an F1 score of 0.994 on the validation set. Sensitivity analysis indicates that under ±10%～±20% perturbation scenarios, small changes in vegetation index and environmental complaints can lead to substantial fluctuations in siting feasibility scores. The study provides an interpretable and reusable technical pathway for early-stage site screening and conflict prevention of NIMBY-type facilities.