生活垃圾源头分类是我国实施的重要废物资源化战略。开发简单易行的垃圾分类效果识别方法，研究源头分类与末端处置效益间的关系，明确处置系统的优化路径，对于提高垃圾分类监管水平、提升居民分类准确度、更好地实现垃圾资源化都具有重要意义。本研究通过在北京和张家港市实地采集1080个袋装垃圾样本，对每个样本分成13种组分，利用3种树结构的机器学习算法构建样本理化特性间的回归关系，提出了基于容重的厨余垃圾-其他垃圾简易识别方法和容重阈值，并开发了容重识别垃圾分类效果智能监测设备。综合运用多目标优化、环境影响评价和成本效益分析方法，探究了经济-环境-能源综合效益优化情景下的终端处置结构与前端厨余垃圾分出率，提出了前端分类与终端处置耦合的系统规划方案。（1）利用不同类型垃圾的容重差异可以简便快速地识别成分复杂的厨余垃圾和其他垃圾。“容重-厨余垃圾组分-其他垃圾组分”的XGBoost机器学习回归模型的决定系数R2为0.65，容重与垃圾分类准确度显著相关，其他垃圾、混合垃圾和厨余垃圾的容重值分别为：0~96、96~229和229~1000 kg/m3。（2）厨余垃圾容重阈值存在季节差异，其他垃圾呈季节均一性。西瓜皮和橙子皮的季节性差异导致厨余垃圾含水率在夏季高于冬季，进而造成夏季和冬季的容重阈值分别为258和229 kg/m3。容重识别的理论方法可应用在不同地区，使用时采集特定地区数据训练模型，对模型进行参数校准得到特定垃圾容重阈值。（3）基于简单易测量的容重值可预测垃圾的多种基础理化特性，建立了垃圾容重与分类准确度、含水率、热值的Logistic定量化模型，决定系数R2为0.6~0.7，可在一定程度上克服传统测量方法因样品用量少而带来的测量误差。（4）设计的智能容重监测设备可有效提高居民的分类准确度。以垃圾容重阈值为基础，利用3D摄像头和质量传感器等多元传感技术创新性地实现对各种颜色袋装垃圾的无接触快速识别，准确率在90%以上，在张家港市居民小区现场连续运行5个月，使分类准确度从38%提升至70%，与人工监管模式的分类效果相当（72%）。（5）建立了垃圾分类处置的经济-能源-环境效益的多目标优化模型，使前端分类后其他垃圾的容重值与终端处置效益动态关联。以张家港市为例，1 t混合垃圾为单位，将0.2 t厨余垃圾厌氧发酵、0.8 t垃圾焚烧处置可实现综合效益最大化（341元/t垃圾），对应前端厨余垃圾分出率为34%，分类后其他垃圾的容重阈值为82 kg/m3。

Municipal solid waste (MSW) source separation is a crucial strategy for China’s waste recycling. Developing practical and efficient methods for waste source separation assessment, investigating the relationship between source separation and end-of-life disposal benefits, and elucidating optimization treatment pathways hold significant importance in enhancing waste classification supervision, improving residential source separation accuracy, and better achieving waste recycling. In this research, 1080 bagged waste samples were collected in Beijing and Zhangjiagang, and each sample was divided into 13 subcomponents. Regression relationship among the samples’ physiochemical properties were constructed by three machine learning algorithms. Then a simple bulk density-based identification method for food and residual waste was proposed, and an intelligent monitoring device for assessing waste classification effectiveness was developed based on bulk density. With multi-objective optimization, life cycle assessment and cost-benefit analysis, this study investigated scenarios for optimizing the economics, environment, and energy benefits. This exploration encompassed the terminal disposal structure and MSW source separation rate, and a comprehensive scheme was put forth to achieve an optimized waste management system.(1) The bulk density threshold can be used to quickly and effectively identify food waste and residue waste. The determination coefficient (R2) of the XGBoost regression model for “bulk density-food waste composition-residue waste composition” is 0.65. There is a significant correlation between bulk density and waste source separation accuracy. The bulk density thresholds for residue waste, mixed waste, and food waste are 0~96, 96~229, and 229~1000 kg/m3, respectively. (2) The density threshold of food waste exhibits seasonal variations, while that of residue waste remains seasonally consistent. The seasonal differences in water content due to items like watermelon and orange peels result in higher bulk density during summer (258 kg/m3) than in winter (229 kg/m3). The theoretical method of bulk density recognition can be applied in different regions. When used, region-specific data is collected to train the model, and the model is then calibrated to obtain specific waste density thresholds for that region. (3) MSW multiple basic physicochemical properties can be predicted using the easily measurable bulk density. Logistic regression expressions were established to quantify the relationships between bulk density and classification accuracy, water content, and calorific value, with determination coefficient (R2) ranging from 0.6 to 0.7. These expressions help mitigate measurement errors due to small sample sizes compared to traditional measurement methods. (4) The designed intelligent bulk density monitoring equipment can effectively improve residents’ source separation accuracy. Via theoretical bulk density threshold, multi-sensor technologies including 3D cameras and mass sensors enables non-contact, rapid identification of bagged waste types, with an accuracy rate exceeding 90%. In a residential community in Zhangjiagang City, the device operated continuously for 5 months, improving the separation accuracy from 38% to 70%, comparable to the performance of manual supervision (72%). (5) A multi-objective optimization model for the economic-energy-environmental benefits of waste classification and disposal was developed. It dynamically associates the MSW bulk density with terminal disposal benefits. Taking Zhangjiagang city as an example, with 1 ton mixed waste as the functional unit, maximizing comprehensive benefits (341 CNY/ton) is achieved by 0.20 tons of food waste anaerobic fermentation and 0.80 tons of residue waste incineration. This corresponds to 34% waste source separation rate and 82 kg/m3 bulk density threshold.