无废城市建设场景下LightGBM垃圾产量预测模型构建

中国于2018年提出“无废城市”建设试点工作方案。通过对某试点研究区域调研发现，近年来当地生活垃圾产量呈现高出城市人口发展速度，且呈非线性快速增长趋势。传统的预测手段已经无法满足当地垃圾产量的精细化管理需求，难以将当地垃圾处理能力的发展与产量增加的趋势相协调。因此，基于社会多元数据，构建对研究区域整体垃圾总产量预测的模型研究方案。通过将灰色关联分析算法与LightGBM机器学习算法结合，获得了多元社会数据中与研究区域垃圾产生增长关联最为密切的几类特征数据，以进行机器学习模型构建与交叉验证调优，获得了MAE为1.48,MAPE为15.42%的生活垃圾产量精准预测模型。最终利用该模型预测，2025年该地的生活垃圾产量将达到17.23万t/a。

CONSTRUCTION OF LightGBM WASTE PRODUCTION PREDICT MODEL IN A SCENARIO OF ZERO-WASTE CITY

In 2018, China put forward a pilot project to construct Zero-Waste City. Through the investigation of the research area of one of the pilot cities, it was found that the domestic waste produced in this area showed a trend of rapid non-linear growth in recent years. Traditional predicting methods can no longer meet the expected delicacy management demand of local waste production. It is challenging to coordinate the development of local waste treatment capacity with the production increasing. Therefore, based on the urban multi-social data with the annual unit, a model research plan for predicting the overall waste production of the study area was proposed. Combining of the grey relational analysis algorithm and LightGBM machine learning algorithm can get several types of feature data in the multi-social data, which are most closely related to the growth of waste generation in the study area. The machine learning model construction and cross-validation tuning to obtain the domestic waste production prediction model with an MAE of 1.48 and an MAPE of 15.42%. Finally, the model predicted that domestic waste production in 2025 will reach 172,300 tons/year.