基于机器学习和车辆路径问题模型的园林绿化垃圾空间分布预测与处理设施选址研究——以武汉市为例

园林绿化垃圾资源化循环利用的基础是处理设施体系建设。以武汉市中心城区为例,分别对园林绿化垃圾空间分布与处理设施选址进行研究。针对前者,基于高清卫片、街景影像、行道树空间点位等大数据,利用机器学习中的分类(卷积神经网络)、聚类(K均值聚类)和回归预测(多项式回归、随机森林)等模型,实现对现状及规划地区的日常和峰值园林绿化垃圾产生量的合理测算;针对后者,采用车辆路径问题类算法中的自适应大邻域搜索算法,带入距离最近、运输周转量最少、有限时间窗口和有限设施容量等限定条件,对组团收集点、就近消纳站、综合处理厂和集中转运站4类设施及有关线路进行智能选址。根据结果,预测研究区域内日常园林绿化垃圾产生量为32.71 万t/年,峰值产生量为7 万t,与同等级城市情况类似。建议布局200处组团收集点、50个就近消纳站、4处综合处理厂和14个集中转运站,以提升园林绿化垃圾收集清运效率,实现区域“产-收”平衡。

Research on Spatial Distribution Prediction andTreatment Facility Location of Greenery WasteBased on Machine Learning and VRP Models: ACase Study of Wuhan City

The system of landscaping waste treatment facilitiesis the fundamental link for its resource utilization and recycling.The research takes Wuhan as an example to study the spatialdistribution of greenery waste and the location of treatment facilities.For the former, based on big data such as high-definition satellitepictures, street view images, street tree spatial points, etc., andusing classification (convolutional neural network), clustering(K-means clustering), regression prediction (polynomial regression),the reasonable calculation of daily and peak greenery wastevolumes in current and planning areas is realized. For the latter,the research adopts ALNS algorithm in vehicle routing problems,which incorporates constraints such as closest distance, minimumtransportation turnover, limited time window, and limited facilitycapacity. It intelligently selects four types of facilities, includingcluster collection points, nearby consumption stations, comprehensiveprocessing plants, and centralized transfer stations, as well as relatedlines. The results show that the daily output of landscaping waste inthe study area is 327,100 tons/year, with a peak output of 70,000 tons,similar to the situation in cities of the same level. It is recommendedto layout 200 cluster collection points, 50 nearby consumptionstations, 4 comprehensive treatment plants, and 14 centralizedtransfer stations to improve collection and transportation efficiencyand achieve regional "production income" balance.