机器学习驱动的智慧停车的运营与管理规划模型研究

Research on the Operation and Management Planning Model of Intelligent Parking Driven by Machine Learning

摘要: 截止2020年底，沈阳市民用汽车保有量达到了263.5万辆。人均汽车保有量位居全国第十。汽车保有量的增长给市民的出行带来了极大的便利，但同时也给道路交通和停车带来了不少压力。近几年，人们已将注意力放在了如何解决“停车难”和“乱停乱放”等问题上，开始发展智慧停车。当前智慧停车的计费方式主要是移动视频采集车计费和人工计费两种。本文分析了视频采集车计费的收益和损失，研究了停车收费路段中视频采集车和人工采集方式的选择和规划问题。针对问题一，在仅考虑一个停车位的情况下，计算移动采集车计费可能产生的计费收益或损失。在采集车采集的频率足够多的情况下，会出现停车时间累积，从而导致人工计费和视频采集车计费不同。在问题一中我们仅以一辆车的停放为例，采用分析列举法后总结得出视频采集车计费方式在单次不超过30分钟，且累次后超过30分钟情况下受益最大。针对问题二，首先，应用聚类分析的方法及pycharm中k-means聚类算法，得到三类中心点类簇；然后，通过最短路径算法，在同一类簇中进行坐标点的线路规划，得到每一类中采集车实际规划道路；最后，根据每一类簇中的停车位数，以及道路情况给出视频采集车最优配置方案，分别配置A类1辆、B类2辆、C类1辆，共四辆视频采集车。针对问题三，首先采用0-1规划模型，根据采集车和人工对应的决策因子决定每个路段采用的采集方式；然后，将采集车与人工计费成本作比，视频采集车与人工计费的人员数量作比，对比两个比值，得到应用视频采集车计费的路段。再次，通过聚类分析及matlab，绘制视频采集车行驶道路；建立多目标线性规划模型，筛选多种影响计费收益因素，将时间和地区作为主要因素；最后，建立道路计费停车总收益的表达式，并根据九种不同情况合并得出最优综合计费方案。

Abstract: By the end of 2020, the number of civilian cars in Shenyang reached 2,635,000. The per capita car ownership ranked the 10th in the country. The growth of car ownership has brought great convenience to citizens’ travel, but it has also brought considerable pressure to traffic and parking. In recent years, people have focused on solving problems such as “difficult parking” and “random parking”, and have begun to develop smart parking. Currently, the charging methods for smart parking mainly include mobile video collection vehicle charging and manual charging. This paper analyzes the gains and losses of video collection vehicle charging, and studies the selection and planning of video collection vehicle and manual collection methods in parking charging sections. For problem one, when only considering one parking space, it calculates the possible charging gains or losses of the mobile collection vehicle. When the collection frequency of the collection vehicle is sufficient, parking time accumulation will occur, resulting in different charging methods between manual charging and video collection vehicle charging. In problem one, we only take the parking of one vehicle as an example, and after using the analysis and enumeration method, we summarize that the video collection vehicle charging method benefits the most when the single parking time does not exceed 30 minutes and the cumulative parking time exceeds 30 minutes. For problem two, first, the k-means clustering algorithm in pycharm is used to apply the clustering analysis method and obtain three types of center point clusters. Then, the shortest path algorithm is used to plan the route of the coordinate points within the same cluster, and the actual planned road of the collection vehicle in each cluster is obtained. Finally, based on the number of parking spaces in each cluster and the road conditions, the optimal configuration plan for the video collection vehicle is given, with one vehicle for class A, two for class B, and one for class C, totaling four video collection vehicles. For problem three, first, the 0-1 programming model is used, and the collection method for each section is determined based on the decision factors corresponding to the collection vehicle and manual charging. Then, the cost of video collection vehicle charging is compared with that of manual charging, and the number of personnel for video collection vehicle charging is compared with that of manual charging. By comparing the two ratios, the sections where video collection vehicle charging is applied are obtained. Next, through clustering analysis and matlab, the driving routes of the video collection vehicles are drawn. A multi-objective linear programming model is established, and multiple factors affecting the charging revenue are screened, with time and region as the main factors. Finally, the expression of the total charging revenue of the road parking is established, and the optimal comprehensive charging plan is obtained by combining nine different situations.