何思远    

       传统的停车管理由于人工采集手段的局限，数据的粒度和广度很难达到研究的深度要求，导致所获得的停车特征信息比较粗糙，应用效果并不理想。随着停车信息化的快速发展，车辆停放特征的数据能够被自动获取，在连续停车数据环境下，传统被动式停车管理模式被主动精细化停车管理模式所取代开始成为可能。本文将在连续数据环境下对停车场的精细化管理进行研究，结合实际的停车场管理问题，研究将停车场精细化管理中的关键问题确定为三个：（1）如何对停车场进行精细化分区？（2）如何对分区后的停车场进行差别化停车收费?（3）如何对分区后的停车场进行准确的停车位诱导？

       首先，研究选取了三个停车场进行停车需求特征研究。根据用户停车特征的不同，研究将停车场用户总结为临时用户与租位用户两类，分别对三个停车场的两类用户进行实际泊位占用量、停车到达率及停放时长的研究，发现临时用户的停车需求变化波动较多，而租位用户每天的实际泊位占用量时变趋势相似，且整体呈现出稳定变化的特征，随后应用同步统计推断法对三个停车场的租位用户进行了停车需求高峰比曲线拟合及检验，发现三个停车场的租位用户均存在停车需求稳定的特征。

       接着，基于停车需求特征的研究，研究将停车场用户总结为两类：需求稳定用户与需求不稳定用户。研究针对不同用户提出了停车场分区管理，将停车场划分为三个区域：（1）固定用户区域，服务于需求稳定用户，区域车位可安排在停车场内相对偏僻的位置；（2）非固定用户区域，服务于需求不稳定用户，区域车位可布局在停车场内相对便捷的位置；（3）缓冲区域，固定用户区域与非固定用户区域的过渡区域，同时为固定用户区域及非固定用户区域提供车位补给，区域泊位位于其它两区域之间。

       随后，针对分区后的停车场进行差别化停车收费研究。固定用户区域，针对停车需求稳定、停车时间长的刚性出行用户，宜采取按期收费方式，研究针对固定用户区域提出了基于基准收益的价格调整方法，停车场可根据自身经济目标需达到的收益增值水平及内部用户结构调整需求对按期收费的费率阈值进行调整。非固定用户区域，针对停车需求不稳定、停放时长较短的弹性出行用户，宜采用按时收费方式。针对非固定用户区域研究提出了限时停车管理方案，并只增加部分便捷车位停车收费。

       最后，针对分区后的停车场进行智能化管理研究，包括内部引导及外部诱导两部分。分区后的内部引导管理基于RFID技术实现，通过发放电子标签对用户进行进出管理以及内部定位追踪。分区后的外部诱导通过设置动态分级诱导标志实现，发布的实时停车位信息可根据实时停车需求的预测结果获得，研究采用深度学习GRU模型与传统时间序列ARIMA模型分别进行短时停车需求预测，发现深度学习的预测模型要优于传统时间序列模型。

关键词：精细化管理，停车场分区，差别化收费，深度学习

Abstract

       Due to the limitations of manual collection methods, the granularity and breadth of data is difficult to meet the depth requirements of the research. As a result, the obtained parking feature information is rough and the application effect is not satisfactory. With the rapid development of parking information, the data of vehicle parking characteristics can be automatically acquired. In the continuous parking data environment, the traditional passive parking management mode is replaced by the active refined parking management mode. This paper will study the refined management of parking lot in the continuous data environment. Combined with the actual parking lot management problem, the key issues in the refined management of parking lot are determined as three: (1) How to partition the parking lot nicely? (2) How to make differential parking charges for the parking lot after the partition? (3) How to accurately predict the parking spaces after the partition?

       First, the paper selected three parking lots for the research of parking demand characteristics. According to the different parking characteristics of the users, the research summarizes the parking users into two categories: temporary users and rented users. The two types of users in the three parking lots are studied on the parking demand, parking arrival rate and parking time. The parking demand distributions of temporary users change fluctuatingly, while the parking demand distributions of the rented users are similar and the overall distribution of the parking demand of the rented users has a characteristic of stability. Then, the synchronous statistical inference method is applied to the rented users of the three parking lots. The parking demand peak ratio curve fitting and inspection found that the rented users of the three parking lots all have the characteristics of stability.

       Then, based on the study of parking demand characteristics, the paper summarizes the parking lot users into two categories: stable users and unstable users. The paper proposes the management strategy of partitioning parking lot for different users, and divides the parking lot into three areas: (1) fixed user area, serving stable users, and regional parking spaces can be arranged in relatively remote locations in the parking lot; (2) non-fixed user area, serving users with unstable demand, regional parking spaces can be placed in a relatively convenient location in the parking lot; (3) buffer area, transition area between fixed user area and non-fixed user area, and provides parking space replenishment for fixed user area and non-fixed user area, and the regional berth is located between the other two areas.

       Subsequently, a differentiated parking charge study was conducted for the parked parking lot. For the fixed user area, for rigid travel users with stable parking demand and long parking time, it is advisable to charge on the lease, and in this paper, the price adjustment method based on the benchmark income for the fixed user area is proposed. The parking lot can increase the charges and adjust the internal user structure according to its own economic goals. For the non-fixed user area, for flexible travel users with unstable parking demand and short parking time, it is advisable to use on-time charging. The parking management schemes of limited parking time are proposed for non-fixed user area research, and the strategy of increasing parking fees only for convenient parking spaces is proposed too.

       Finally, the intelligent management of the parking lot is studied, including internal guidance and external induction. The internal guidance after parking lot partitioning is implemented based on RFID technology, and the user is managed by issuing electronic tags. The external induction after partitioning is realized by setting the dynamic grading induction flags. The published real-time parking space information can be obtained according to the prediction of the real-time parking demand. The paper adopts the deep learning GRU model and the traditional time series ARIMA model to predict the short-term parking demand. And find that the predictive model of deep learning is superior to the traditional time series model.

Key Words：refined management, parking lot partitioning, differentiated charging, deep learning