姜屿

        近些年，精细化的停车管理已成为越来越多停车场投资管理者追求的目标，它不仅可以很大程度地均衡停车的供需问题，还可以为经营管理产生更多的附加经济价值，从而缩短停车场的投资回报周期。为了实现这样的目标，在精细化停车管理中的一个关键技术就是对停车需求全面、准确的分析。但是，由于传统分析中受限于停车数据量，停车需求分析常常计算一些含有集中趋势意义的指标，而无法考虑停车需求的波动情况，导致了停车需求分析不够全面，进而无法真正达到精细化停车管理的要求。幸运的是，随着连续停车数据的获取越来越方便，不仅可以为停车需求分析提供更加全面的数据基础，还可以为停车管理的风险决策提供更多信息支持，使得在解决这一问题上看到了新的机遇。

       本研究以单一建筑类型的自备路外停车场为研究对象，拟解决的关键问题是分析停车需求的波动性特征以及构建考虑停车需求波动性的预测模型与方法流程。在分析多个建筑类型停车场停车需求特征指标的基础上，提出了停车需求波动性特征的含义与分析思路；结合常用统计检验方法，提出了各类停车需求特征指标时变特征集的概念与验证方法；结合时间序列分析与预测方法，建立白天时段考虑停车需求波动性的时变预测模型；对停车时长进行生存时间分析，从而建立夜晚时段过夜驻留停车需求的预测方法。

       首先，本文分析了多个建筑类型停车场的停车需求特征并提出了停车需求波动性特征的含义与分析思路。基于此，在考虑波动性的停车需求预测方法时，提出分别对白天与夜间的停车需求采用不同的方法进行分析与建模。

       接着，本文进一步探究白天时段各高峰比指标呈现出较小范围波动的性质，结合常用的统计检验方法，提出了各类停车需求特征指标时变特征集的概念与验证方法。

       然后，本文结合时间序列分析与预测方法，建立了白天时段考虑停车需求波动性的时变预测模型。通过停车需求时序图的分析，发现在一定区间内，停车需求存在趋势变化的情况。因此，通过模型预测精度检验以及考虑利于决策信息的解释，本文采用残差自回归模型进行短期高峰停车需求的预测。同时，结合高峰比指标特征集，得到不同风险下的全日各时段的停车需求预测值，为日间停车场管理提供信息决策支持。

       最后，本文基于停车时长的生存时间分析，建立了夜晚时段过夜驻留停车需求的预测模型。在分析过夜驻留停车的生存问题基础上，本文提出了用停车时长分布来估计过夜驻留停车概率的方法思路。通过对比三类生存分析方法，本文最终采用半参数模型方法，构建了多种因素对停车时长影响的Cox比例风险模型，从而获得不同影响因素下的停车时长生存时间曲线以及过夜驻留停车概率，为停车场的夜间需求管理提供决策支持。

关键词：停车需求预测；停车需求波动；精细化停车管理；同步统计推断法；时间序列分析；生存时间分析

     In recent years, the delicacy management of parking has become the goal pursued by more and more parking investment managers. It can not only largely balance the supply and demand of parking, but also generate more additional economic value for business management, thereby shortening the ROI cycle of parking lots. To achieve this goal, the key technology to be addressed in delicacy management is the comprehensive and accurate analysis of parking demand. However, because traditional analysis is limited to the amount of parking data, parking demand analysis often calculates some indicators that have central trend significance but cannot consider the fluctuation of parking demand. As a result, the analysis of parking demand is not comprehensive enough, and the requirement for delicacy management cannot be achieved. Fortunately, with the increasing convenience of continuous parking data collection, it not only provides a more comprehensive data base for parking demand analysis, but also provides more information for parking management risk decision-making. Makes us see new opportunities in solving the problem.

    In this study, the off-street parking lot with a single building type is taken as the research object. The key issues to be solved are the analysis of the characteristics of the fluctuation of parking demand and the construction of forecasting models and method flows based on that. Based on the analysis of parking demand characteristic indicatiors of multiple types of buildings, this study proposed the implication and analysis ideas of the characteristics of parking demand fluctuation. Combining with common statistical testing methods, the concepts and verification methods of time-varying feature sets of various peak ratio indicators were proposed. Combining time series analysis and forecasting methods, a time-variant forecasting model that considers the fluctuation of parking demand during daytime was established. After analyzing the survival time of parking time, a forecasting model for overnight parking demand during night time was established.

    First, this study proposed the implication and analysis ideas of the characteristics of parking demand fluctuation based on the analysis of parking demand characteristic indicatiors of multiple types of buildings.Therefore, the study should use different methods to analyze and model the parking requirements for daytime and night.

    Second, this study proposed the concepts and verification methods of time-varying feature sets of various peak ratio indicators, combining with common statistical testing methods.

    Third, this study established a time-variant forecasting model that considers the fluctuation of parking demand during daytime, combining time series analysis and forecasting methods. By testing model predictive accuracy and considering the interpretation of decision-making information, the study used the residual autoregressive model to predict short-term peak parking demand. At the same time, combined with the peak ratio index feature set, the forecasted parking demand for different periods under different risks was obtained. It provided informational decision support for daytime parking management.

    Finally, this study established a forecasting model for overnight parking demand during night time based on the analysis of the survival time of parking time. The study proposed a method for estimating the overnight parking probability by using the parking time distribution. Through the Cox proportional hazards model, the parking time duration curve and the overnight parking probability under different influencing factors were obtained, which provided decision support for the night-time demand management of the parking lot.

Key Words: parking demand prediction, the fluctuation of parking demand, the delicacy management of parking, simultaneous inference technique, time series analysis, survival time analysis