Abstract: Bike-sharing is a green, low-carbon and sustainable mode of transportation. It has potential benefits such as improving physical health, promoting travel safety and reducing carbon emissions. It is an important part of a sustainable public transport model. It will also be a crucial method to accomplish the strategic objective of “carbon peaking and carbon neutrality” in urban transportation systems. In the bike-sharing system, a small number of bikes are out of service every day. But unfortunately, the collection of availability information is complicated due to the maintenance management of shared bikes. Although modern technologies such as GPS, internet of things and cloud computing provide a rich information environment for the bike-sharing system, the fault and maintenance information of the bike-sharing system is still lacking. The fault feedback function is embedded in the bike-sharing system. However, users’ willingness to feedback the bike fault is not strong, and the availability of feedback information is also insufficient. The lack of fault information about shared bikes indeed increases the difficulty of quickly identifying unusable shared bikes from the perspective of system reliability. When analyzing the travel data of users in this paper, it is found that users frequently rent-out shared bikes within a short period of time, which may imply information about the availability of shared bikes. Therefore, a Bayesian model is proposed in this paper. According to the rental transaction data of users’ trips, the rent-out data is extracted, site attributes and the time information of users’ rent-out shared bikes are introduced, and a Bayesian extended model with covariates is constructed to estimate the unusable probability and number of unusable shared bikes in the site by using online transaction data. This paper verifies the effectiveness of the proposed method based on real data of Hohhot bike-sharing system database.
   In the application of the method, we first analyze the data of one day on August 1, 2017. Based on the hypothesis of KASPI et al., the prior probability applies the Bayesian extended model with covariates to obtain the unavailability probability of each shared bike and the number of unusable shared bikes in the site according to the cumulative rental and rent-out times of shared bikes. The results show that in the sites with high activity, if the shared bikes are canceled for several consecutive times, the availability of the shared bikes is low. At the same time, if there are multiple bikes with the same number of rentals in a day’s running time, the rentals in the peak period will have a higher probability of unavailability than in other periods. In addition, since the prior probability and values are both assumed values, the optimal parameter values cannot be determined. Therefore, this paper simulates different prior probabilities and values, and obtains the unusable probabilities of shared bikes under different circumstances. However, which specific parameter value has better simulation effect?Further verification needs to compare the actual number of unusable shared bikes at a specific time point with the estimated one of unusable bikes in this paper. Due to the lack of such data required by this project, this paper does not make a comparison, but only studies the unavailability probability and unavailability quantity of shared bikes under different assumed parameters.
   Based on the results obtained in this paper, the following suggestions can be put forward for the operation and management personnel of shared bikes: (1)The operation and management personnel can timely understand the availability level of shared bikes according to the unavailability probability of shared bikes, arrange maintenance tasks in a planned way, prevent the accumulation of maintenance tasks during peak periods, and make full use of human resources. (2)Small faults of shared bikes from deteriorating into bigger faults can be prevented, maintenance costs reduced and sustainable development achieved. (3)Timely maintenance of shared bikes can reduce the probability of users riding faulty bikes, reduce the risk of users travelling, improve the satisfaction of users travelling, boost users’ retention rates, and bring greater benefits to the long-term development of enterprises.
   Since the information of specific bikes in a single station in the bike-sharing system is incomplete, the number of unusable bikes estimated in this paper is a comprehensive number of unusable bikes for all stations. In further research, we will determine the optimal parameter value based on this, and accurately calculate the number of unusable bikes at each station.

Key words: shared bikes, availability, Bayesian model, cancelling rental data, failure

摘要： 共享单车的运营管理对于减少碳排放、实现城市交通的可持续发展具有重要意义。在共享单车系统中,每天都有一小部分的单车无法使用。目前共享单车维护管理的可用性信息收集复杂。尽管GPS、物联网、云计算等现代技术为共享单车系统提供了丰富的信息环境,但共享单车自身的故障和维护信息仍然缺乏,没有可靠的在线信息表明共享单车的可用性。本文在对用户出行数据进行分析时,发现存在短时间内用户频繁退租共享单车的现象,这种现象可能隐含着共享单车可用性的信息。因此,本文提出了一个贝叶斯模型,根据用户出行的租赁交易数据,提取退租数据,引入站点属性和用户退租共享单车的时间信息,构建带有协变量的贝叶斯扩展模型。利用在线交易数据估计站点中共享单车的不可用概率以及不可用自行车的数量。本文基于呼和浩特市共享单车系统数据库的真实数据验证该方法的有效性。

关键词: 共享单车, 可用性, 贝叶斯模型, 退租数据, 故障