Application of Simulation Method and Regression Analysis to Optimize Car Operations in Carsharing Services: A Case Study in South Korea

A carsharing service is a form of public transportation that enables a group of people to share vehicles based at certain stations by making reservations in advance. One of the common problems of carsharing is that companies can have difficulty optimizing the number of vehicles in operation. This paper reports on investigations of the relationship between the number of cars and the number of reservations per day with either the acceptance ratio or utilization ratio based on the commerciallyoperational dataset of a carsharing company in Korea. A discrete event simulation is run to analyze a round-trip service for every possible number of cars and number of reservations with the output acceptance ratio and utilization ratio. The simulation data revealed that increasing the number of reservations with respect to a certain number of cars will decrease the acceptance ratio, thus increasing the percentage of the utilization ratio. Based on the simulation data results, a rational regression model can achieve high precision when predicting the acceptance ratio or the utilization ratio compared to other prediction algorithms such as the Multi-Layer Perceptron (MLP) and the Radial Basis Function (RBF) models. K-means clustering was * Corresponding author Journal of Public Transportation, Vol. 17, No. 1, 2014 122 used to understand the pattern and provide additional policies for carsharing companies. Consequently, opening a carsharing business is very promising in terms of profit, escalating the level of customer satisfaction. In addition, a small reduction in the utilization ratio by operators will create a large increase in the acceptance ratio. Introduction As the world population grows, private vehicles are becoming more attractive, leading to high energy consumption and high vehicle emission levels. Carsharing is one of the transportation strategies that can reduce personal transportation usage and its negative impacts. Because of the worldwide environmental benefits involved, carsharing evolved out of the economic motivations of individuals who could not afford to purchase a vehicle into a mainstream, worldwide transportation system. In recent carsharing systems, customers can access the portal of a carsharing company and easily make a reservation via an Internet connection or by phone. The information, including traveled distances and rent duration, is recorded and charged as to the customer’s bill. An intelligent transportation system can play an important role in making a carsharing system user-friendly, easy to manage, and efficient. Because of these benefits, carsharing as an alternative transportation paradigm has become increasingly popular in many countries (Barth and Todd 1999). Previous research has demonstrated that the benefits of carsharing include reducing costs and the negative impacts of private vehicle ownership and the environmental impacts of auto usage (e.g., congestion, energy consumption, vehicle emissions, and inefficient land use). In North America, the impact of carsharing includes the reduction of emissions as a result of less driving and a 27 percent reduction in the average number of observed vehicle kilometers traveled per year (Martin and Shaheen 2011). According to another review, an additional benefit is cost savings, which was reported to be the main motivation for new memberships from 2006 to 2010. In addition, there has been a change in carsharing activity, as can be seen from the number of worldwide carsharing memberships. In 2006, Europe was the epicenter, but it shifted to North America in late 2010. Stabilized growth in neighborhood residential carsharing and rapid growth in the business and university markets in North America from 2006–2011 was the key trigger. Importantly, there was a worldwide increase in the number of carsharing memberships and in total vehicles and member-vehicle ratios from 2006–2010. As carsharing increasingly becomes a mainstream transportation mode, it is expected that it will be further Application of Simulation Method and Regression Analysis to Optimize Car Operations in Carsharing 123 integrated into metropolitan transportation, land use strategies, and multimodal nodes (Shaheen and Cohen 2013). Up-to-date carsharing systems enable a car to be driven among multiple stations (one-way service), whereas traditional service (round-trip/two-way) allows users to use a car and return it to the same station only. Although one-way service can provide convenience for customers, the cars from each station become disproportionally distributed. Thus, a strategy of vehicle relocation is necessary to elevate the satisfactory level of users. A carsharing system must be efficient, user-friendly, easy to manage, and advantageous to both companies and customers (Barth et al. 2001). Studies concerning data mining have been intensively conducted in carsharingrelated research areas. In particular, the forecasting technique is used to predict the net flow of vehicles in a three-hour period by using neural networks and support vector machines (SVM) (Cheu et al. 2006), and the results show that multilayer perceptron has slightly better accuracy compared to SVM. In another case, such as the one-way type, it is difficult to maintain the distribution balance of parked vehicles among stations. A method for the optimization of vehicle assignment is used according to the distribution balance of parked vehicles; thus, it is possible to maintain distribution balance of parked vehicles and keep the convenience of the carsharing system (Uesugi et al. 2007). In regard to car optimization, one study shows an international comparison regarding carsharing services (Shaheen and Cohen 2007). The paper shows that the member-vehicle ratio is an important key factor that characterizes worldwide carsharing operations. The comparison demonstrates that the member-vehicle ratio based on the survey of each country is different; Asia, Australia, Europe, and North America are 26:1, 17:1, 28:1 and 40:1, respectively. The estimation for the average national ratios are approximately 20:1 and are lower in new markets where carsharing companies must first position their vehicles to gain membership. However, in other research (Morency et al. 2007; Habib et al. 2012; Costain et al. 2012), studies about user behavior in carsharing transaction data sets show interesting results. The data are from Communauto, Inc., a carsharing company in Montreal from January –December 2004. The result reveals that there is variability in the number of transactions and distance traveled by each customer. Another study (Costain et al. 2012) found that increasing the home-to-parking-lot distance reduces trip duration. Thus, it is important to evaluate the member-vehicle ratio with respect Journal of Public Transportation, Vol. 17, No. 1, 2014 124 to other parameters such as variability of the number of transactions, traveled distance, and traveled time by the customer. Advanced simulations in carsharing have focused on developing a relocation model to evaluate one-way car availability (Kek et al. 2009). In addition, a forecasting model for relocation has been suggested to optimize the results of relocation and predict efficient routes (Cheu et al. 2006; Wang et al. 2010; Karbassi and Barth 2003; Correia and Antunes 2012). However, to implement those models, it is important for carsharing companies to decide first on the initial vehicles before focusing on relocation models. Because it is difficult to predict the initial number of cars needed without losing customer interest and company profits, this paper aims to demonstrate that a simulation model must be developed first to evaluate the acceptance ratio and utilization ratio for traditional, round-trip services based on traveling frequency, number of vehicles, and Vehicle Hours Traveled (VHT) and Vehicle Kilometers Traveled (VKT) patterns. Two output parameters were used in this paper. The first was the acceptance ratio, which can be simply explained as successful reservations over total reservations made by customers; this parameter can be expected to reveal general customer satisfaction. The second parameter is the utilization ratio, which is the percentage of total actual driving hours of rented cars over the total possible driving hours of cars, which elucidates company profits. Later, the simulation data results are analyzed using regression and other forecasting techniques to generate a prediction model. This paper aims to focus on how to develop a model that can be used to optimize the number of cars needed with respect to a certain number reservations per day, time patterns, and thresholds of either the acceptance ratio or the utilization ratio. Section 2 of this paper provides an overview of the results of the literature review. Section 3 describes the methodology of the simulation and algorithm analyses. Results and a discussion of the proposed model in are presented in section 4, and limitations and future research of this paper are discussed in section 5.