Simulation-Based Regression Models to Estimate Bus Routes and Network Travel Times

This paper presents an approach to estimate bus route and network travel times using micro-simulation. This can be used in predicting the effectiveness of bus route designs using some network traffic measures or indicators. The used indicators are average network traffic intensity, posted speeds, route length, frequency of bus operation, and average passenger loadings (boarding and alighting). Regression models are calibrated to predict both route and overall network travel times. The prediction errors of these models were investigated and analyzed, and regression models were validated. Results indicated the validity of the calibrated regression models. Conclusions are made on how the devised models can be validated in reality and used for route planning purposes to determine best operating conditions such as the frequency. Introduction The performance of a transit network depends on the effective planning and design of transit routes. To ensure effective planning of transit networks, it is important to develop tools or methods to characterize network effectiveness as a function of frequency, route design, and other factors such as traffic network intensity and passenger loadings. Such methods or tools are eventually needed to assist transport agencies in transit planning applications, alteration of service Journal of Public Transportation, Vol. 16, No. 4, 2013 108 schedules, devising of enhancement policies, macro-management of operation, and, ultimately, better service for transit users. Transit effectiveness measures are needed to quantify how efficiently transit system inputs are used in producing a given output (Nash 2006). Among the common effectiveness indicators of transit network design are the overall network and route travel times. The lesser the travel times of the designs needed to provide service to specific transit demand, the better is the design and the more attractive is the service to transit users. The effectiveness indicators are influenced by many factors, such as number of bus stops on routes, number of passengers boarding and alighting, speed restrictions, route length and alignments, etc. In general, the factors that affect travel times include human, vehicular, and facility aspects. Different drivers and road conditions could cause large differences in journey times. For the same time interval and on the same link, different vehicles can have quite different travel times (Li and McDonald 2002). Free-flow travel speed is another factor that affects network travel time. Journey speed along an arterial road depends not only on the arterial road geometry but also on the traffic flow characteristics and traffic signal coordination (Lum et al. 1998). Other main factors cited in previous studies include incidents (Karl et al. 1999), signal delay (Wu 2001), weather conditions (Chien and Kuchipudi 2003), and traffic congestion levels (Lin 2005). Speed (Chien 2003), frequency, and number of boarding and alighting passengers of bus service (Tetreault 2010) have been used for route and network average travel time prediction. The use of travel time information is essential for long-term design of transit service as well as scheduling. In relatively stable light traffic conditions, with light transit demand, fairly simple estimation procedures may be used to estimate travel times. On the other hand, in rapidly-changing traffic conditions, using sophisticated prediction models is essential (Van Grol et al. 1999). Different studies suggested different techniques for estimating or predicting travel times (Kwon et al. 2003; Chakraborty and Kikuchi 2004; Zhang and Rice 2003; El-Geneidy et al. 2010; Tetreault and El-Geneidy 2010). Kwon et al. (2000) used linear regression and advanced statistical methods to develop models for predicting travel time. Chakraborty and Kikuchi (2004) developed a simple linear equation using regression analysis to predict automobile travel time based on bus travel time. Zhang and Rice (2003) developed a linear model with time varying coefficients for short-term travel time prediction. Multivariate