Performance Analysis of Hublidharwad Bus Rapid Transit System

Abstract

The concept of ‘Smart Mobility’ is one of the innovative solutions to tackle many urban transportation-related issues; that will connect various elements of technology and mobility, and Intelligent Transport System (ITS) is a step toward implementing it. The ITS integrates transportation system users with vehicles and infrastructure using information and communication technology. Bus Rapid Transit System (BRTS) is a state-of-the-art smart mobility system and is a boon for urbanized areas, which are affected by numerous transportation-connected glitches. The role of BRTS has now been recognized as essential for physically active, economically sound, and energyefficient cities. The BRTS has a combined structure of various exclusive features with a strong identity and distinctiveness. A dedicated lane of bus operation is the critical parameter of any BRTS, which will enhance its performance from all the perspectives. The interference of mixed traffic with the operation of BRTS buses, although it only occurs on a few road segments, can compromise the end-to-end travel time of the whole system due to congestion and contribute to reliability-related problems. Many internal and external factors will also influence the Travel Time Reliability (TTR) and Travel Time Variability (TTV) temporally as well as spatially and finally cause an impact on whole system performance. The main motivation behind this research is to study the impact of such nondedicated, and dedicated lanes of BRTS bus operation on its overall system performance from multiple perspectives by identifying bus stations, routes, and segments that are critical in nature. The current study used Automatic Vehicle Location (AVL) data and Automatic Passenger Count (APC) data from the recently implemented Hubli-Dharwad Bus Rapid Transit System (HDBRTS) as a case study. HDBRTS buses operate as express and non-express routes along the single linear corridor between twin cities Hubli and Dharwad. Express route buses serve the limited bus stations, whereas non-express route buses serve all the bus stations. Most of the buses of both environments will run from terminal to terminal, such as the terminal ii at the Hubli side to the terminal at the Dharwad side, which is named as UP direction, and the terminal at the Dharwad side to the terminal at the Hubli side which is named as a DOWN direction in the current study. Most of the length of this corridor has a dedicated nature for the bus operation, and a small part of it has non-dedicated nature, too; hence HDBRTS is considered as a hybrid-based BRT system. The BRT corridor from Hosur Circle of Hubli City to the Jubilee Circle of Dharwad is dedicated in nature, in UP and DOWN directions and the corridor from Hosur Circle Hubli to CBT Hubli is completely non-dedicated in nature. For the current research work, express routes and non-express routes were considered for the route level analysis, and one dedicated segment at the Dharwad side, one dedicated segment, and one non-dedicated segment towards Hubli were considered for the segment level analysis. From the preliminary study carried out for the HDBRTS, it was understood that, higher dwell time, bus bunching at the stations, signal delays at intersections, peak, and off-peak traffic hours of the day were few of the general incidences that were actually influencing the travel time variability of the buses and further leading to the less travel time reliability of the system. Keeping all those points in observance, in the first part of the current study, systematic smart data-based end-to-end travel time variability and reliability analysis have been carried out for the HDBRTS. Analysis has been done for two routes (express and non-express) and three segments exclusively (Two dedicated and one non-dedicated) in two stages. Travel time data points have been extracted for all the days of the week and different hours of the days as different aggregations. In the first stage, descriptive statistics and TTR analysis of the selected data points were done, whereas, in the second stage of the study, probability distribution fitting was carried out for both the routes and selected segments separately with seven potential continuous distributions to characterize the travel time. In the analysis, distribution parameters were extracted using the Maximum Likelihood Estimations (MLE) method. Kolmogorov-Smirnov (K-S) test was used to extract the distribution parameters and check for the goodness of the fit of each distribution. Hence based on the K-S p-value, the robustness of best-fit distribution was selected and iii ranked amongst all the choices, for describing the travel time data points under different conditions considered. In conclusion, as per the total number of cases passed by each selected distribution model, distribution performance was established at different ratios for all routes and segments. At the end of the probability distribution fitting with the travel time data points, the best fit distribution parameters were tried to compare with the passenger demand of that particular time stamp. From the analysis, it was found that peak and off-peak hours have a direct influence on the change in the characteristics of route and segment travel time and subsequent reliability indices. Except for the higher values of reliability indices during peak hours, the performance of the express routes seems to be more reliable. From the distribution study, the Generalized Extreme Value (GEV) distribution stood first on the best performance distributions list for the routes, dedicated segments, and even a non-dedicated segment. Hence it shows the robustness of GEV in explaining the heterogenous Travel Time (TT) characteristics. Based on TTR analysis with GEV distribution, it was inferenced that passenger demand and Buffer Time Index (BTI) have a direct correlation with the variations in the GEV shape parameter ‘k’. In the second part of the study, travel time reliability modelling was carried out with observed and unobserved independent variables obtained from HDBRTS operations. The travel time data points have been extracted according to the selected segments. Modelling was carried out with the Multiple Linear Regression (MLR) technique. Average travel time (ATT) and buffer time (BT) were the two dependent variables chosen from the operator’s and passengers’ point of view. Independent variables were selected based on permutation and combination of multiple covariables. Length of the segment, passenger demand, bus stop density, intersection density, peak and off-peak periods, and land use type were the finalized independent variables. Finally, two MLR models were developed in relation to the two dependent and eight independent variables. The performance of both models was examined with the adjusted R square values and t-statistics and significance values of individual covariables of both the developed models. With the higher adjusted R square values of iv 0.795 and 0.804, respectively, ATT and BT as dependent variables have shown superior explanatory power in describing the system's reliability. In the third part of the current study, as passenger demand forecast for the public transit system is a crucial and inevitable step in keeping the public transit system in the direction of continuous upgrading mode in their performance; hence forecasting of passenger demand was done with Long Short-Term Memory (LSTM) using the three months Automatic Passenger Counter (APC). Then the forecasting of passenger demand was also done with Seasonal Autoregressive Integrated Moving Average (SARIMA) models, and the comparison of the forecasting accuracy of both methods was made using Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). Furthermore, to validate the results, a novel approach has been adopted for the process, by following some more time series resampled with different time intervals. The study shows that LSTMs will be used satisfactorily in the traffic conditions present between Hubli- Dharwad, for forecasting passenger demand using APC data. As the last objective, the travel time reliability-based Level of Service (LOS) of the HDBRTS has been established for three operating conditions, such as route, dedicated segment, and non-dedicated segment. Planning Time Index (PTI), Buffer Time Index (BTI), and Travel Time Index (TTI) were the three reliability indices used to establish LOS. K-mean clustering method was used to develop clusters, and silhouette analysis was carried out to validate the quality of the clusters. Most of the clusters were found to be reasonable and opt with an average silhouette coefficient of more than 0.5. Hence LOS development in the current study better suits with selected data points of travel time reliability indices. Based on the analysis and obtained results of current research work, finally elaborated, three stages of recommendations were made to the operator for improving the performance of HDBRTS.