Design of Operational Strategies for Public Bus Transit System Considering Variations in Passenger Mobility Pattern

Abstract

In developing countries, the migration of people to urban areas has led to an increase in the urban population. Growth in urban population directly influences the vehicle population, leading to traffic congestion, pollution and accidents. To overcome this, people should be made to choose public transit over private vehicles by making it more efficient and attractive. The transit system can be made efficient and attractive when the travel time of passengers is reduced. This can be achieved by the implementation of operational strategies including limited-stop services, short-turning, dead-heading and so on. However, before the implementation of these strategies, variability in passenger mobility patterns should be assessed to identify which strategy is best suited for a given corridor. Thus, the present study aimed to design operational strategies for public bus transit system (PBTS) considering variability in passenger mobility patterns. Initially, seasonal variations in passenger mobility patterns were analysed to evaluate the impact of weather on passenger behaviour. Based on the factors influencing passenger flow, passenger demand forecasting models were developed at route and stop levels. Then, the operational strategy was designed for PBTS by integrating limitedstop service with the existing all-stop service. Further, a regression model was developed based on variability in passenger mobility patterns to evaluate the feasibility of limited-stop service before implementing it in any corridor. Based on the outcomes obtained from the analysis, recommendations were given to the transport planners to improve the efficiency of the PBTS. The research work was carried out by considering PBTS from two cities i.e., Davanagere and Udupi of Karnataka state, India. The PBTS of both cities is operated by Karnataka State Road Transport Corporation (KSRTC). The Electronic Ticketing Machine (ETM) is used by the systems to collect ticket data. The ETM data of both cities was procured from KSRTC to obtain passenger flow information. Two cities were considered because of their different weather conditions, making it possible to compare the influence of weather on the variability in passenger mobility patterns. To analyse the impact of different intensities of weather on passenger flow, the weather data of both cities was collected from the Indian Meteorological Department (IMD). Further, land use data was also procured from urban development authorities to evaluate variations in passenger mobility patterns at different spatial regions. The impact of weather on passenger mobility patterns was analysed at different temporal and spatial scales. The change in passenger flow with a change in the intensity of weather was computed to evaluate the impact of different intensities of weather on passenger mobility. The analysis was carried out at system, stop and route levels to have microscopic insights into the relationship between weather and passenger flow. Time-series analysis was carried out at the system and stop levels to evaluate the hourly variations in passenger flow at different intensities of weather. At the system level, the were categorised based on land use and aggregated to evaluate the impact of weather on different spatial regions with different land use. Further, at the route level, the variation in passenger flow between the stops of a route was analysed based on the coefficient of variation (CV). The outcome from the analysis indicated that, at high rainfall, the passenger flow increased in Davanagere and decreased in Udupi. At moderate rainfall, only during weekend variation in passenger flow was observed. The village stops of Davanagere experienced a significant increase in passenger demand during the rainy season. Further, high CV was observed at the routes of Davanagere and Udupi under different weather conditions indicating the concentration of demand among a few stops of a route. These results proved that there is a weather-related impact on passenger flow, but it is region-specific. It was not possible to identify a specific pattern in the relationship between passenger flow and weather. Passenger demand forecasting models were developed with the incorporation of factors influencing passenger flow. The models were developed at the route level and stop level using long short-term memory (LSTM) and graph convolutional neural network (GCN), respectively. Based on the analysis of variability in passenger mobility patterns it was evident that passenger flow is influenced by weather attributes at different temporal and spatial scales. As a result, the LSTM model was developed with the incorporation of weather and temporal attributes and the GCN model was developed with the incorporation of land use as a spatial attribute. In the LSTM model, temporal features such as recent time intervals (R), daily periodicity (P) and weekly trend (T) were included along with the weather attributes (W) and named RPTW-LSTM. Recent time intervals are the most recent steps in time-series data that account for short-term fluctuations in the data. Periodicity refers to patterns that repeat at regular intervals in time-series data. Daily periodicity specifically involves daily cyclic repetitions at the same hour. The weekly trend is encompassed in the model to account for long-term changes observed in the passenger flow data. The output (predictions) of hourly variation, daily periodicity, weekly trend and weather models are fused in the final model. The fused model uses LSTM for multivariate time series analysis, accounting for multiple factors that change over time. In the GCN model, the bus network is constructed as a graph with bus stops were considered as nodes and link between two nodes specifying the passenger flow between the stops. For each node, the land use around the bus stops within a service range of 500m was extracted and included as a node feature to capture spatial correlations in passenger flow. The developed models were compared with the baseline models to evaluate the impact of different features. The developed model showed better accuracy than the baseline models. This indicates that the performance of the model can be improved by incorporating the factors that influence passenger flow. Thus, the development of the prediction models assists transport planners in identifying the possible passenger flow variations in future, based on which transit schedules and frequencies can be designed. The operational strategies were designed for the PBTS by integrating limitedstop service with the existing all-stop service. The frequency of the existing all-stop service was shifted to the limited-stop service to avoid expansion of the fleet size, thereby avoiding additional investments. An optimisation model was developed to optimise stopping strategy and frequency of limited-stop service by maximising the total cost savings (TCS). The optimisation model included maximisation of the operator cost and in-vehicle travel time of passengers and minimisation of waiting time for passengers not served by limited-stop service. The limited-stop service was designed for PBTS of both cities. Genetic Algorithm (GA) was used to optimise the stopping pattern and frequency of limited-stop service by maximising the TCS. During each iteration, a single-point crossover and bit-flip mutation were applied to the population to introduce genetic diversity. To maintain the quality of the solutions, two elite individuals with the best fitness values were preserved in each generation. From the results, it was observed that the implementation of limited-stop service resulted in significant savings for both PBTS. The model strategically included important stops and skipped stops with low demand. Further, it was observed that the savings and stopping patterns were dependent on the number of stops and demand variability, respectively. As a result, to further identify the impact of different characteristics of corridors on the performance of the limited-stop service, a feasibility analysis was carried out. The feasibility analysis was carried out to identify when and where limited-stop service can be implemented. A regression model was developed to evaluate the influence of different characteristics of corridors on the savings incurred due to the implementation of limited-stop services. Based on the factors identified, the adaptability of limited-stop service can be verified before implementing it in any corridor. The scenarios were generated based on variability in passenger mobility patterns at different weather and temporal conditions. The limited-stop service was optimised for each of these scenarios and TCS was determined. Then, cost percentage savings (CPS), achieved due to the implementation of limited-stop service compared to total cost (TC) from all-stop service, was computed. Employing CPS as the dependent variable and various characteristics of corridors as independent variables, a regression model was calibrated. The model was found to have a better fit with the variables considered. The features like CV, dwell time, operator cost and number of stops had a significant positive impact on the model and wait time cost had a negative impact on the model. The variable CV with a positive impact indicates that the limited-stop service could be feasible when variability in passenger mobility patterns is high and it was observed that variability was high in different weather and temporal conditions. Similarly, limited-stop service could be feasible in the corridors where operator cost, dwell time and number of stops are high. However, care must be taken to reduce the waiting time of passengers whose stops are not served. Thus, the developed methodology can be adopted by transport planners to assess the variability in passenger mobility patterns and based on the observed variations different operational strategies can be designed to make the PBTS efficient and attractive.