Electrifying transport, particularly through battery electric vehicles (EVs), is essential to decarbonise the sector. Light commercial vehicles (LCVs) are the third-largest contributor to road transport GHG emissions. Electrification is promising, but LCV operators face challenges in planning charging capacity, managing infrastructure, and integrating fleet charging with the electricity system. The links between charging-infrastructure planning, operations and electricity costs remain underexplored, especially when smart-charging and infrastructure investment are considered together. This study proposes an integrated Mixed-Integer Linear Programming model to optimise charging infrastructure sizing and operational planning for electric LCV fleets. The model selects charger numbers and types at the distribution terminal and optimises vehicle charging schedules within operational, technical, and economic constraints. To address limited data availability, 15 case studies are generated using a schedule generator with different operating patterns. Results demonstrate that operational behaviour influences infrastructure requirements, with the optimal charger-to-vehicle ratio ranging from 15% to 30%. Slow charging at the distribution terminal is the most cost-effective solution, while on-route charging is necessary only for high-intensity fleets. The analysis highlights the importance of optimising both infrastructure investment and charging operations, including peak power demand, showing how fleet scheduling decisions affect electricity-network requirements. The findings prove that smart-charging and operational flexibility can reduce infrastructure needs and system costs, supporting evidence-based planning. For fleet operators, the results show how schedules affect charger requirements and operating costs. This framework enables efficient, grid-responsive electrification and gives fleet operators and planners a practical way to assess infrastructure and operational decisions across varied fleet profiles.
Shadow driving refers to the informal operation of commercial freight vehicles by truck assistants who are not legally licensed or otherwise authorized to drive. Although the practice appears to occur within Bangladesh’s freight sector, empirical evidence remains limited. This study examines factors associated with Shadow Driving Practice (SDP) and its relationship with Perceived Road Safety Risk (RSR). A cross-sectional survey of 492 truck assistants was conducted across major freight settings in all eight administrative divisions of Bangladesh using purposive recruitment and self-reported measures. Partial Least Squares Structural Equation Modeling (PLS-SEM) was used to evaluate Career Aspiration (CA), Economic Pressure (EP), Weak Enforcement (WE), Perceived Risk (PR), Driver Fatigue (DF), SDP, and perceived RSR. Predictive performance was assessed using PLSpredict and the cross-validated predictive ability test (CVPAT), while Importance–Performance Map Analysis (IPMA) was applied to examine relative importance and performance. EP showed the strongest positive association with SDP (β = 0.290), followed by WE (β = 0.255), DF (β = 0.236), and CA (β = 0.224), whereas PR was negatively associated with SDP (β = − 0.165). SDP was strongly associated with perceived RSR (β = 0.729). The model explained 52.4 % of the variance in SDP and 53.1 % of the variance in perceived RSR. PLSpredict indicated positive predictive relevance for both endogenous constructs, and CVPAT showed that the PLS-SEM model outperformed benchmark models under repeated cross-validation. IPMA further identified SDP as the most important construct for perceived RSR, with EP showing the greatest importance among the upstream predictors. The findings suggest that shadow driving reflects a combination of economic, regulatory, occupational, cognitive, and operational conditions rather than a single dominant factor. This study provides an empirical framework for an under-studied form of informal freight operation and points to formal assistant-to-driver training pathways, improved employment conditions, credible regulatory monitoring, fatigue management, and targeted safety education as relevant policy directions. Given the cross-sectional, self-reported, and purposively sampled design, the results should be interpreted as statistical associations and internally cross-validated predictive evidence, not as causal effects or objectively observed road-safety outcomes.
The study’s primary contribution is to generate emission factors using a representative driving cycle (RDC) for Dhaka, one of the most congested cities in the world. GPS data from 665 trips conducted within Dhaka from light duty vehicles were processed into the second-by-second speed-time profile. Markov chains using dynamically adjusted transition probabilities synthesized RDCs from speed-time profiles for different driving conditions. The generated RDCs were incorporated in the International Vehicle Emission (IVE) model to estimate the emission factors (EF) of CO, NOx, and CO2 across different relative humidity and temperatures. ANOVA tests for different pollutants revealed that weekday and weekend EFs differ significantly at a 95% confidence level. Additionally, the study found a significant impact of stop-and-go traffic conditions – a typical characteristics of Dhaka traffic – on emission factors. Increasing stop-and-go intensity significantly elevated vehicular emissions, with the highest stop-and-go class exhibiting average CO and NOx emission factors that were 50.3% and 18.7% higher, respectively, than those under low stop-and-go conditions. A case study with EV-adoption scenarios indicated that achieving the unconditional 1.90 MtCO2eq reduction target would require approximately 20.7% replacement of older Internal Combustion Engine (ICE) vehicles and 26.5% EV adoption among new registrations. Achieving the combined 6.00 MtCO2eq reduction target would require corresponding rates of 70.5% and 76.2%. CO2 emission savings decreased when well-to-wheel scenario that included CO2 emission for electricity generation required for charging EVs is considered. These findings highlight the importance of disaggregated RDCs for capturing stop-and-go emission effects and their application for policy analysis.
Accurate ridership forecasts underpin public transport planning, resource allocation, and resilience, yet forecasting models often fail when demand undergoes sudden structural change. This study examines how the relative performance of statistical and machine learning models changes across demand regimes, using monthly Gautrain ridership data from April 2017 to March 2025. Three statistical models (ARIMA, SARIMA, and Holt-Winters) and two machine learning models (Random Forest and Gradient Boosting) were evaluated over 24 successive quarterly forecast tests using an expanding training window. Structural-break tests confirmed regime boundaries at April 2020 and July 2022, separating pre-COVID-19, COVID-19 disruption and post-COVID-19 regimes, and Diebold-Mariano tests assessed whether accuracy differences were statistically significant.Forecasting performance was regime dependent. Random Forest recorded the lowest errors pre-COVID-19, Gradient Boosting during disruption, and SARIMA in the post-COVID-19 regime, where it was most accurate on all four error measures. Gradient Boosting achieved lowest RMSE across the full period. However, differences between models were not statistically significant before or during the disruption; afterwards, SARIMA and Gradient Boosting and Random Forest significantly outperformed ARIMA. No model performed best in every validation window, and the most accurate model on average was not the most consistent.This study contributes a regime-based validation framework combining walk-forward validation, structural-break testing and significance testing, and provides rail-specific evidence from an emerging economy where exogenous data are scarce. The findings indicate that model selection should depend on the prevailing demand regime rather than on the assumed superiority of either model class, and that models should be judged on consistency and statistical significance as well as average accuracy.
The COVID-19 pandemic accelerated remote work, online shopping, and home-based activity, raising questions about how passenger travel and residential goods movement are evolving together. This study examines whether changes in physical workplace travel and online home delivery reflect interconnected and asymmetric adjustments in household activity organization. Using 6,186 employed respondents from the 2022 NextGen National Household Travel Survey, we analyze reported changes in both activities relative to pre-pandemic conditions and distinguish the direction of change from respondents’ perceived durability. Although “about the same as before” is the most common response for both activities, respondents reporting change predominantly indicate increased online home delivery and reduced physical workplace travel.A bivariate ordered probit model, supplemented by average discrete changes in predicted probabilities, reveals substantial socioeconomic asymmetry. Higher-income and advanced-degree respondents show greater movement toward online home delivery and reduced workplace travel, whereas lower-income respondents exhibit the opposite pattern. Advanced-degree and higher-income respondents have approximately 10–11 percentage-point higher probabilities of reporting reduced physical workplace travel than their respective comparison groups. Mobility resources, lifecycle, and geographic context introduce additional differences across the two activities. A significant negative cross-equation error correlation further indicates that delivery and workplace-travel adjustments remain related through unobserved factors after accounting for observed differences. Rather than supporting direct trip-for-trip substitution, the combined evidence points to interconnected but uneven reorganization of work and consumption activities. These findings support jointly considering passenger travel and residential goods movement while accounting for socioeconomic and spatial heterogeneity in transportation forecasting and scenario analysis.
