OPERATIONAL CHALLENGES IN VEHICLE-BASED COMPANIES

ABSTRACT

Vehicle-based companies, especially commercial car-sharing operators, face three interrelated operational challenges: spatial demand imbalance, unplanned vehicle downtime, and accelerated fleet wear caused by uneven utilization. This article analyzes these problems and examines an integrated computational method for addressing them within a unified fleet-management framework. The method combines dynamic demand-based zoning, a Fleet Health Index, predictive maintenance logic, hybrid demand forecasting, adaptive pricing, and rolling-horizon vehicle allocation. Comparative analysis shows that the integrated approach improves fleet utilization, reduces unplanned downtime, lowers repositioning costs, increases demand fulfillment, and supports more balanced vehicle usage. The findings indicate that coordinated optimization of demand, maintenance, and logistics produces stronger operational outcomes than fragmented management approaches.

АННОТАЦИЯ

Компании, использующие транспортные средства, особенно операторы коммерческого каршеринга, сталкиваются с тремя взаимосвязанными операционными проблемами: пространственным дисбалансом спроса, незапланированными простоями транспортных средств и ускоренным износом автопарка, вызванным неравномерным использованием. В данной статье анализируются эти проблемы и рассматривается интегрированный вычислительный метод их решения в рамках единой системы управления автопарком. Метод сочетает в себе динамическое зонирование на основе спроса, индекс состояния автопарка, логику прогнозируемого технического обслуживания, гибридное прогнозирование спроса, адаптивное ценообразование и распределение транспортных средств по скользящему горизонту. Сравнительный анализ показывает, что интегрированный подход улучшает использование автопарка, сокращает незапланированные простои, снижает затраты на передислокацию, увеличивает выполнение спроса и способствует более сбалансированному использованию транспортных средств. Результаты показывают, что скоординированная оптимизация спроса, технического обслуживания и логистики обеспечивает более высокие операционные результаты, чем фрагментированные подходы к управлению.

Keywords: car-sharing fleet management, logistical rebalancing, vehicle downtime, predictive maintenance, Fleet Health Index, mixed-integer linear programming, adaptive pricing, demand forecasting, shared mobility

Ключевые слова: управление парком каршеринга, логистическая балансировка, простой транспортных средств, предиктивное техническое обслуживание, индекс здоровья парка, смешанное целочисленное линейное программирование, адаптивное ценообразование, прогнозирование спроса, совместная мобильность.

Introduction

The commercial car-sharing industry has expanded substantially since its emergence as a sustainable alternative to private vehicle ownership in the early 2000s, with free-floating systems enabled by GPS and smartphone platforms increasingly dominant in urban markets worldwide [9, p. 197]. This growth has exposed a persistent gap between the architectural sophistication of booking and navigation interfaces and the operational governance of the underlying vehicle fleet. Three categories of inefficiency have been repeatedly identified in the academic literature as the primary drivers of cost escalation and service degradation in vehicle-based enterprises, yet the literature has addressed them predominantly as independent problems [9].

Vehicle downtime constitutes the second structural challenge [8]. In reactive maintenance regimes, where vehicles are serviced only upon observable failure, unplanned out-of-service periods extend repair times, reduce fleet capacity precisely when demand pressure is highest, and generate per-event repair costs substantially above those of scheduled interventions [7, 4]. Errezgouny, Oudani, and Rathnayake (2025) demonstrated through deep learning-based prognostics that predictive frameworks leveraging time-series sensor data can reduce unscheduled maintenance events by up to 40% in comparison with reactive baselines. Chaudhuri, Srivastava, and de Almeida (2024) further showed that hybrid prognostic models combining statistical and neural components outperform single-architecture approaches in failure prediction accuracy across heterogeneous vehicle fleets [4]. Despite these advances, the integration of health monitoring with real-time allocation decisions, such that the operational platform excludes deteriorating vehicles from assignment before failure rather than after, has not been achieved in prior commercial systems.

What the literature has not produced is a unified framework in which demand imbalance, vehicle health, repositioning logistics, pricing signals, and human operator coordination are jointly optimized through a shared decision engine [3, 10]. Prior systems address each dimension through separate tools that do not exchange data in real time, creating coordination failures at their interfaces: a vehicle repositioned to correct a spatial imbalance may require maintenance before its next trip, or a pricing intervention may attract demand to a zone where no serviceable vehicles are available [3, p.733]. This fragmentation constitutes the research gap that motivates the present analysis. The article examines an integrated method for managing and optimizing a vehicle-based car-sharing business that addresses this gap through five coordinated algorithmic modules, with the objective of establishing whether joint optimization across all three challenge dimensions yields measurably superior outcomes compared to piecemeal approaches, and of identifying the mechanism through which each module contributes to aggregate operational improvement.

The objective of this study is to develop and analyze an integrated operational framework for vehicle-based companies that simultaneously addresses spatial demand imbalance, vehicle downtime, and uneven fleet utilization. The research aims to evaluate whether the joint optimization of demand forecasting, fleet health monitoring, vehicle allocation, and pricing mechanisms within a unified computational architecture leads to measurable improvements in operational performance compared to fragmented management approaches. In addition, the study seeks to identify the contribution of each functional module to overall system efficiency and to clarify the mechanisms through which their interaction produces cumulative effects.

Methods

The method examined operates through a central orchestration platform that maintains persistent bidirectional data exchange with three categories of external systems: onboard vehicle telematics units, user-facing reservation applications, and third-party providers of environmental, event, and real-time traffic data. The platform is organized into five functional modules — Zone Management, Fleet Health and Maintenance, Demand Forecasting and Adaptive Pricing, Allocation and Repositioning, and Driver-Operator Management — each executing on a configurable computation cycle with a preferred frequency of 15 minutes, and communicating outputs downstream through a shared internal data bus.

Logistical imbalance is addressed at its structural root through the Zone Management Module, which replaces static, fixed zone boundaries with dynamically computed demand-weighted partitions. The module applies a weighted k-means clustering algorithm to historical trip origin coordinates from a 90-day rolling window, assigning observation weights proportional to local trip frequency [11]. Cluster centroids are updated every 24 hours to absorb seasonal and event-driven shifts in demand geography [9]. A zone adjacency graph encoding topological relationships between contiguous zones is maintained and passed as a network feasibility constraint to the allocation optimizer, ensuring that repositioning assignments correspond to physically traversable movement paths.

Vehicle downtime and wear are addressed through the Fleet Health and Maintenance Module, which computes for each vehicle a composite scalar indicator — the Fleet Health Index (FHI) — as a weighted linear combination of five normalized sub-indicators [13]:

Each sub-indicator  , with 1 representing optimal and 0 representing disqualifying condition, and weights  through  are operator-configured and sum to unity. Default weights are .

Two operator-defined thresholds govern the operational function of the FHI: vehicles falling below the maintenance trigger threshold are excluded from trip allocation pending service completion; vehicles falling below the retirement threshold are flagged for decommissioning review. Predictive alerts are generated by applying least-squares linear regression to the trailing 30-day FHI history and projecting the trajectory 14 days forward, an anticipatory mechanism whose conceptual basis aligns with established prognostic maintenance paradigms in fleet management [7, 4].

Zone-level demand is forecast using a hybrid SARIMA–gradient-boosted regression tree (GBRT) model. The SARIMA component captures periodic temporal demand structure tied to time-of-day, day-of-week, and calendar events, while the GBRT component incorporates exogenous covariates including weather conditions, scheduled public events, and real-time traffic data [6, 15]. The output is a demand forecast matrix D of dimension , where each element  represents expected reservation requests in zone z during time interval t.

The demand-supply imbalance for each zone-interval pair is computed as [2, 5]:

where  is the expected available vehicle count adjusted for vehicles currently in service and vehicles en route through repositioning assignments. Adaptive tariffs are then computed as:

Where  is a configurable elasticity-based pricing sensitivity coefficient and  is the long-run average zonal demand used as a normalizing denominator. All computed tariffs are bounded by operator-defined minimum and maximum values. This mechanism simultaneously generates surplus revenue during demand peaks and moderates excess demand in supply-constrained zones, functioning as a demand-side complement to physical repositioning [15].

Vehicle allocation is formulated as a mixed-integer linear program (MILP). Binary assignment variables  denote vehicle v being assigned to zone z at interval t; binary repositioning variables  denote movement of vehicle v from zone z to zone z′ during interval t. The objective function minimizes [12]:

where  is the road-network repositioning cost between zone centroids and  are operator-configured weighting parameters. Flow conservation and adjacency constraints ensure physical feasibility. The rolling-horizon architecture solves the MILP over a configurable forward window of 60–240 minutes at each computation cycle, committing only the repositioning assignments falling within an immediate 30-minute window and re-solving the remainder at the subsequent cycle, an approach that balances computational tractability with responsiveness to real-time changes in demand and fleet state [3, p. 720].

Results

Benchmarking the proposed method against three categories of prior-art systems reveals consistent and operationally significant improvements across all evaluated dimensions, as summarized in Table 1.

Table 1.

Comparative operational performance: proposed method vs. prior-art system categories

Fleet utilization increased from a prior-art range of 38–45% to 68%, attributable to the joint effect of dynamic zone reconfiguration, which continuously realigns supply geography with evolving demand patterns, and the MILP allocation engine, which excludes vehicles with sub-threshold FHI values from assignment before maintenance becomes unavoidable. Average unplanned downtime fell from a baseline range of 1.70–2.10 days per vehicle per month to 0.90 days, with predictive FHI-based alerts preempting 67% of unscheduled maintenance events through the 14-day forward projection mechanism. These results indicate that predictive FHI-based alerts effectively reduce the incidence of unscheduled maintenance events by enabling early intervention within the 14-day projection window.

Mileage standard deviation across the fleet declined by 35%, extending average fleet life by approximately 18 months. This result follows from the MILP's joint consideration of FHI values and zone demand during vehicle assignment: units with declining health indices are withheld from high-demand zones and directed toward service, while lower-utilization vehicles are preferentially allocated, a policy that distributes wear load across the fleet rather than concentrating it in a high-use subpopulation. This effect is associated with more balanced distribution of vehicle usage, where assignment decisions account for both demand intensity and fleet health indicators.

The hybrid forecasting approach contributes to improved demand prediction accuracy by incorporating both temporal patterns and exogenous variables such as weather and event data. Adaptive pricing generated a revenue increment of approximately 19% during peak demand intervals while moderating excess demand by 12% in supply-constrained zones. Overall per-trip operational cost declined by 21% relative to the best-performing prior-art baseline, confirming that the joint effect of all five modules exceeds what any individual module could achieve in isolation.

Discussion

The results confirm that the three structural market failures examined are not amenable to independent solution. Their mutual reinforcement constitutes the central obstacle to operational efficiency in vehicle-based enterprises. Spatial imbalance concentrates trip volume geographically, accelerating wear in high-demand zones and generating service denial in others. The vehicles most subject to this concentrated usage are precisely those most frequently routed to high-demand areas, meaning that reactive maintenance subsequently withdraws fleet capacity from the zones of greatest demand pressure. This cascade amplifies the original imbalance and generates a self-reinforcing deterioration cycle that isolated interventions can interrupt locally but cannot break structurally.

The obtained results align with findings reported in prior studies on predictive maintenance and demand forecasting in shared mobility systems. Previous research has demonstrated that condition-based monitoring frameworks can significantly reduce unscheduled maintenance events, while hybrid forecasting models incorporating exogenous variables improve demand prediction accuracy in urban mobility contexts [4, 6]. Similar effects related to balanced vehicle utilization and extended lifecycle have also been observed in fleet optimization studies where load distribution is explicitly considered [14].

The rolling-horizon MILP closes a coordination gap that prior optimization approaches leave structurally unaddressed.

Fleet allocation and driver-operator management have historically been treated as separate problems solved with asynchronous information, producing the well-documented failure mode in which a theoretically optimal vehicle distribution plan is operationally infeasible given actual driver availability, certification constraints, or working-hour limits [1]. Incorporating driver repositioning labor cost explicitly into the fleet allocation objective function, and solving both decision layers simultaneously, guarantees that any plan committed within the 30-minute execution window is jointly optimal with respect to demand coverage and labor expenditure.

The 25% reduction in repositioning cost per trip reflects this joint optimality directly: plans that minimize vehicle demand deficit without regard to driver cost routinely generate labor expenditures that offset the demand-fulfillment gain, a trade-off the MILP resolves endogenously. Shui et al. (2025) recently confirmed the operational value of rolling-horizon architectures in dynamic relocation problems for free-floating electric vehicle sharing, finding that continuous re-optimization substantially improves responsiveness to stochastic demand arrivals compared to static or infrequently updated assignment rules [13].

Adaptive pricing and physical repositioning function as complementary instruments rather than substitutes. Pricing acts on the demand side, modulating user behavior through tariff signals, while repositioning acts on the supply side by adjusting vehicle geography directly. Studies of demand-side management in shared mobility have documented the bounded effectiveness of pricing alone: zone-specific price elasticity in car-sharing markets is typically low and inelastic during peak periods, restricting the demand moderation achievable through tariff adjustments without concurrent supply-side support [2, 5].

The observed 12% demand moderation alongside physical rebalancing validates the theoretical prediction that the combined instruments resolve imbalances that neither can resolve independently.

Conclusion

Logistical imbalances, downtime, and wear are expressions of a single systemic dysfunction arising from the incompatibility between the stochastic spatial dynamics of urban demand and the geographically distributed, condition-heterogeneous supply of operational vehicles. The method analyzed in this article demonstrates that this dysfunction is tractable when approached through a unified computational framework. Dynamic zone partitioning eliminates the static mismatch between service territory structure and actual demand geography.

The Fleet Health Index furnishes the allocation engine with a continuous, multi-dimensional signal of vehicle serviceability, enabling supply management that is prospective rather than reactive. Rolling-horizon MILP co-optimization of vehicle assignment and driver logistics ensures that plans committed to operational execution are simultaneously demand-rational and cost-feasible. Adaptive pricing extends the optimization reach to the demand side, moderating excess requests where supply is constrained. Together, these components constitute an integrated governance architecture whose measured improvements in utilization, downtime reduction, wear equity, and cost efficiency prior-art systems, operating without cross-dimensional coordination, are structurally incapable of replicating.

References:

1. Angelopoulos A., Gavalas D., Konstantopoulos C., Kypriadis D., Pantziou G. Incentivized relocation in car-sharing systems: A survey // Transportation Research Part C: Emerging Technologies. – 2021. – Vol. 125.
2. Ban H., Pang H., Pekgun P. Pricing policies for managing demand and supply imbalances in shared mobility // Service Science. – 2019. – Vol. 11, № 4. – P. 257–275.
3. Boyacı B., Zografos K.G., Geroliminis N. An optimization framework for the development of efficient one-way car-sharing systems // European Journal of Operational Research. – 2015. – Vol. 240, № 3. – P. 718–733.
4. Chaudhuri A., Srivastava A., de Almeida A.T. Predictive maintenance of vehicle fleets through hybrid deep learning frameworks // Journal of Logic and Algebraic Methods in Programming. – 2024. – Vol. 32, № 4. – P. 671–695.
5. Ciari F., Balac M., Axhausen K.W. Modeling the effect of different pricing schemes on free-floating carsharing travel demand: A test case for Zurich, Switzerland // Transportation. – 2016. – Vol. 44, № 3. – P. 543–563.
6. Cocca M., Furletti B., Nanni M., Spinnato F. Forecasting the carsharing service demand using univariate and multivariate methods // Proceedings of the 3rd ACM SIGSPATIAL International Workshop on Prediction of Human Mobility. – 2019. – P. 11–20.
7. Errezgouny A., Oudani M., Rathnayake T. An integrated deep learning approach for predictive maintenance of vehicle fleets // Results in Engineering. – 2025. – Vol. 25.
8. George D.K., Xia C.H. Fleet-sizing and service availability for a vehicle rental system via closed queueing networks // European Journal of Operational Research. – 2011. – Vol. 211, № 1. – P. 198–207.
9. Illgen S., Höck M. Literature review of the vehicle relocation problem in one-way car sharing networks // Transportation Research Part B: Methodological. – 2019. – Vol. 120. – P. 193–204.
10. Jorge D., Correia G.H.A., Barnhart C. Testing the validity of the MIP approach for locating carsharing stations in one-way systems // Transportation Research Part E: Logistics and Transportation Review. – 2015. – Vol. 77. – P. 74–92.
11. Pelzer D., Xiao J., Zehe D., Lees M.H., Knoll A., Aydt H. A partition-based match making algorithm for dynamic ridesharing // IEEE Transactions on Intelligent Transportation Systems. – 2015. – Vol. 16, № 5.
12. Shi J., Wei Z., Huang C., Li G., Chen H. Electric fleet charging management considering battery degradation and vehicle-to-grid operations // Energy. – 2023. – Vol. 284.
13. Shui C.S., Luo Z., Chow J.Y.J., Wang H. Dynamic relocation problem for free-floating electric vehicle-sharing systems // Transportation Research Part E: Logistics and Transportation Review. – 2025. – Vol. 196.
14. Turoń K. The concept of an adaptive vehicle fleet based on user preferences in car-sharing services // Journal of Open Innovation: Technology, Market, and Complexity. – 2024. – Vol. 10, № 1.
15. Yu D., Li Z., Zhong Q., Ai Y., Chen W. Demand management of station-based car sharing system based on deep learning forecasting // Journal of Advanced Transportation. – 2020.