Car-sharing services face challenges in balancing fleet availability due to fluctuating demand. This paper simulates car-sharing demand using probability distributions, incorporating factors like daily and weekly patterns, seasonal trends, and external influences such as weather and public holidays. Model training considers convergence speed since it dictates how quickly algorithms arrive at solutions, which affects deployment time and computational resources. Using numerical analytical techniques and optimization frameworks, this research looks at how optimization is controlled in the dynamic field of machine learning. The simulation provides a flexible framework for demand analysis, aiding in more efficient fleet management and forecasting.
Car-sharing has become a crucial solution to urban mobility challenges, providing efficient
resource utilization and access to transportation. However, operators face fleet imbalance, where
the supply of cars does not align with fluctuating demand. Accurate demand prediction is essential
for addressing this issue, but there is often insufficient data available for car-sharing services, using
time series simulation tools [
To overcome this limitation, this paper focuses on simulating car-sharing demand [
2.1.
A time series consists of sequential observations recorded at uniform time intervals. These records
—whether measurements, counts, or other numerical data—are arranged in order of occurrence.
Examples include weather readings, financial market data, and heartbeat signals. The proliferation
of sensors and tracking systems has led to a surge in sequential data collection, making time series
data both widespread and vital. Such data helps in discovering causal relationships, identifying
trends and even prediction of next values in many applications such as, but not limited to,
medicine, weather-forecast, and economy deciding equivalences [
Data simulation is the process with which synthetic datasets are generated that resemble the
statistical properties and trends of real data. For instance, simulating visitors to a restaurant may
create a bimodal distribution that reflects lunch and dinner hours. This method is commonly used
in statistics, economics, finance, social sciences and engineering, to test a model or hypothesis
before using it on real data [
3.1.
Drawing on the analysis conducted by [
This scenario adopts a free-floating model; however, the simulation here does not distinguish
between one-way, round trip, and free-floating services. Instead, its primary objective is to
accurately replicate real-world behavior by incorporating all essential parameters for an effective
and scalable simulation model [
The figure above isolates hourly demand trends for Monday and Tuesday, offering a closer look
at daily car sharing usage. Both days display similar patterns, indicating consistent behavior that
could impact operational planning and resource allocation [
Our simulation is currently designed around a fixed fleet size of ~400 vehicles, which considerably limits the solution space for type of operational scenarios. Therefore, to create a more generic solution, we need to go beyond those limitations and enable variable fleet sizes. A proper scalable simulation should allow the users to configure a fleet that best fits their requirements. This will involve augmenting the model by incorporating fleet size as a time-varying parameter. This central change reengineers the primary mechanics of the simulation, so that hourly demand numbers directly react to the number of available vehicles. So, as the fleet expands/shrink, it will definitively paint a more realistic picture of what is happening in the world (at least in the simulation). The last result simulated dataset will have one additional column for fleet size now, giving a complete picture of what vehicles mean in terms of their market presence, offering a comprehensive view of how vehicle availability influences demand dynamics.
The final stage of the simulation has a notably unique feature that noticeably enhances the simulation dynamics: mobility in fleet size, which encourages the simulation to be more realistic according to the operational aspect. The upgrade allows stakeholders to investigate the impact of altering fleet sizes on meeting demand and assist car-sharing companies optimize resource allocation.
We address this gap by developing a scalable, robust simulation model of car-sharing demand, which is contextually tied to real-world parameters like hourly utilization profiles, weather effects, and fleet size constraints. It then uses mathematical models, such as cosine and bimodal distributions, to reflect daily and weekly demand variations in the simulation.
Incorporating the probabilistic perspective gives a better representation of demand fluctuations over longer periods. Moreover, the incorporation of weather data improves each model's realism, accounting for the role of variable weather in the usage of car-sharing. To reflect monthly or annual demand changes in the simulation, we can use mathematical models, such as sine and modal probability distributions or even other models to the next future. These results contribute to the recent literature on demand forecasting, provide guidance for operational planning, and demonstrate how simulation can support decision-making in economic trends, or city-specific mobility policies, to enhance predictive accuracy.
The authors have not employed any Generative AI tools. [11] demand