|
Computational intelligence-based model for passenger demand forecasting and fleet management of autonomous taxis in smart cities |
|---|---|
| รหัสดีโอไอ | |
| Title | Computational intelligence-based model for passenger demand forecasting and fleet management of autonomous taxis in smart cities |
| Creator | Adeel Munawar |
| Contributor | Mongkut Piantanakulchai, Advisor |
| Publisher | Thammasat University |
| Publication Year | 2567 |
| Keyword | Smart cities, Machine learning, Travel demand forecasting, Federated learning, Vehicle dispatching, Demand uncertainty, Vehicle trajectory, Intelligent transportation systems, Autonomous taxis, Sustainable cities and communities, Urban transportation optimization, SDG 11 |
| Abstract | Transforming cities into smart cities is an essential and aspirational goal in the rapidly evolving technology landscape. An integral part of this change is the emergence of Autonomous Taxis (ATs), which represents a significant development in the taxi service industry. These ATs systems hold the potential to transform urban mobility by mitigating traffic congestion and reducing environmental pollution. Beyond their environmental benefits, they promise to enhance urban residents’ overall quality of life. Despite their potential, the effective management of ATs, particularly in passenger demand forecasting and dispatching, poses significant challenges due to privacy concerns, data sharing limitations, and the dynamic nature of urban mobility demands. This study addresses these challenges by focusing on two key areas: forecasting ATs demand and optimizing ATs fleet management. In this study, we proposed a model that integrates computational intelligence approaches to enhance the operational efficiency of ATs in smart cities. The proposed model integrates a novel collaborative privacy-preserving Federated Learning (FL) approach for demand forecasting. It enables ATs across various city regions to refine their forecasting models collaboratively without direct data sharing, thereby reducing privacy risks and communication overheads. It integrates a centralized agent-based model to optimize the dispatching of ATs, with an emphasis on maximizing fleet profitability while considering both minimizing distance traveled and reducing passenger waiting time. The model employs proactive decision-making to predict demand patterns and dynamically reposition vacant ATs to high-demand areas, ensuring efficient and profitable fleet management. We evaluate the proposed model using a real-world dataset of over 4,500 taxi trips in Bangkok, Thailand, by using MATLAB R2024b. The results of our simulations demonstrate significant operational improvements over traditional dispatching and demand forecasting approaches. Specifically, our collaborative privacy-preserving FL approach has been proven effective in enhancing demand forecasting accuracy for ATs while preserving the privacy of passenger data. We use several backpropagation neural networks as local models for collaborating to train the global model without directly sharing their data, addressing privacy concerns and communication costs efficiently. Our approach significantly outperforms existing methods regarding model accuracy, privacy preservation, and key performance metrics such as RMSE, MAE, and R-squared (R²). Concurrently, our centralized agent-based model effectively optimizes AT dispatching. It dynamically reallocates vacant ATs to areas with demand gaps, directly addressing the challenges of fluctuating spatiotemporal supply-demand dynamics. Validated through various approaches, including random assignment, modified ant colony optimization, proximity greedy heuristic, hill climbing, and beam search models. The simulation illustrates the proposed model using real-world scenarios of the demand pattern in Bangkok, Thailand. Results show that the proposed model outperforms other approaches, achieving higher fleet profitability, reduced waiting times, and shorter travel distances. This study contributes significantly to intelligent transportation systems and supports the vision of sustainable smart cities. |
