The Dial-a-Ride Problem (DARP) is a well-known routing problem in the field of shared mobility systems. In this problem, instead of isolated trips, people share their rides. In addition to the operational constraints, the convenience of passengers is a primary concern in the DARP, which leads to a variety of timing restrictions. In this regard, the feasibility checking procedure is an essential component of solving methods for the DARP. In this paper, utilizing a novel shortest path algorithm, a linear-time feasibility checking technique is introduced that takes advantage of difference constraints systems. This method can also be employed to minimize the route duration value and to find the earliest and latest feasible departure times of the nodes in a DARP path with O(n) time complexity. Thus, the DCFA could also be used to accelerate procedures in the literature that rely on these timing values. In comparison with some other feasibility checking techniques in the literature, our method is substantially easier to comprehend and implement. Moreover, it outperforms these techniques in terms of time complexity. Beyond this theoretical superiority, experimental results indicate the efficiency of this algorithm in practice, particularly when large DARP paths are considered.
Automated valet parking is a use case for autonomous driving. It allows customers to leave their vehicles at the entrance of a parking garage. The vehicles then independently steer to a parking space. Besides increasing comfort, automated valet parking also enables parking garage operators to provide additional services, such as charging an electric vehicle, to which the vehicle can be autonomously routed. We formulate automated valet parking service provision as a scheduling problem and propose solution methods that focus on runtime and quality of results. The results show that the proposed optimization approach is capable of slightly outperforming heuristic benchmarks in many scenarios. However, we also find that the First-Come-First-Served and Earliest-Due-Date-First heuristics are often satisfying scheduling approaches. In addition, the potential of optimizing automated valet parking service provision is compared to traditional parking garages, showing at least a 46.4
Resource allocation in 5G communication systems is referred to as the process of logically allocating available network resources to end-users, devices, and applications at any point in time in order to provide effective network performance and quality of service. The initial systems employed the semi-static approach in resource allocation, which proved to be ineffective in meeting real-time network changes and hence failed to use the resources effectively. Also, the existing task scheduling approaches lacked real-time responsiveness, causing delays in execution, failed to meet strict task deadlines, and poor user experience. These issues are overcome in this research by developing a hybrid deep reinforcement learning framework for joint resource allocation as well as adaptive task scheduling. A novel Revolution Crisscross scheduling Optimization Algorithm (RCOA) is introduced to minimize makespan and maximize system throughput by dynamically assigning computational tasks across edge and core resources. Also, it operates through a dual-space model, switching among global exploration and local exploitation for efficient load balancing and adaptive policy evolution. Additionally, this research proposed the Deep Quantum Supercell Reinforcement-based Support Vector Regression (DQSR-SVR), which combines the deep reinforcement learning algorithm with a quantum support vector regression model to detect and fine-tune resource allocation under the conditions of latency as well as load. The supercell optimizer further improves the convergence and global efficiency. Furthermore, the proposed DQSR-SVR model achieved the lowest average time delay of 0.25 min, latency of 0.55 min, execution time of 200 ms, makespan of 750 s, and highest throughput of 99
This paper opens a special issue of the Journal of Scheduling, devoted to selected papers presented at the Fourth International Workshop on Dynamic Scheduling Problems, held in June 2023 in Winterthur, Switzerland.
Demand for large area coverage is pervasive over a variety of emergency response application domains. Using multiple Earth observation satellites to collect information through image acquisition in efficiently covering such a task is increasingly drawing attention. But multi-satellite collection scheduling, a NP-hard problem, still faces many computational challenges. In this paper, a new approach combining a single objective multi-satellite collection scheduling decision model and an innovative 2-stage algorithm is proposed to solve the large area coverage problem. The proposed mixed overlapping subarea composition coverage (MOSAIC) approach aims at maximizing collection value defined over the quality of individual image acquisitions weighted by their corresponding relative swath coverage, while minimizing area overlap. The underlying MOSAIC algorithm first maximizes collection value using mutually non-overlapping swaths and then fills remaining gaps while minimizing overlap. The algorithm initially partitions each imaging opportunity swath segment in sub-opportunities (time subintervals) forming contiguous subareas in order to reduce mutual overlaps. Following the 2-stage procedure, it then recombines a mixture of segmented swath elements to near-optimally maximize collection value in covering the original large area. The resulting solution is a mosaic of swath subarea segments. A computational experiment conducted on the basic large area coverage maximization problem proves the value and quality of the approach exhibiting small optimality gap.
