Distributed Simulation-based Optimization for Resource Allocation of Multimodal Operation System on Container Terminals

Faculty of Symbiotic Systems Science, Fukushima University, Fukushima City, 960-1296, Japan This paper focuses on efficiency of simulation-based optimization for container terminal operation system. Firstly, the discrete-event simulation model of multimodal operation system on container terminal was presented with the queuing network, for which the mathematical optimization model of resource allocation was formed. To solve the huge computational effort problem generated by the considerate simulation time cost, a distributed simulation-based optimization method combined simulation model with distributed optimization was proposed. Finally, the feasibility and effectiveness of the proposed method was validated and analyzed with an actual example. The results show its feasibility for the efficiency improvement of simulation-based optimization on resource allocation of the multimodal operation system on the container terminals. INTRODUCTION Simulation models often support the analysis of complex discrete systems, such as container terminal multimodal operation system. Recently, when the project aims to obtain a better solution based on the existing simulation model, simulation-based optimization is an effective method to solve optimization problems which are random or hard to describe by mathematical formula. Unfortunately, the greatest obstacle it suffers is the huge computational cost, shown as follows: 1) Huge solution spaces of a certain optimization algorithm correspond to a great many simulation experiments. 2) Each individual simulation is time-consuming 3) For each solution, a large number of simulation replications are usually required to eliminate the random influence. Such tremendous simulation time makes it not feasible enough to some application. Therefore, it has a positive and practical significance to reduce computing cost of simulation execution. Until now, research on the problem raising is divided into two kinds: One way is to reduce overall computational demand, including controlling the simulation sample amount by sample average approximation [Lee], guiding the subsequent heuristic searching of the algorithm using sensitivity analysis [Chen]. Although the computational cost can be reduced to a certain degree, it still limits to a certain problem in a certain domain. Another method focuses on the distributed simulation environment by using parallel computing technologies incorporated into the solution process. It can cut down runtimes considerably and is more suitable to cope with more complex simulation-based optimization problem. The main distributed/parallel technologies/implementation platforms are parallel virtual machine (PVM) [Pierreval], Java RMI framework [Gehlsen], web service technology [Luo], message passing interface (MPI)[Li], grid technology [Grauer], and so on. At present, study on multimodal operation system simulation on container terminals centralizes on the system modelling methods, but no concern for the time efficiency of the simulation optimization process, which sets back to the application and development of simulation-based optimization. In terms of improving the time efficiency, this paper focuses on the distributed simulation-based optimization (DSO) approach to solve the coordination of resource allocation of multimodal transport system on container terminals, which is supported by the HLA (high level architecture) technology. SIMULATION-BASED OPTIMIZATION MODEL Simulation Model By analysis on the multimodal operation process on container terminals, the discrete event simulation model with queue network is established. Fig. 1 shows the resources structure based on queue theory, the ring represents customs of the system, and the rectangle represents resource of the system. Based on the above, the coordination optimization model for resource allocation of multimodal transport system is formulated as follows. Vessel Inner Truck Out Truck