Sen Kubota

Traveling salesman problem solving method fit for interactive repetitive simulation of large-scale distribution networks

Based on experimental comparison, this paper discusses approximate solution methods of medium-scale traveling salesman problems (TSPs) which suit repetitive use in interactive simulation for globally optimizing a large-scale distribution logistic network. For constructing a globally optimized large-scale logistic network, the problem is decomposed into hundreds of sub-problems, and each sub-problem including TSPs should be repetitively solved. Thus, it is essential to find approximate solution methods of medium-scale TSPs that suit the heavily repetitive use in interactive simulation for globally optimizing a large-scale distribution logistic network. Accordingly, we carried out an experiment for comparison among approximate methods using a random restart strategy that iterates the combination of random initialization and local search. As a result of this experimental comparison, we discovered that one of the approximate methods could obtain solutions ensuring errors below 2-3% within 0.1 second. Thus, this method is considered to be promising for realizing a system that enables one to carry out interactive simulations repetitively for constructing a globally optimized large-scale logistic network.

A Multi-world Intelligent Genetic Algorithm to Interactively Optimize Large-scale TSP

To optimize large-scale distribution networks, solving about 1000 middle scale (around 40 cities) TSPs (traveling salesman problems) within an interactive length of time (max. 30 seconds) is required. Yet, expert-level (less than 3% of errors) accuracy is necessary. To realize the above requirements, a multi-world intelligent GA method was developed. This method combines a high-speed GA with an intelligent GA holding problem-oriented knowledge that is effective for some special location patterns. If conventional methods were applied, solutions for more than 20 out of 20,000 cases were below expert-level accuracy. However, the developed method could solve all of 20,000 cases at expert-level

A multi-inner-world Genetic Algorithm to optimize delivery problem with interactive-time

Building a delivery route optimization system that improves the delivery efficiency in real time requires to solve several tens to hundreds cities Traveling Salesman Problems (TSP) within interactive response time, with expert-level accuracy (less than 3% of errors). To meet these requirements, a multi-inner-world Genetic Algorithm (Miw-GA) method was developed. This method combines two types of GApsilas inner worlds such as a 2-opt type mutation world and an NI type mutation world, randomly selecting either one of these mutation methods (inner worlds) each generation in a GA world consisting of the whole generations. This method is compared with other related works based on experimental results.

A method for solving nested combinatorial optimization problems - a case of optimizing a large-scale distribution network

The optimization of a large-scale distribution network is considered to be a nested combinatorial problem consisting of the following steps: (1) the decision about part delivery volume per part manufacturer; (2) the decision about depots and trucks for the transportation of parts; and (3) the generation of delivery routes for each truck. In such a nested combinatorial problem, a high-level and mathematically strict optimization is desirable as the first step. In addition, at each step, human multi-sided inspection is desired, which requires interactive simulation. Thus, for the first step, a method using linear programming (LP) is proposed. For the second and third steps, a method using a genetic algorithm (GA) is proposed. The latter guarantees interactive responsiveness and realizes expert-level accuracy, through enabling the solution of 1000 mid-scale traveling salesman problems (TSPs) for a distribution network within 30 seconds and within a 3% error. Experimental results proved that the proposed method enables the optimization of a nationwide large-scale distribution network.

A multi-inner-world Genetic Algorithm using multiple heuristics to optimize delivery schedule

Building a delivery route optimization system that improves the delivery efficiency in real time requires to solve several tens to hundreds cities Traveling Salesman Problems (TSP) within interactive response time, with expert-level accuracy (less than 3% of errors). To meet these requirements, a multi-inner-world Genetic Algorithm (Miw-GA) method is developed. This method combines several types of GAs inner worlds. Each world of this method uses a different type of heuristics such as a 2-opt type mutation world and a block (Nearest Insertion) type mutation world. Comparison based on the results of 1000 times experiments proved the method is superior to others.

Constraint pre-checking and gene build-in delaying GA for optimizing large-scale distribution networks

To optimize large-scale distribution networks, it is required to solve large-scale vehicle routing problems (VRP) with in interactive response time, with practicable optimality. To satisfy the requirement, a constraint pre-checking method and a gene build-in delaying GA are proposed. This gene build-in delaying GA postpones the reproduction of genes whose fitness cannot be determined only by local conditions, and preferentially reproduces genes which have stronger relation to just reconstructed genes. Moreover, constraint pre-checking method adopts the constraint propagation method into the GA. This method statistically analyses constraints on the VRP, before starting GA and generates a two dimensional array. This array shows the node, which cannot be contained together in one truck route. Therefore, the calculation efficiency of the crossover and mutation operations is improved by using this array in reconstructing individuals. Our experiments prove that this method enables interactive simulations of large-scale distribution networks.

Locally-Selfish-Gene Tolerant Dynamic Control GA for Time Constraint Delivery Problem

Building ¿just in time¿ distribution systems to improve the delivery efficiency requires solving at least several tens to hundred cities time-constraint Traveling Salesman Problems (TSP) within interactive response time, with practicable optimality. To meet these requirements, a Locally-Selfish-gene Tolerant Dynamic Control GA is proposed. Here, each gene of an individual satisfies only its constraints selfishly, disregarding the constraints of other genes in the same individual. Further, to some extent, even individuals that violate constraints can survive over generations and are given the chance of improvement. Moreover, evolution is promoted by dynamically changing the degree of the tolerance and GA operations. Our experiment proved that this method provides expert-level solutions for several tens to hundred cities time constraint TSPs within a few seconds.

Selfish-gene Tolerant Generic Algorithms to solve large-scale constraint TSPs

Large-scale distribution network simulation applicable to supply-chain management requires to solve hundreds of time-constraint large-scale (max 2000 cities) traveling salesman problems (TSP) within interactive response time, with practicable optimality. To meet this requirement, a selfish-gene tolerant type GA is proposed. Here, each gene of an individual satisfies only its constraints selfishly, disregarding the constraints of other genes in the same individual Further, to some extent, even individuals that violate constraints can survive over generations and are given the chance of improvement. Our experiment proves that this method provides expert-level solutions for time constraint large-scale TSPs within a few seconds