Masaru Tezuka

Linkage Identification by Nonlinearity Check for Real-Coded Genetic Algorithms

Linkage identification is a technique to recognize decomposable or quasi-decomposable sub-problems. Accurate linkage identification improves GA’s search capability. We introduce a new linkage identification method for Real-Coded GAs called LINC-R (Linkage Identification by Nonlinearity Check for Real-Coded GAs). It tests nonlinearity by random perturbations on each locus in a real value domain. For the problem on which the proportion of nonlinear region in the domain is smaller, more perturbations are required to ensure LINC-R to detect nonlinearity successfully. If the proportion is known, the population size which ensures a certain success rate of LINC-R can be calculated. Computational experiments on benchmark problems showed that the GA with LINC-R outperforms conventional Real-Coded GAs and those with linkage identification by a correlation model.

Genetic Algorithm to Optimize Fitness Function with Sampling Error and its Application to Financial Optimization Problem

In this paper we discuss the optimization problems with noisy fitness function. On financial optimization problems, Monte-Carlo method is commonly used to evaluate the optimization criteria such as value at risk. The evaluation model is often very complex which needs considerable computational overheads. In order to realize efficient optimization of financial problems, we propose a method to decide the number of samples used to estimate the optimization criteria. Selection efficiency proposed in this paper is a index that shows how close the population approaches to the convergence to a good solution. In general, it is difficult to calculate selection efficiency analytically. Thus we also employ bootstrap method to estimate selection efficiency. The resulting algorithm is applied to the optimization of the procurement plan optimization problem. The result shows that value at risk of the problem is optimized efficiently by the proposed method.

Modeling manufacturing resources based on agent model

In order to improve the responsiveness to changes and uncertainties around production activity, we should have the integrated methodology which allows us to uniformly model and design the total system including production and planning resource and related information exchanged among resources. We propose the agent model and the modeling methodology. There are two types of the agent in our model. One is a service agents that offers some processing functions. The other is a mobile agent that moves among service agents. A manufacturing resource is modeled as one of the service agents since it provides manufacturing functions. On the other hand, each lot or production instruction information controlling production progress can be model as one of the mobile agents.

Maintenance schedule optimization based on failure probability distribution

Organizations related to infrastructure, such as utilities and railway companies, manage a large number of facilities, the failure of which can have a huge impact on society. The cost of maintaining these facilities is a combination of regular maintenance costs and urgent recovery costs. Generally, the urgent costs are much higher than regular costs. Regular maintenance work should result in fewer sudden failures, and thus reduce these urgent costs. However, if the regular maintenance is too frequent, its cost becomes too high. Therefore, it is important to balance the regular and urgent costs to minimize the overall maintenance cost. We propose a maintenance schedule optimization method based on the failure probability distribution of the facilities. The total cost is mathematically modeled, with the regular maintenance schedule included via decision variables and the occurrence of failures modeled as stochastic variables. The stochastic total maintenance costs are evaluated using a Monte Carlo method, and a genetic algorithm is employed to optimize the maintenance schedule. The proposed method is evaluated using data provided by a Japanese railway company, and our results confirm that the method produces an excellent maintenance schedule. A statistical test shows there is a significant difference between the proposed and conventional methods.

Demand forecasting model of service parts with different failure rate

Manufacturers are required to provide repair service for their users after sales. Therefore manufacturers usually hold inventory of service parts for many years to keep their repair service level satisfactory. On the other hand, they have to decide the production volume of service parts before the assurance period expires because they cannot keep the production facility for each product for many years. Thus, demand forecasting of service parts close to the end of the assurance period is becoming more important. An accurate demand forecasting enables manufacturers to keep appropriate inventory level of service parts for the remaining period. In this paper, we propose a new demand forecasting model. Our model is developed on the basis of the three factors: the failure rate of a part, the replacement rate of a failed part and the declining rate of the population of products in use. The failure rate is derived from mixing the Weibull distribution. The replacement rate and declining rate are modeled as exponential function. We examined the forecasting performance of the proposed model, using actual shipment records of service parts at a home appliance manufacturer. The results show that the proposed model outperforms the conventional model based on renewal theory.

FORECASTING TOTAL SALES OF HIGH-TECH PRODUCTS - Daily Diffusion Models and a Genetic Algorithm

A vertebral axial decompression table is operated by applying a baseline tension to the two table parts, increasing tension to about 50% of maximum above baseline, then logarithmically increasing tension to maximum tension. Thereafter, tension is linearly relaxed back to baseline. This cycle is repeated a programmed number of times to effect a therapy session. Data concerning the table operation is transmitted to allow remote monitoring and re-programming of the table.

A New Genetic Representation and Common Cluster Crossover for Job Shop Scheduling Problems

This paper describes a genetic algorithm approach for sequencing problems especially for job shop scheduling. In actual problems, setup time should be optimized. In order to reduce setup time, we developed a new genetic representation and an efficient crossover operator called Common Cluster Crossover (CCX). In our representation, chromosomes represent the shift of the order in sequence. To preserve sub-sequences in crossover operations, we implemented the process to identify the cluster of the sub-sequences and applied CCX that exchanges common clusters between two parents. The approach was tested on two standard benchmarks and applied to and audio parts manufacturer. CCX achieved remarkable results on the actual job shop scheduling problem.