In some systems which involve human beings, such as management systems or social systems, there are cases that if we can observe the data with human vagueness then the observation becomes easily. Then, the methods which treat these data as fuzzy interval data have been proposed. On the other hand, to investigate the distribution type of observed data, the test of statistical hypothesis has been used. For this purpose, we use the likelihood ratio test in this paper. But, when we treat the fuzzy interval data, the method for usual data has to process the data in which the vagueness is ignored, and there are cases that we obtain the incorrect result by that method. Then, in this paper, we propose the method to process the vague data for likelihood ratio test by considering the vagueness which is included in the fuzzy interval data. Furtheremore, we apply our method to the test of distribution type, and its usefulness is illustrated using computer simulation.

In this paper, we propose a novel method based on artificial life (A-Life) approach. We apply it to prediction of relations among some concepts and to extract causal reasonings which dynamically change. The mechanism in this proposed method is explained as follows. First, agents which have their own rules of causal relations are created from numerical data. Then agents change their rules through their interactions with the other agents and calculate the prediction values of each concept by integrating the rules. The integrated rules (macro rules) are fed-backed to the agents so that the prediction is adapted to the environment. In addition, macro rules are expressed in the form which the users can understand easily. Computer simulation results indicate the following effectiveness of the proposed method : 1)better prediction ability, 2)extraction of dynamically changing rule of causal reasonings, and 3)estimation of difficult prediction.

A lot of methods for solving the fuzzy mathematical programming problem have been proposed. In many cases of them, we translate the original fuzzy problem into a crisp problem. Then, we can solve the translated crisp problem using some ordinal methods. Hence, it becomes one of the important factors how to incorporate the information of fuzzy environment in the original fuzzy problem, to the translated crisp problem. In this paper, we propose a two-phase approach for solving the multiobjective linear programming problem with fuzzy parameters. In the first phase, we solve an α level for the constraints and fuzzy parameters, considering the feasibility of objective functions. In the second phase, based on the optimal α level obtained in the first phase, we search again the ranges of objective functions' values. Then, we solve a well-balanced solution for the original multiobjective fuzzy problem.

There are many large systems composed of several subsystems such as a chemical plant and an iron-manufacturing plant. It is common to formulate these systems by the use of a mathematical programming because an optimal operation of these systems forms a multivariate problem with multiple constraints and purposes. Introducing the new method, m-operator, into a sequential fuzzy linear programming problem and applying it to the control problem of a closed ecological life support system (CELSS) as an example to be solve, we show usefulness of this method.

It is difficult to observe a whole image of object, when a view through fiberscopes or microscopes with higher magnifying power is too narrow to inspect them. So, It is necessary to assemble a set of subpictures automatically into a mosaic picture. This paper presents a method of assembling subpictures using structural information on junctions of branches in the pictures. Moreover, to reduce processing time, an idea of approximate junctions computed by fuzzy reasoning is introduced to decide the image processing area, which contains the optimum numbers of junctions for assembling. Some demonstrations, exemplified by assembling microscopic, are given to show feasibility of this method.

The conventional information processing usually deals with verbal and certain information. A computer needs to deal with non-verbal or ambiguity information as well as verbal and certain information in future. This paper considers a facial expression that is a kind of non-verbal information and expresses human emotion, and constructs an interaction model between the recognition model of emotion and the route decision system as an example of human-computer interaction using facial expressions and situations. The recognition model of emotion recognizes emotions from not only facial expressions but also human situations in which a person is placed. On the other hand, the route decision system reaches an instructed destination or sometimes takes a wrong way, where instructions are given by linguistic expressions. The route decision system also shows its emotions through facial expressions. This paper shows interactions between the recognition model of emotions and the route decision system, which are implemented on two computer systems. Interaction examples show the usefulness of the present model.

Fuzzy modeling is a process to identify input-output relationship of unknown systems. One of the major problems of the fuzzy modeling is the determination of the number of membership functions for each input. Automatic methods for the allocation of membership functions by generating them one by one have been proposed. Other automatic methods were to search for an appropriate allocation of membership functions by deleting them one after another. This paper presents a new method for the allocation of membership functions by inserting/deleting them in the modeling process. The membership functions are unevenly allocated on the universe of discourse by the proposed method depending on the characteristics of the target system. Experiments are done to compare the proposed method with that only adds membership functions and that only deletes membership functions. The results show that the proposed method obtains more concise models with higher generality than the other methods.

In this paper, we propose a method to learn skills automatically by using operation data which human carried out. Tetris is used as an example to study skills. The acquisition and improvement of skills are essential for learning in humans. As for machine learning, the research on examining learning process of human by computer simulations has been carried out. Our objective is to examine acquisition and improvement of skills for Tetris by using Fuzzy Inference Neural Networks(FINNs). FINNs can express the acquired knowledge of the neural network with if-then rules. The rules can be generated automatically by giving the operation data to the proposed system. The skills can be improved by using more operation data. We can analyze acquired skills by examining the rules generated from FINNs. We confirmed the improvement of skills by using more operation data. We could express skills explicitly by the rules that FINNs generated. Also we examined the difference of skills between a veteran and a beginner, and it is suggested that a veteran was doing Tetris considering further steps.

Distribution system loss minimization problem is to accomplish an optimal loss reduction of power distribution system by changing the open/closed states of the sectionalizing switches. This problem is a combinatorial optimization problem, and the number of possible system configurations is extremely huge in practical applications. For this problem, we have already proposed a method based on genetic algorithms(GA), in which the number of infeasible offsprings generated in genetic operations can be decreased, considering four necessary conditions of feasible solutions. However, when the method is applied to large-scale problems, it has a tendency to decrease the probability of obtaining optimal solution. In this paper, we introduce dynamic parameter adaptation technique into our method described above. A variety of genetic parameters, such as crossover rate, mutation rate, and population size, can be automatically tuned by means of fuzzy reasoning rules for the purpose of effective searching in solution space. Experimental results are shown to demonstrate that the proposed method can also solve practical large-scale problems effectively.

The Genetic Algorithms(GAs) based on multi-point search method and crossover operation are one of the useful search procedures for combinatorial optimization problems and also applied to many kinds of practical optimization. However, in general, the GAs have a tendency to go down rapidly of the diversity of population in the process of searching. In order to improve this drawback, some researchers have proposed new algorithms for maintaining the diversity of population. On the other hand, the Immune Algorithms(IAs) are optimization techniques which imitate the immune systems in an organism. The IAs are able to obtain plural semi-optimum solution with maintaining the diversity of population compared with the GAs. In this study, in order to consider the application to optimal design problems in structures, the improvement of convergent and the maintenance of the diversity of population are attempted. Furthermore, improved IA is applied to the impact resistance design problem. It is found that application of improved IA can be used as effective method of structural optimization design base on several simulation results.

This paper proposes an automatic generation technique for distributed simplification fuzzy rules in order to make fuzzy If-Then rules having high general-purpose capability and simple algorithm. The rules are generated by multilayer fuzzy neural network paying attention to fuzzy partitions. This technique improves the fault of conventional methods, which have a constraint in the calculation of membership functions in the antecedent part because the automatic tuning is carried out only to the consequent part parameters in the distributed simplification fuzzy rules. The technique is also simpler than the conventional methods. Results of identification of some nonlinear functions were shown and compared with those by conventional methods. Finally, it was verified that this technique had higher general-purpose capability and higher identification accuracy than those of the conventional methods.

We study the implementation of a meat-quality grading system, using the concept of the "marbling score", as well as image processing, neural network techniques and multiple regression analysis. The marbling score is a measure of the distribution density of fat in the rib-eye region. We investigate as the basis for determining the grade of meat, based on the results of a questionnaire put forth to graders. From the results, we identify five features used for grading meat images. For the evaluation of the five features, we propose a method of image binarization using a three-layer neural network developed on the basis of inputs given by a professional grader and a system of meat-quality grading based on the evaluation of three of five features with multiple regression analysis. Experimental results of grading of unknown images show the proposed method to be effective.