Journal of Intelligent & Fuzzy Systems: Applications in Engineering and Technology

This paper reviews genetic algorithms and considers their application in the domain of attributed point pattern matching, specifically star pattern recognition. Conventional algorithms employed in this area are first reviewed and the suitability of the genetic algorithm approach considered. A test environment is built and used to provide feedback on the viability of this approach. Algorithm testing is carried out using this environment, and results compared with the performance of the traditional star pattern search algorithms. The genetic algorithm approach discussed in this paper is shown to be a viable alternative to the conventional search algorithms (Benelli and Mecocci, ESA J. 17 (1993), 185–198; Murtagh, Astronom. Soc. Pacific 104 (1992), 301–307; Junkins and White, J. Astronom. Sci. XXV(3) (1977), 251–270), achieving on average fewer multiple matches and producing 'higher quality' solutions.

In this paper a method of designing a neural pattern recognition system for a rotated coin recognition problem using a genetic algorithm (GA) with deterministic mutation (DM) and partial fitness (PF) is presented. In this method, chromosomes of individuals in the GA are divided into several parts and their PF functions are evaluated for GA operations. Furthermore, the DM, which is based on neural network learning, is introduced. The DM can evolve chromosomes of individuals to increase their fitness functions in a deterministic manner. In the pattern recognition system described in this paper, the Fourier transform is used as a preprocessor which produces rotation invariant features. These features are recognized by a multilayered neural network. The GA is utilized to reduce the number of signals, Fourier spectra, and input to the neural network. This approach using the GA is a type of feature selection problem. It is shown that the present method is better than conventional GAs with respect to convergence in learning, and results in the formation of a small neural network.

In the field of failure diagnosis of plant machinery, one of the most important and most difficult factors is the identification of Symptom Parameters (SP). Failures of machinery can be sensitively detected and the failure types can be distinguished by using the optimum SP. Currently, however, there is no acceptable method for extracting the optimum SP. In order to overcome this difficulty and ensure highly accurate failure diagnosis, in this paper a new method called "Self-reorganization of Symptom Parameters" has been proposed by using Genetic Algorithms (GA). The new method can also be applied to other pattern recognition problems. It has been proved that the optimum SP can be quickly discovered by applying the method to many practical machinery diagnoses.

In this paper, we propose a hybrid model of the fuzzified Kohonen's Self-Organizing Map and the Genetic Algorithms with numerical chromosomes, and automatic fuzzy rule extraction method that uses our model. The results show the possibility of superiority of our hybrid model in a Lamarckian stance to both of the individual models in cases where there is a tendency for data to change dynamically and quickly.

In our research, the evolutionary algorithm is applied to behavior learning of an individual agent in multi agent robots. Each robot, which is an agent, is given two behavior duties, collision avoidance from other agents and target (food point) reaching for recovering self-energy. Addressing the problem of two conflicting behaviors, collision avoidance and target reaching motion of multi-agent robots, the learning method to change the self-energy and the behavior gain of each agent is discussed in this paper. Each agent has the same rules and is controlled as a homogeneous distributed system without any central or hierarchical control. Furthermore, we perform a simulation with the additional algorithm of a group evolution in which the parameters of the most excellent agent are copied to a dead agent, that is, an agent that has lost its energy. The simulation confirmed that each agent has the abilities of behavior learning and group evolution.

We had presented an automatic generation technique for fuzzy rules using hyper-cone membership functions by genetic algorithms (GA). However, there remain some problems. The shape of fuzzy subsets is limited to a hyper-sphere. Since there was not a regulation which determines the order of rules, two rules having different antecedent part structure are crossed in Crossover, and high-performance rules may not be inherited in the next generation. In this paper, we expand the shape of fuzzy subsets to be elliptic and present an automatic generation technique for fuzzy rules using hyper-elliptic-cone membership functions by GA. We also present a rules sorting technique which efficiently reduces the number of rules and obtains high-performance fuzzy rule sets. We applied presented methods to a line pursuit control problem and a trailer-truck back-up control problem. The method using hyper-elliptic-cone membership functions can obtain very accurate fuzzy system with as high performance as the method using hyper-cone membership functions. The technique of a sort of rules can not only reduce the number of rules, but also can get a high-performance rule set.

Compared with the conventional approaches, the evolutionary algorithms (EAs) are more efficient for system design in the sense that EAs can provide higher opportunity for obtaining the global optimal solution. However, in most existing EAs, an individual corresponds directly to a possible solution, and a large amount of computations is required for designing large-scaled systems. To solve this problem, this paper proposes a co-evolutionary algorithm (CEA). The basic idea is to divide and conquer: divide the system into many small homogeneous modules, define an individual as a module, find many good individuals using existing EAs, and put them together again to form the whole system. To make the study more concrete, we focus the discussion on the evolutionary learning of neural networks for pattern recognition. Experimental results are provided to show the procedure and the performance of the CEA.

In this paper we describe an evolutionary fuzzy system to control a mobile robot effectively, and apply it to the simulated mobile robot called Khepera. The system gets input from eight infrared sensors and operates two motors according to fuzzy inference based on the sensory input. In order to robustly determine the shape and number of membership functions in fuzzy rules, genetic algorithm has been utilized. This approach reduces the burden of human operators to decide the structure of fuzzy rules. With the simulation of Khepera robot, we confirm that the evolutionary approach might find out a set of optimal fuzzy rules which make the robot to reach the goal point, as well as to solve autonomously several subproblems such as obstacle avoidance and passing-by narrow corridors.

Power plant start-up scheduling is aimed at minimizing the start-up time while limiting turbine rotor stresses to acceptable values. In order to increase on-line performance of searching an optimal or near-optimal start-up schedule during power plant operation, we propose to integrate neural network-based reinforcement learning with evolutionary computation implemented by means of Genetic Algorithms (GA). GA guides reinforcement learning to learn optimal schedules with respect to a number of representative sets of stress limits prior to the start-up process. During start-up, GA combined with reinforcement learning will search an optimal or near-optimal start-up schedule at a given set of stress limits. This approach significantly reduces the time needed for learning and searching. On a SPARC station 20, experiments show that it can search an optimal or near-optimal schedule within tens of seconds of CPU time, a time range which should be acceptable in power plant operations.

Genetic algorithms are powerful and robust heuristic adaptation procedures suggested by biological evolution and molecular genetics. Fuzzy set theory and fuzzy logic have been proposed in order to provide some means for representing and manipulating imprecision and vagueness. In this paper genetic algorithms and fuzzy logic are combined in a uniform framework suitable for fuzzy classification. We discuss how a fuzzy classification methodology introduced in previous papers has been improved by becoming part of a genetic algorithm. The resulting genetic fuzzy classification technique shows increased sensitivity of solution, avoids the effect of fuzzy numbers grouping and allows for more effective search over solution space.

This paper proposes a new input method for human operators of an interactive GA to reduce their psychological burden. This method utilizes discrete fitness values to reduce the psychological stress involved in the input procedure. The advantage and disadvantage of the proposed method are evaluated and discussed. We conducted simulations to investigate the influence of quantization noise stemming from the use of discrete fitness values on convergence. Results show that the quantization noise does not significantly worsen convergence in practical use. The proposed method and a variation of the method which combines both discrete and continuous fitness values are evaluated using two subjective tests involving the task of drawing faces. The results of the subjective tests indicate that our proposed method and its variation can significantly reduce the psychological stress level of human interactive GA operators. The experimental results of the proposed method are discussed on its universality.

We illustrate a method of controlled mathematical comparison for determining design principles of biological control circuits. The method involves the comparison of alternative circuits that differ in a single process, that are constrained so as to eliminate extraneous differences, and that are compared quantitatively on the basis of a priori criteria for functional effectiveness. We use this method to compare two forms of coupling that are important in the regulation of repressible gene expression. The first form, which is called complete uncoupling, is marked by a constant level of regulator protein. The second form, which is called perfect coupling, is marked by a level of regulator protein that varies coordinately with the level of repressible enzyme under control. Our results indicate that, for a large class of systems, performance is better with perfect coupling than with complete uncoupling if the regulator protein is a repressor, and conversely, that performance is better with complete uncoupling than with perfect coupling if the regulator protein is an activator. These results lead to testable predictions that are consistent with available data.

The soft computing methods are used for the control of refrigerator temperature. An axis having blades is introduced in the refrigerator compartment to control the direction of cold air, and the directions of the blades are controlled by a neural networks controller so that more cold air is sent where the temperature is high in the refrigerator compartment. The neural networks controller is taught by a TSK fuzzy supervisor. The TSK fuzzy supervisor has the form of TSK fuzzy model. For the identification of fuzzy sets parameters in TSK fuzzy models, a new method made by combining the genetic algorithm with the complex method is suggested. With the new method, more optimal parameters of fuzzy sets can be searched more speedily.