Genetic Algorithms (GAs) are search procedures for combinatorial optimization problems. Because GAs are based on multi-point search and use crossover operator, GAs have an excellent global search ability. However GAs are not effective for searching the solution space locally due to crossover-based-search, and the diversity of the population sometimes decreases rapidly. In order to overcome these drawbacks, we propose a new algorithm called Immunity-based GA (IGA) combining features of the Immune System with GAs. IGA is expected to improve the local search ability of GAs and maintain the diversity of the population. We apply IGA to the VLSI floorplan design problem. Experimental results show that IGA performs better than GAs.

抄録全体を表示

Start-up scheduling is aimed at minimizing the start-up time while limiting turbine rotor stresses to an acceptable level. This scheduling problem has a wide search space. In order to improve the search efficiency and robustness and to establish an adaptive search model, we propose to integrate evolutionary computation, based on Genetic Algorithms (GA), with reinforcement learning. The strategies with our proposal include: multi-boundary-based enforcement operator and multi-elitist plan. By setting a second boundary, located right outside the existing boundary containing those feasible schedules, we extend our proposed enforcement operator and the conventional elitist plan into the multi-boundary-based enforcement operator and multi-elitist plan. These two strategies work together to focus the search along the boundary, around which the optimal schedule is supposed to exist, so as to increase the search efficiency as well as its robustness. During a search process, GA guides the reinforcement learning to concentrate its learning on those promising areas instead of the entire space. In return, reinforcement learning can generate a good schedule, in the earlier stage of the search process. We obtain encouraging test results. In this paper, we propose the GA-based search model with these strategies and discuss the test results.

抄録全体を表示

The hypercycle model is a promising framework for emergent computations because of its powerful self-organizing behavior and flexible reacting schemes. In particular, emphasizing system code evolution based on hypercycles, it is important to develop and maintain various kinds of hypercycles. However, the theoretical studies have elucidated their powerful self-organizing ability don't allow more than two kinds of hypercycles to coexist, and so some missing links have remained unsolved still now concerning a problem how hypercycles conserve themselves from the others. This paper demonstrates a process whereby macro conservative hypercycles emerge through self-organizing compartmentation in a chemical evolution model named Atomoid The Atomoid is a simulation model, the reaction dynamics of which are analogous to those found in the physical world based on energy levels. In the Atomoid, self-reproducing hypercycles emerge under dissipative structures, and a potential field is provided for the purpose of compartmentation of them. Local interactions between genotypic atomic reactions and phenotypic potential field result in emergence of macro conservative systems. This paper discusses these dynamics through simulation results.

抄録全体を表示

The combinatorial optimization problem is important problem for practical use in any factory. However the problem includes very difficult problem to solve automatically. One of the famous difficult problem of combinatorial problem is the Traveling Salesman Problem (TSP). This paper approaches to TSP using new information processing method that is an adaptive distributed system on the vibrating potential field. The autonomous units proposed in this paper uses field information and does analogous search in the problem Euclidian environment like as artificial ameba. In this paper, we propose its framework and analyze it by computational experiments.

抄録全体を表示

This paper aims to evaluate the fault tolerance of hierarchical decentralized systems (HDSs) by autonomous controllability. A HDS consists of a number of identical modules, each of which consists again of a number of identical submodules. Autonomous controllability at level α means that the entire system remains controllable, in spite of failures in arbitrary α submodules. Conversely, if the entire system becomes uncontrollable because of failures in a set of β submodules, this number β of failed submodules manifests the level of fault tolerance. We will thus find such sets of submodules to evaluate fault tolerance. A HDS considered here is built by combination of three basic structures, such as ring-type, chain-type and wheel-type structures. Therefore, the structural information of a HDS is described with the type of interconnections among modules and that for submodules. We will show that the results of evaluation for a HDS are obtained by combination of the results based on the interconnections among modules and the results on those among submodules. Various examples are also shown and it is discussed the fault tolerance of different types of structures.

抄録全体を表示

In order to solve job shop scheduling problems (JSSPs) by a genetic algorithm (GA), one should first design an encoding scheme, on which a search space is constructed. This paper proposes two encoding formats; one is a string code format that leads to the redundancy of the code space, and the other is a matrix code format that overcomes the redundancy but only insures an approximate representation. Two corresponding genetic algorithms (GAs) are designed for investigating the encoding effectiveness. Complex problems like the JSSPs usually require complicated code structures, which in turn call for delicate design of genetic operations such as crossover. The code structures of the two encoding formats are analyzed and compared. Test-runs of the two GAs on several tough JSSP benchmarks are performed for a demonstration of the validation of the proposed methods.

抄録全体を表示

In recent years, in the fields of artificial intelligence (Al), much attention has been focused on the reactive planning approach such as behavior-based AI, new AI, and so on. However, in these approaches, there is some criticism that their arbitration among competence modules are fixed against dynamically changing environments. On the other hand, biological information processing systems have various interesting characteristics viewed from an engineering standpoint. Among them, the immune system plays an important role in maintaining its own system against dynamically changing environments. Based on this fact, we have been investigating a new decentralized consensus-making system for the behavior arbitration of autonomous mobile robots inspired from the idiotypic network hypothesis in immunology. In this paper, we report the useful_??_ss of our proposed method by carrying out some computer simulations and real experiments.

抄録全体を表示

In a multihop radio network, packets are transmitted from source nodes to destination nodes by activating several links between nodes. Each node can either send a packet to, or receive a packet from, at most one of its adjacent nodes simultaneously. In order to minimize the transmission time for given requests, the problems must be solved by selecting a transmission route for each request (the routing problem) and by finding a link activation schedule (the link activation problem). The routing problem is decomposed into two sub-problems: the candidate extraction problem and the route selection problem. In this paper, we propose a neural network algorithm using the maximum neuron model for the route selection problem. We verify, through simulations, that our algorithm finds better solutions in shorter time than the existing algorithms. We also prove the NP-hardness of this problem.

抄録全体を表示

We propose a method to evolve adaptive behavior of an artificial neural network (ANN). The adaptive behavior emerges from the coordination of learning rules. Each learning rule is defined as a function of local information of a corresponding neuron only and modifies the connective strength between the neuron and its neighbor neurons. The learing rule is exposed to the selective pressure based on the fitness value, which represents the importance in producing the adaptive behavior of the ANN. The learning rules with lower fitness values are replaced by new ones generated by the Genetic Programming techniques. Experimental results demonstrate that the proposed method produces adaptive behavior of single-layered ANNs and two-layered ANNs. This means that efficient learning rule evolves. The learning rules in the two-layered ANN coordinate each other and the macroscopic adaptive behavior of the ANN emerges.

抄録全体を表示

This paper presents a new genetic algorithm for job-shop scheduling problems. When we design a genetic algorithm for difficult ordering problems such as job-shop scheduling problems, it is important to design encoding/crossover that is excellent in characteristic preservation. We regard a sub-sequence on each machine as a characteristic to be preserved between parents and their children. The proposed method uses a job sequence for encoding. This paper proposes a new crossover, the sub-sequence exchange crossover (SXX), that can preserve the characteristic very well. Since the children generated by SXX are not always feasible, we propose a technique to transform them into active schedules by using the GT method with a few modifications. Maintaining a diversity of population is important for preventing premature convergence. We present a mutation based on the shift change operator for efficiently introducing a diversity. Furthermore, we design a generation alternation model that is excellent in diversity maintaining. By applying the proposed method to Fisher's and Thompson's 10×10 and 20×5 problems, we show its effectiveness.

抄録全体を表示

In this paper. we propose a practical method for solving the delivery scheduling problem and discuss its implementation. The method is based on the distributed cooperative problem solving. In the delivery scheduling method, the covered area is partitioned into multiple sub-areas and each sub-area is assigned a sub-problem solving agent which takes care of the scheduling of drivers and trucks in the sub-area. Integrating those sub-problem solving agents, an integration-and-evaluation agent solves the total problem.
By employing the distributed cooperative problem solving framework for the delivery scheduling problem, we achieved an easy incorporation of various evaluation parameters in the process of scheduling, efficient use and management of scheduling knowledge of various levels, and reduction of computer processing by partition of the problem into sub-problems.

抄録全体を表示

The speed control ability of dc servomotors is affected by parameter variations and disturbance torque. In this paper, the robust speed control method for dc servomotors with the disturbance torque observer and a feedback controller is proposed. The disturbance torque observer is used for the compensation of parameter variations and disturbance. The feedback gain is adjusted in the aspect of stability by the adaptive gain law based on Lyapunov's direct method and is used to restrain the influence of estimation error for the disturbance observer.
Moreover, the robust current control scheme for a voltage source PWM inverter with disturbance observer is presented, since current control ability affects the control performance of dc servomotors. The validity of this control scheme is verified by numerical simulations and experiments.

抄録全体を表示

The principle of a novel gain grating method, which enables an on-line monitoring of the Bragg wavelengths of fiber gratings (FG's) to be formed in Er-doped fibers (EDF's), is experimentally confirmed. The 532nm second harmonic of a mode-locked Nd: YAG laser is used as a writing light source in this experiment. Firstly, it is observed that the 532nm light can induce a refractive index change in EDF's which is enough for FG formations. Secondly, the formation of gain gratings in EDF's is confirmed, and a wavelength controllability better than 0.2nm is expected.

抄録全体を表示

The aim of this work was to investigate the feasibility of forming self-organized artificial neural networks in a virtual environment for information processing machine. A computer model on the organization process of nerve cell systems was constructed with a simplified model of eukaryotic cell as its basic raw element. The processes of protein metabolisms were computed according to gene codes of chromosome in the cell. The result of computer simulation with this model showed the processes of ontogenesis of nerve cells and nerve fibers, and also the phylogeny process of self-organizing formation of neural network.

抄録全体を表示

We have developed a high peak power pulsed vacuum ultraviolet (VUV) flash lamp with extremely broadened emission spectra by using a pulsed silent discharge of a binary mixture of rare gases. In appropriate rare gas mixtures, certain energy transfer processes between rare gases were possible, resulting in the simultaneous multi-excimer emissions in a single discharge tube. In the case of a Kr/Xe mixture, an emission spectrum of almost flat-top shape was obtained with wavelengths ranging between 150nm and 175nm.
The frequency-averaged peak power of the broadened VUV emission was 0.4kW. Pulse widths of the VUV emission ranged from 1.5μs to 2.5μs (FWHM), depending on emission wavelengths. Kinetic analyses supported the production of hetero-nuclear excimers (KrXe^*) with a center emission wavelength of 158nm as well as the production of homo-nuclear excimers (Kr₂^* and Xe₂^*) in a pulsed Kr/Xe silent discharge, that are the origins of the extremely broadened VUV spectra with a bandwidth of 26nm (FWHM).

抄録全体を表示

The human emotional model called “Kansei” is vague. That is why it is difficult to understand. Kansei Engineering is a method which shows relationships between Kansei and physical design elements, and it is a poweful method for understanding Kansei. Quantification Method I is one of the techniques in Kansei Engineering, a very famous regression method. But the problem of this technique is that a treatment of non-categorical design elements such as color is difficult.
In this paper, we try to solve the above problem by fuzzy theory, and apply this technique to Kansei Engineering for handkerchief design. Finally, we construct a handkerchief design model and show the validity of the model.

抄録全体を表示

We often encounter with a signal composed of a base level signal slowly varying in time and a relatively fast varying signal of interest, such as electrocardiogram and video signal of cars running on a road. In order to set the base level signal, we usually estimate the base level signal by a lowpass filter using difference of characteristics in frequency domain between these two signals. However, a base level estimator of the linear filter is directly affected by amplitude of the signal of interest.
We propose the new base level signal estimator using a bilateral clamper that consists of two linear filters and a non-linear element, moreover, analyze precisely and give roughly design criteria. This estimator has nice response to the change of the base level signal, and well protects the affection of the signal of interest.
Finally, we demonstrate usefulness of this estimator showing some examples.

抄録全体を表示

We have developed a scheduling algorithm, which simultaneously levels resourcece profiles in both the total time span and local time intervals, for plant construction schedule optimization. Resource leveling in local time intervals minimizes resource peaks through a heuristic search method with a target profile.
The developed scheduling algorithm has been experimentally evaluated with realistic construction data of a power plant. The resulting profile from overall resource leveling have been improved with the local resource leveling function. From the results it is confirmed that the local leveling is helpful in resource peak reduction.

抄録全体を表示

A Compact and simple distance meter based on self-coupling effect of semiconductor laser has been developed. Using the interference in the laser resonator with the returned light from the external surface, the sensor of the distance meter is composed of only a laser diode and a lens. As it has no external interference optics, it could be much smaller than the interference measurement. The drive current of the semiconductor laser is modulated with not only a triangular wave current but also a sinusoidal wave current. It is confirmed that the device modulated with sinusoidal wave can measure the distance with the same accuracy as one with triangular wave, and that the former is superior to the latter in the temperature characteristic.
The measurement error is found to be below one percent between 20cm and 90cm.

抄録全体を表示

In the centralized architecture of the Fieldbus protocol, the synchronous phase executions are being guaranteed timeliness and throughput, by a centralized scheduling scheme on a timely basis. Whereas, the asynchronous phase executions are being served using an emulation of a “circulated” timed token. In such a scheme, the maximum token circulation (i.e., cycle) time is determined by the maximum token holding time parameter within which a station can serve its asynchronous traffic. Furthermore, the maximum token circulation time determines the maximum waiting time an urgent data can incur.
In order to ensure that the maximum delivery bounds of urgent data are not violated, and that the maximum token circulation time is kept within bounds, an asynchronous bandwidth allocation mechanism is proposed. In this approach, first the allocatable asynchronous bandwidth is derived. Then, A modified version of the Normalized Proportional Allocation Scheme (NPAS) is proposed, to determine the bandwidth share of each station (i.e. the maximum token holding time). In case the NPAS is not applicable, due to protocol constraints, a procedure allowing its partial application is proposed.
Analytical and simulation analysis are used to evaluate the proposed approach. The results show that the asynchronous bandwidth allocation scheme and the parameter setting policy can ensure, in one hand, that the delivery delay bounds of urgent messages are not violated, and that the maximum token cycle time is kept within bounds. On the other hand, the results show that parameter tuning can further lead to improve the protocol performance.

抄録全体を表示

A new model of Generalized Fuzzy Inference Neural Network (GFINN) is proposed in this paper. The network consists of three layers: the input-output layer, the if layer, and the then layer. In each layer, there are the operational nodes. The GFINN can perform three representative fuzzy inference methods by changing the connectivity and the operational nodes. There are three learning precesses in the GFINN: the self-organizing process, the rule-integration precess, and the LMS learning process. Especially in the rule-integration process, the GFINN employs two feature maps in order to integrate appropriate rules effectively. Computer simulations are carried out to show the superiority of the GFINN over the back-propagation networks.

抄録全体を表示

The control of the Distributed Parameter System(DPS) has some problems like as that 1)the construction of the mathematical model of the DPS is difficult and 2) the allocation of control equipments must be optimized. For the former problem, we applied the fuzzy technique to the control of the DPS because it has dominance of that one can construct the control system with the experience rules but without the mathematical model. For the latter problem, the genetic algorithm(GA) was used because one can optimize the controller allocation with low calculation time by using it. Simulation experiments were performed for the temperature control problem on a finite length stick. As the result, effectiveness of the proposed method was confirmed.

抄録全体を表示

Proposed in this paper is a novel approach to inverse optimization problems by the learning of neural networks. Inverse optimization here means to estimate a diagonal positive semidefinite quadratic criterion function which optimizes a given solution under predetermined constraint conditions. This task is reduced to obtaining gradient vectors which lie in the intersection of two polar cones: one determined by the Kuhn-Tucker condition and the other by the positive semidefinite condition. A new network architecture for inverse optimization problems is proposed to simultaneously satisfy the two conditions. It is a special purpose network architecture with inequality constraints on some of the connection weights. Applications of the proposed method to examples with 3 variables demonstrate that it can actually obtain a criterion function satisfying these conditions. Detailed analyses are given to the results with different initial gradient vectors. Furthermore, real data on second-hand houses are analyzed by the proposed method. The present method, however, is restricted to the estimation of only diagonal positive semidefinite quadratic criterion functions. Its extension to estimating positive semidefinite quadratic criterion functions is left for further study.

抄録全体を表示