IEEJ Transactions on Electronics, Information and Systems
Online ISSN : 1348-8155
Print ISSN : 0385-4221
ISSN-L : 0385-4221
Volume 113 , Issue 6
Showing 1-16 articles out of 16 articles from the selected issue
  • Masumi Ishikawa
    1993 Volume 113 Issue 6 Pages 371
    Published: June 20, 1993
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Download PDF (81K)
  • Yoshiro Miyata
    1993 Volume 113 Issue 6 Pages 372-377
    Published: June 20, 1993
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Download PDF (935K)
  • Yoji Uno
    1993 Volume 113 Issue 6 Pages 378-383
    Published: June 20, 1993
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Download PDF (1065K)
  • Yasuo Sugai, Ken Kobayashi
    1993 Volume 113 Issue 6 Pages 384-393
    Published: June 20, 1993
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Although there are some neural network models which show adaptive self-organizing characteristics, they have complicated architectures and complicated actions, and there comes to be many parameters which the values are very difficult to be adjusted.
    In this paper, we propose a new neural network model for adaptive learning by unsupervised learning. The model consists of simple architacture and so has simple action compared with others. Learning of analog input is also possible by only one representation of input vectors. Moreover, by changing the parameter value, we can make the model forget old patterns. Simulation results indicate the effectiveness of the proposed model.
    Download PDF (1000K)
  • Hiroshi Kinjo, Sigeru Omatu, Tetsuhiko Yamamoto
    1993 Volume 113 Issue 6 Pages 394-401
    Published: June 20, 1993
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    In this paper, we propose a method to construct a suboptimal controller for a nonlinear system using neural networks. First we roughly approximate a nonlinear plant by a linear model. To derive the optimal control law for the approximated linear system, we introduce teh Luenberger observer and optimal feedback control law which minimizes a quadratic cost function. We consider two kinds of compensators based on neural networks. One is to compensate the observer and the other is to adjust the linear feedback controller. The neural networks are basically used to add the nonlinearlity of the plant to the approximated linear model.
    After training the network, nonlinear observer could work well to estimate the state of the nonlinear plant. To represent the dynamical structure of the plant, we introduce a recurrent neural networl for the compensator of the observer. Finally, simulation results show the effectiveness of the proposed method for nonlinear optimal control problems.
    Download PDF (1028K)
  • Yutaka Maeda, Yakichi Kanata
    1993 Volume 113 Issue 6 Pages 402-408
    Published: June 20, 1993
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    This paper proposes two learning rules for the recurrent neural networks using an ordinary simple perturbation and a simultaneous perturbation.
    The first learning rule uses the simple perturbation signals. Then, we can obtain values of an evaluation function with and without perturbation. By using a difference approximation, the rule estimates the partial derivatives of the evaluation function. The estimators are used to update all weights. On the other hand, the second one uses the simultaneous perturbation where perturbations are added to all weights simultaneously. By using a difference approximation, we obtain the updating quantities for all weights.
    When a neural network is used as a direct controller for an unknown plant, a desired signal corresponding to an output of the plant is usually given. Generally, the evaluation function is squared error between the desired output and the practical output of the plant. Therefore, we need a sensitivity function of the plant in order to obtain the updating quantities for all weights. However, the proposed rules are applicable to this problem without this information about the plant. Some numerical results are shown.
    Download PDF (1164K)
  • Katsuhisa Endo, Yoshihisa Ishida, Takashi Honda
    1993 Volume 113 Issue 6 Pages 409-416
    Published: June 20, 1993
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    This paper describes a gain adjustment method of I-PD control systems using a neural network.
    PID or I-PD controller has been used for several control systems. It is important to decide the gains of PID or I-PD controller, because the control performance of the system depends on such gains. Recently, many studies have been performed to apply neural networks to identification and control of dynamical systems.
    In this paper, we propose a new method to automatically adjust the gains of the I-PD control system using a neural network. The neural network used here consists of three layers. Coupling coefficients of the network are regarded as gains of the I-PD controller. The coupling coefficients are adjusted to minimize the error between the controlled value and the target value, that is, the output value of a model.
    Firstly, we try simulation experiment and show that the controlled value nearly equals the target value. Next, we experiment on a DC servomotor system and confirm that the proposed method is effective for a real control system.
    Download PDF (858K)
  • Takeshi Fukao, Jijun Wu
    1993 Volume 113 Issue 6 Pages 417-423
    Published: June 20, 1993
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Discrete optimization problems are generally difficult to solve since they are NP-complete. They may have complex structure in solution space with huge number of local minima and the simulated annealing is well known as an effective stochastic algorithm for these problems. The determination of the cooling schedule is essential in annealing which is under the control of the critical temperature.
    This paper discusses on the critical temperature in the neural net-like algorithm of the discrete quadratic optimization problems, derived through mean field approximation, and the validity of our evaluation method of the critical temperature is shown by many examples. It is also pointed out that the critical temperature can be made to be higher by modifying the structure of the objective function without changing the original discrete optimization problem. This derives the possibility of finding more efficient algorithms.
    Download PDF (774K)
  • Yasuhiro Kojima, Yoshio Izui, Sumie Kyomoto, Tadahiro Goda
    1993 Volume 113 Issue 6 Pages 424-429
    Published: June 20, 1993
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Recently, many controllers using neural networks are proposed. The hybrid and direct type controllers are two major categories. While the former one tunes up parameters of the conventional controller using neural networks, the latter constructs the controller by learning inverse dynamics of control target.
    Meanwhile, the electric power system requires voltage and reactive power control (VQ control) to avoid voltage collapse. The conventional VQ control, however, meets this requirement unsatisfactory because of approximated control.
    In this paper, we will propose a new algorithm for VQ control using recurrent neural networks which have the ability to treat system dynamics. Firstly, we will propose the learning algorithm for inverse dynamics of controlled target using recurrent neural networks. Secondly, we will apply this algorithm to the VQ control. We will call this controller ‘neuro VQC’. Finally, usefulness of the neuro VQC will be shown in comparison with conventional VQ controller.
    Download PDF (1004K)
  • Shigenori Matsumura, Sigeru Omatu, Hiromasa Higasa
    1993 Volume 113 Issue 6 Pages 430-439
    Published: June 20, 1993
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    In order to devolop an efficient driving system of electric vehicles (EVs), a testing system has been devised, which enables us to simulate the driving performance of EVs. In this system, running resistance of the EV is realized by torque control of the motor operated as a generator. Nowadays, torque control is performed by using conventional PID controllers, and reasonably good results have been obtained. However, when using conventional PID controllers, it generally needs much effort and time to tune the PID gains.
    In this paper, instead of conventional PID controllers, three types of control systems by neural networks are considered. At first, the testing system and its conventional PID controller are described. Then self tuning PID type neurocontroller is discussed from the viewpoints of how to self-tune the PID gains automatically and parallel and series types are examined to improve the control performance. Finally, we show numerical results of the effectiveness of the presented methods through simulations and experiments, compared with the conventional PID control.
    Download PDF (1161K)
  • Masao Kawachi
    1993 Volume 113 Issue 6 Pages 440-445
    Published: June 20, 1993
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Download PDF (2372K)
  • Norio Fukuma, Takehiro Mori
    1993 Volume 113 Issue 6 Pages 446-451
    Published: June 20, 1993
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Download PDF (855K)
  • Kazuhisa Matsuo, Toshimi Katayama, Takehiko Tomikawa
    1993 Volume 113 Issue 6 Pages 452-453
    Published: June 20, 1993
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Download PDF (756K)
  • Norio Nishizuka, Minoru Tahara, Kimio Sato
    1993 Volume 113 Issue 6 Pages 454-455
    Published: June 20, 1993
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Download PDF (603K)
  • Shigeki Yamagata, Osamu Yoshie
    1993 Volume 113 Issue 6 Pages 456-457
    Published: June 20, 1993
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Download PDF (568K)
  • 1993 Volume 113 Issue 6 Pages e1
    Published: 1993
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Download PDF (63K)
feedback
Top