Shunting scheduling problems in railway stations can be regarded as a sort of resource constrained project scheduling problems (RCPSPs). But unlike normal RCPSPs, shunting scheduling problems have such requirements that the number of works which consist of a project has to be dynamically changed in the process of solving the problem, some of the works have to be performed at prescribed timing, etc., We propose an efficient algorithm for shunting scheduling problems combining probabilistic local search and PERT. Local search and PERT is combined so that the candidates for answers in the local search process are evaluated by PERT. This enables to reduce the search space of the local search to a great extent and to make the algorithm work quite fast. We have confirmed effectiveness of our algorithm through several experiments using practical train schedule data.
In this paper, we propose an identification method of discrete nonlinear systems using Chebyshev polynomials. A nonlinear function of a given system is assumed to be represented by a linear combination of some Chebyshev polynomials. Each coefficient of the Chebyshev polynomials is easily evaluated by the least squares method. An error bound of the identification is estimated under an assumption of measurement errors being negligible small. Numerical examples show that the accuracy of this identification for nonlinear systems is considerably well improved as the order of the Chebyshev polynomials is increased.
In this paper, we describe learning rules by means of the time difference simultaneous perturbation method. This recursive method was proposed by the first author to find a maximum or minimum of unknown functions by using only one value of the objective function at each iteration. We applied this to learning of neural networks. This approach is simpler than the well-known backpropagation method in the point that our rule needs only one value of the error function instead of the complicated derivation of derivatives in the backpropagation method. Some simple numerical examples are shown.
This paper proposes a fast algorithm to compute the fuzzy reasoning with functional consequent part. This type of fuzzy reasoning is able to represent a complex input/output relation. The relation is difficult to be represented by the simplified fuzzy reasoning (SFR). The fuzzy reasoning with functional consequent part can be easily optimized in the same way as the SFR can. This will be advantageous in such a case that the fuzzy reasoning is applied to an adaptive controller and is optimized. The execution time can be ignored when the fuzzy reasoning is applied to a non-adaptive controller, but will be a serious problem when the fuzzy reasoning is applied to such an adaptive controller or its optimization. As a method to improve this difficulty, we propose a fast computational algorithm for the fuzzy reasoning with functional consequent part. This algorithm can be expressed by a simple computational flow. Possibility of its parallel execution is also discussed in this paper.
In this paper, are discussed a determination method of weighting matrix of quadratic performance index and a design method of optimal deadbeat control system. First, the relationship between weighting matrix of quadratic performance index and extra pole of over-parameterized pulse transfer function model is indicated. Next, a design method of optimal deadbeat control system is explained. Lastly, to demonstrate the effectiveness of the proposed method, some numerical examples are also presented.
Frequency domain subspace identification algorithms have been studied recently by several researchers in the literature, motivated by the significant development of the more popular time domain counterparts. Usually, this class of methods are focused on discrete-time models, since in the case of continuous-time models, the data matrices often become ill-conditioned if we simply rewrite the Laplace operator s as s=jω, where ω denotes the frequency. This paper proposes an efficient and convenient approach to frequency domain subspace identification for continuous-time systems. The operator ω=(s-α)/(s+α) is introduced to avoid the ill-conditioned problem. Hence the system can be identified based on a state-space model in the ω-operator. And then the estimated ω-operator state-space model can be transformed back to the common continuous-time state-space model. An instrumental variable matrix in the frequency domain is also proposed to obtain consistent estimates of the equivalent system matrices in the presence of measurement noise. Simulation results are included to verify the efficiency of the proposed algorithms.
A methodology for identifying nonlinear dynamic systems using NARX model structure and three layered feed forward neural networks is presented. The neural network, which maps between regressor vector of the NARX model and the output, is viewed as comprising of number of linear models, each of which is represented by a hidden neuron. An algorithm that uses orthogonal least squares based regressor selection procedure for the estimation of the structure and weights of the hidden neurons is developed for training partially connected neural networks. The use of regressor selection algorithm facilitates implicit identification of unknown dead times and dynamic orders by the neural network during the training process. The conventional error back propagation is used for online adaptation of the identified neural network model. The validity of the proposed approach is demonstrated by applying it to obtain the model of a wastewater pH neutralization process.
According as information systems become more important for business activities, they should be evaluated adequately their effectiveness and suitability. However, it is difficult to evaluate the latest information systems that provide the work support functions, such as groupware, internet and so on, as to compare with the legacy information systems that provide the automation functions. In this paper we propose a method to evaluate the suitability of work support functions for business information systems. The method provides the ability to evaluate quantitatively the functional suitability between the business process and the business resources, such as information systems and workers, using the functional process model and fuzzy relation. The method consists of three schemes: (l) derivation scheme of the functional process model, (2) evaluation scheme of the functional suitability, and (3) analysis scheme of the functional suitability on the type of information processing, the level of information processing, and the type of human-machine interaction. Furthermore, we will show the validity and the usefulness of the method in a case study.
This paper presents a method of adaptive control for nonlinear systems using neural networks. The control input is given by the sum of the output of an adaptive controller and the output of the neural network. The neural network is used to compensate the nonlinerity of plant dynamics that is not taken into consideration in the usual adaptive control. The role of the neural network is to construct a linearized model by minimizing the output error caused by nonlinearities in the control systems. New parallel neural networks are proposed. The learning time required for convergence and the network size of each parallel neural network can be reduced when compared to conventional neural network based control systems.
We propose a new on-line recognition method to recognize handwritten cursive-style Japanese characters correctly. Our method simultaneously uses both directional features, otherwise known as off-line features, and direction-change features which we designed as on-line features. The direction-change features express where in the mesh and in which direction the character's coordinates change. These features express both written strokes in the pen-down state and unwritten imaginary strokes in the pen-up state. The recognition rate was improved by our method over the traditional method using only directional features.
To reduce the product development period of embedded software, one effective approach is review and validation of product specifications at an early stage of the design. One method of doing this is to describe Statechart (Harel's state transition diagram) of the specifications. However, design methods of Statechart have not yet been established. The article proposes a method of making states hierarchical in Statechart. To make states hierarchical, we created Deciding State Hierarchy Matrix that expresses relation among states, and transformed a matrix into three types of Complete Form Matrix. Besides we propose an algorithm to transform into Complete Form Matrix. We have applied the proposed method to a development of control software for a prototype of airconditioning equipment. The preparation of the Statecharts for that equipment demonstrated the effectiveness of this approach.
In this paper, a decoupling control for a non-minimum phase system is considered from a limiting form of linear optimal regulator problems point of view, and an explicit relationship between the limiting properties and a decoupling control is shown. Moreover, using an appropriate performance index, this paper deals with the new method on the decoupling control to keep out the effect of the natural poles by pole-zero cancellation. The properties of stability and sensitivity are improved by this method. This paper also gives numerical examples to illustrate effectiveness on decoupling control with a limiting form.
In order to improve the response property of the closed-loop control system constructed by generalized minimum variance control (GMVC), in general, the control-weighting polynomial included in the cost function is adjusted. But, the tracking response to the reference-signal and the response property for disturbances are changed together when the control-weighting polynomial is tuned. A method solving this problem is a two-degree-of-freedom GMVC. In this paper, we propose a two-degree-of-freedom GMVC with the reference-signal-filter. On the other hand, we propose the control system allowing the reference-signal-weighting to be a rational function in the conventional control structure. This system can set up the feedback property and the tracking performance independently.
Genetic Algorithm (GA) has been applied to many difficult combinatorial optimization problems. It is known that GA can find the global solution rapidly if the population holds both varieties and concentration sufficiently. However, it is difficult to satisfy both requirements at the same time, because they often tradeoff each other. Usually, existing methods have several parameters to controle this tradeoff balance. But, it is difficult to know proper values for good performance before search. In this paper, we propose GSA, Genetic algorithm with Search area Adaptation. GSA controls dynamically this tradeoff balance and needs not tune parameters. We applied it for the floorplan design problem. The experimental results show better performance than the existing methods, such as GAPE, SA and ISA.
In this paper, we present an analysis of the problems involved in both centralized and distributed network management architectures and present a scheme of Hierarchical Network Management based on Extended SNMP as our solution. The status field in SNMP PDU Format is newly defined so that various management policies can be encapsulated into the SNMP PDU and sent from main manager to sub manager. The management task, which achieves concrete management activities on side of the sub manage, is created and executed according to management policy. One of the advantages of the proposed method is distribution of all monitoring tasks and data processing tasks to each sub manager so that the load of main manager and the traffic of management information on the backbone network can be significantly decreased. In addition, it is possible to obtain high-level management result from the sub manager directly by using free-designed management policy. The management framework is easy to implement on all versions of SNMP and makes the management system more flexible and scalable. Implementation examples are also given.
As mechatronics consists of various engineering fields such as mechanics, electronics, information etc., it is one of the most important subjects for the engineering students to study. So, it is becoming a well-received curriculum in Japan, Europe and USA. Especially, for the manufacturing, mechatronics is one of the most suitable subjects. The “Department of Systems and Control Engineering” in Osaka Prefectural College of Technology has opened a course named “Case Study on Design of Mechatronics Systems” for these five years. In this curriculum, each student makes a robot in group. Through the robot-making in group, they learn not only “technical engineering” but also “solving method for problems”. Recently, it is said that the balance between “learning” and “humanity” in engineering education is very important. In other word, it is important for the students to understand the methodology for system configuration in the fields of not only hardware and software but also human-ware. This paper describes how the “Case Study” was performed and how the “humanity” was obtained in the group study.
The light reflection characteristic on the surface of chip-electronic-parts is deeply dependent on the wavelength of the light source. In this letter, we discuss the design problem of an optical filter which is placed in front of the photo sensor of the CCD camera By this method, the good quality of the binary image of chip-electronic-parts can be obtained.
In this paper, the numerical simulations are performed in order to compare ability of HONNs with one of the traditional model in the case where the same number of parameters is used, and effectiveness of HONNs is shown.
This paper presents a new method for prediction of the remaining battery discharge capacity by using the backpropagation algorithm for neural networks (NN). The input-layer of NN has two neurons corresponding to the battery terminal voltage and discharge current. From the input data, the NN predict the remaining battery discharge capacity according to the study results. This method can be adapted to the every battery. The network can be easily implemented in hardware using standard circuit element that is translated from the computed weight of the simulation networks.