Kunihiko Nakazono

Vibration control of load for rotary crane system using neural network with GA-based training

A neuro-controller for vibration control of load in a rotary crane system is proposed involving the rotation about the vertical axis only. As in a nonholonomic system, the vibration control method using a static continuous state feedback cannot stabilize the load swing. It is necessary to design a time-varying feedback controller or a discontinuous feedback controller. We propose a simple three-layered neural network as a controller (NC) with genetic algorithm-based (GA-based) training in order to control load swing suppression for the rotary crane system. The NC is trained by a real-coded GA, which substantially simplifies the design of the controller. It appeared that a control scheme with performance comparable to conventional methods can be obtained by a relatively simple approach.

Load Swing Suppression in Jib Crane Systems Using a Genetic Algorithm-trained Neuro-controller

A neuro-controller for load swing suppression in a jib crane system involving only rotation about the vertical axis is proposed. The controller is trained by a genetic algorithm, substantially simplifying the design of the controller. As such a system is nonholonomic, the conventional control problem is difficult to solve, requiring knowledge of complex control theory. Using a simple three-layered neural network as a controller genetic algorithm-based training, it is demonstrated that a control scheme with performance comparable to conventional methods can be obtained by a relatively simple approach.

Force and position control of robot manipulator using neurocontroller with GA based training

In this paper, we propose a force and position controller for a robot manipulator using a neurocontroller (NC) with genetic algorithm (GA) based training. It is very difficult to design the controller which applies both force and position control to the robot manipulator. We use a simple three layered neural network as the controller, and the training method of the NC is GA based. Inputs to the NC are errors of the position and force. Furthermore, we input the integral information of the position error to the NC because it eliminates the steady-state position error. Simulation shows that the proposed NC has better performance for both position and force control than the conventional neural network, for the robot manipulator.

Backward control of multitrailer systems using neurocontrollers evolved by a genetic algorithm

We propose a design method for neurocontrollers (NCs) evolved by a genetic algorithm (GA) for the control of the backward movement of multitrailer truck systems. The difficulty of controlling backward movement increases with the number of connected trailers. In order to search for the best NCs for multitrailer systems, we propose a step-up training method. The step-up training sequence is as follows. First, the initial NCs, that are set to random values, are trained for an easy control object. Second, the set of NCs is trained for more difficult control objects. In this study, the initial NCs are first trained for a two-trailer connected truck system, then the NCs are trained for a three-trailer system, and finally the NCs are trained for a four-trailer system. The step-up training method is able to advance to NCs which can successfully control multitrailer systems. Simulation results show that the step-up training method is useful for multitrailer systems.

Simulation study of multitrailer control system using neurocontrollers evolved by a modified GA

In this paper, we propose a design method of neurocontrollers (NCs) evolved by a genetic algorithm (GA) for the backward movement control of multitrailer systems. In a previous study, we proposed a control method using NCs evolved by a modified GA to solve the backward movement control problem for a two-trailer-connected truck system. In the report, the modified GA which adaptively changes the number of offspring and the mutation rate according to the diversity of NC population has good control performance. However, the difficult and complexity of the backward movement control depends on the number of connected trailers, then the method of NCs with GA evolution is often not able to produce a better controller, or the GA process takes too much time to obtain NCs that can control the trailer-truck system successfully. For the multitrailer (three of more connected trailer) system, the control performance of NCs by the modified GA method is not clear. In this paper, we apply the modified GA to the multitrailer control system. The simulation results show that the modified GA still improves the search performance of NCs for the three- and four-trailer-connected systems.

Solution searching for multivariable optimization problems by a momentum genetic algorithm

Genetic algorithms (GAs) have emerged as powerful solution searching mechanisms, especially for nonlinear and multivariable optimization problems. Generally, it is time-consuming for GAs to find the solutions, and sometimes they cannot find the global optima. In order to improve their search performance, we propose a fast GA algorithm called momentum GA, which employs momentum offspring (MOS) and constant range mutation (CRM). MOS, which generates offspring based on the best individuals of current and past generations, is considered to have the effect of fast searching for the optimum solutions. CRM is considered to have the ability to avoid the production of ineffective individuals and maintain the diversity of the population. In order to verify the performance of our proposed method, a comparison between momentum GA and the conventional mean will be implemented by utilizing optimization problems of two multivariable functions and neural network training problems with different activation functions. Simulations show that the proposed method has good performance regardless of the small values of the population size and generation number in the GA.

Investigation of multi-layer neural network performance evolved by genetic algorithms

This paper presents a study on the investigation of multi-layer neural networks (MLNNs) performance evolved with genetic algorithm (GA) for multi-logic training patterns applied to various network functions. Specifically, we have concentrated on the Sigmoid, Step and ReLU functions to evaluate and simulate their performances in the network. We have revealed that GA training gives good training results in evolutionary computation by changing of Sigmoid, ReLU and Step as the activity functions in MLNN performance. Sigmoid function has proved to train all patterns for all outputs without any challenge as compared to ReLU function and Step in this study. We are still trying to see how a ReLU function could be trained with GA for MLNNs performance for the two input and four output training patterns termed as the multi-logic pattern training about multiple training parameters.

Identification of time-series signals using a dynamic neural network with GA-based training

We propose a dynamic neural network (DNN) that realizes a dynamic property and has a network structure with the properties of inertia, viscosity, and stiffness without time-delayed input elements, and a training algorithm based on a genetic algorithm (GA). In a previous study, we proposed a modified training algorithm for the DNN based on the error back-propagation method. However, in the previous method it was necessary to determine the values of the DNN property parameters by trial and error. In the newly proposed DNN, the GA is designed to train not only the connecting weights but also the property parameters of the DNN. Simulation results show that the DNN trained by the GA obtains good performance for time-series patterns generated from an unknown system, and provides a higher performance than the conventional neural network.

Neurocontrol of Bicycle using GA with Rough Evaluations.

Genetic algorithms (GAs) with rough evaluations can prompt the evolution of neural networks that are able to control unstable dynamic systems. The proposed control method exploits the advantage of GAs that time-varying evaluations can be easily incorporated. First an easy evaluation in GAs induces the appearance of neural networks with controllability. Second, an evaluation of settling time prompts the evolution of neural networks that show high performance. The method is applied to the stable control of a bicycle. Neurocontrol of the steering at direction change causes reverse response like that of a human cyclist.