Learning soft computing control strategies in a modular neural network architecture

Chapter 3 - Neural Networks: A Methodology for Modeling and Control Design of Dynamical Systems

In this paper, the use of radial basis function network (RBFN) for simultaneous online identification and indirect adaptive control of nonlinear dynamical systems is demonstrated. The motivation of using RBFN comes from the simplicity of its structure and simpler mathematical formulation, which gives it an advantage over multi-layer feed-forward neural network (MLFFNN). Since most processes are nonlinear, the use of conventional proportional-integral-derivative controller is not useful. Most of the time plant’s dynamics information is not available. This creates another limitation on the use of conventional control techniques, which works only if plant’s dynamics information is available. The proposed controller is tested for parameter variations and disturbance effects. Simulation results showed that RBFN is able to capture the unknown dynamics as well as simultaneously able to adaptively control the plant. It is also found to compensate the effects of parameter variations and disturbances. The comparative analysis is also done with MLFFNN in each simulation example, and it is found that performance of RBFN is better than that of MLFFNN.

We describe a new approach for face recognition using modular neural networks with a fuzzy logic method for response integration. We proposed a new architecture for modular neural networks for achieving pattern recognition in the particular case of human faces. Also, the method for achieving response integration is based on the fuzzy Sugeno integral. Response integration is required to combine the outputs of all the modules in the modular network. We have applied the new approach for face recognition to a real database of faces of students at our institution. Recognition rates with the modular approach were compared against the monolithic single neural network approach to measure the improvement. The results of the new modular neural network approach were excellent overall and also in comparison to the monolithic approach. © 2005 Wiley Periodicals, Inc. Int J Int Syst 20: 275–291, 2005.

A new dynamical sliding mode control algorithm is proposed for robust adaptive learning in analog multilayer feedforward networks with a scalar output. These type neural structures are widely used for modeling, identification and control of nonlinear dynamical systems. The zero level set of the learning error variable is considered as a sliding surface in the space of network learning parameters. The convergence of the algorithm is established and conditions are given. Its effectiveness is shown when applied to on-line learning of nonmonotonic function using a two-layered feedforward neural network.KeywordsHide LayerFeedforward Neural NetworkSliding ModeScalar OutputLinear Activation FunctionThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Preface to the special issue on soft computing for modeling, simulation, and control of nonlinear dynamical systems

We describe in this paper the evolution of modular neural networks using hierarchical genetic algorithms for pattern recognition. Modular neural networks (MNN) have shown significant learning improvement over single neural networks (NN). For this reason, the use of MNN for pattern recognition is well justified. However, network topology design of MNN is at least an order of magnitude more difficult than for classical NNs. We describe in this paper the use of a hierarchical genetic algorithm (HGA) for optimizing the topology of each of the neural network modules of the MNN. The HGA is clearly needed due to the fact that topology optimization requires that we are able to manage both the layer and node information for each of the MNN modules. Simulation results prove the feasibility and advantages of the proposed approach.

This paper presents a method for obtaining optimal stable control for general nonlinear nonautonomous dynamical systems. The approach is inspired by recent developments in analytical dynamics and the observation that the Lyapunov criterion for stability of dynamical systems can be recast as a constraint to be imposed on the system. A closed-form expression for control is obtained that minimizes a user-defined control cost at each instant of time and enforces the Lyapunov constraint simultaneously. The derivation of this expression closely mirrors the development of the fundamental equation of motion used in the study of constrained motion. For this control method to work, the positive definite functions used in the Lyapunov constraint should satisfy a consistency condition. A class of positive definite functions has been provided for mechanical systems that meet this criterion. To illustrate the broad scope of the method, for linear systems it is shown that a proper choice of these positive definite functions results in conventional LQR control. Control of the Lorenz system and a multi-degree of freedom nonlinear mechanical system are considered. Numerical examples demonstrating the efficacy and simplicity of the method are provided.

The authors present a neural network architecture For model reference adaptive control of nonlinear dynamical systems. The main feature of the network is its combined supervised and unsupervised learning strategy for controlling unknown dynamical systems.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

The goal of this paper is to introduce a new neural network architecture called Sigmoid Diagonal Recurrent Neural Network (SDRNN) to be used in the adaptive control of nonlinear dynamical systems. This is done by adding a sigmoid weight victor in the hidden layer neurons to adapt of the shape of the sigmoid function making their outputs not restricted to the sigmoid function output. Also, we introduce a dynamic back propagation learning algorithm to train the new proposed network parameters. The simulation results showed that the (SDRNN) is more efficient and accurate than the DRNN in both the identification and adaptive control of nonlinear dynamical systems.

We describe in this chapter a new approach for fingerprint recognition using modular neural networks with a fuzzy logic method for response integration. We describe a new architecture for modular neural networks for achieving pattern recognition in the particular case of human fingerprints. Also, the method for achieving response integration is based on the fuzzy Sugeno integral. Response integration is required to combine the outputs of all the modules in the modular network. We have applied the new approach for fingerprint recognition with a real database of fingerprints obtained from students of our institution.

Neural networks have been applied very successfully in the identification and control of nonlinear dynamic systems. The paper presents a design of neural network based control system for 2DOF nonlinear laboratory helicopter model (Humusoft CE 150). The main objective of this paper is to develop artificial neural networks to control helicopter's motors, or consequently elevation and azimuth angles. Neural networks are obtained by cloning various type of controllers designed in our previous papers. Those procedures included a cloning linear PID controller, gain scheduling controller and fuzzy controller.

This chapter presents three modular neural network architectures as systems for recognizing persons based on the iris biometric measurement of humans [80]. In these systems, the human iris database is enhanced with image processing methods, and the coordinates of the center and radius of the iris are obtained to make a cut of the area of interest by removing the noise around the iris. The inputs to the modular neural networks are the processed iris images and the output is the number of the person identified. The integration of the modules was done with a gating network method [59].KeywordsHide LayerIdentification RateAdaptive LearningBiometric MeasurementScaled Conjugate GradientThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

We describe a general method for adaptive model-based control of nonlinear dynamic systems using neural networks, fuzzy logic and fractal theory. The new neuro-fuzzy-fractal method combines soft computing techniques with the concept of the fractal dimension for the domain of nonlinear dynamic system control. The new method for adaptive model-based control has been implemented as a computer program to show that our neuro-fuzzy-fractal approach is a good alternative for controlling nonlinear dynamic systems. We illustrate our new methodology with the case of controlling aircraft dynamic systems. For this case, we use mathematical models for the simulation of aircraft dynamics during flight. The goal of constructing these models is to capture the dynamics of the aircraft, so as to have a way of controlling this dynamics to avoid dangerous behavior of the system.

We have proposed a constructive learning method, called multi-sieving learning, for implementing automatic decomposition of learning tasks and a parallel and modular multi-sieving network architecture in our previous work (1995). In this paper we present a new parallel and modular multi-sieving neural network architecture to which multiple control networks are introduced. In this architecture the learning task for a control network is decomposed into a finite set of manageable subtasks, and each subtask is learned by an individual control sub-network. An important advantage of this architecture is that the learning tasks for control networks can be learned efficiently, and therefore automatic decomposition of complex learning tasks can be achieved easily.

SummaryThis research article presents an artificial neural network (ANN)‐based indirect adaptive control method for nonlinear dynamical systems. In this article, a modified Elman recurrent neural network (MERNN) is proposed as an identifier and controller for controlling nonlinear systems. The architecture of the proposed controller is a modified form of the existing Elman recurrent neural network. The parameter training of ANN‐based controllers is obtained by using the most popular optimization algorithm which is known as the back‐propagation algorithm. A comparative study includes Elman, Diagonal, Jordan, feed‐forward neural network (FFNN), and radial basis function network (RBFN)‐based controllers to compare with the proposed MERNN controller. To determine the controller's robustness, parameter variations, and disturbance signals have been considered. The performance analysis of the proposed controller is illustrated by two simulation examples. The simulation results reveal that MERNN can not only identify the unknown dynamics of the plant but also adaptively control it compared to the others.