Active Training On The Cmac Nonlinear Adaptive System

Informatica

The conventional neural network (NN) CMAC (Cerebellar Model Articulation Controller) can be applied in many real-world applications thanks to its high learning speed and good generalization capability. In this paper, it is proposed to utilize a neuro-evolutional approach to adjust CMAC parameters and construct mathematical models of nonlinear objects in the presence of the Gaussian noise. The general structure of the evolving NN CMAC (ECMAC) is considered. The paper demonstrates that the evolving NN CMAC can be used effectively for the identification of nonlinear dynamical systems. The simulation of the proposed approach for various nonlinear objects is performed. The results proved the effectiveness of the developed methods. Povzetek: Razvit je postopek za evolucijsko iskanje najbolj prilagojene CMAC (Cerebellar Model Articulation Controller) nevronske mreže za probleme z Gaussovim šumom.

Proceedings of 1994 IEEE 3rd International Fuzzy Systems Conference, 1994

In the past, various neural network-bed controllers are proposed to master the nonlinear control problems with different level of success. The recent trend is to incorporate fuzzy logic to this process. This article compares a neural network-based controller, both local and global networks, with Fuzzy associative memories (FAM) on a M)nliDBpT problem. CMAC and F A M are chosen as representatives of local generalization networks. CMAC controller is tmbd off-line, therefore, it can response to the incoming input immediately. CMAC can intrapolate its memory m d give a reosonoble control signal even the input bas not been trained on. Backpropagation is picked =a*-'ve of global genedization networks. All three systems are studied on a simple simulated control problem. This p " r y te88Luch will be adapted later to control the laser cutting machine. A performance measure that depends on the transient response and the steady state response of the controlled system is used. The d t s indicate that CMAC and F A M are comparable.

DESCRIPTION The convergence of parameters in model reference adaptive control (MRAC) requires that a restrictive persistence of excitation (PE) condition be satisfied. A recent data-driven approach, concurrent learning, uses past input-output data in conjunction with standard adaptive laws to ensure the parameter convergence without needing the PE condition. However, the concurrent learning method assumes the knowledge of the state derivative, which is a limitation. This paper combines a state derivative estimator with concurrent learning to guarantee parameter convergence, thus eliminating the need for both the PE condition and the knowledge of the state derivative. Simulation results are presented to demonstrate the effectiveness of the proposed control method.

1998

ttractive methods for learning the dynamics and improving A the control of robot manipulators during movements have been proposed for more than 10 years, but they still await applications. This article investigates practical issues for the implementation of these methods. Two nonlinear adaptive controllers, selected for their simplicity and efficiency, are tested on 2-DOF and 3-DOFmanipulators. The experimental results show that the Adaptive FeedForward Controller (AFFC) is well suited for learning the parameters of the dynamic equation, even in the presence of friction and noise. The control performance along the learning trajectory and other test trajectories are also better than when measured parameters are used. However, when the task consists of driving a repeated trajectory, the adaptive lookup table MEMory is simpler to implement. It also provides a robust and stable control, and results in even better performance.

A one-by-one learning algorithm similar to traditional incremental learning for CMAC is suggested. The convergence properly is investigated based on the principles of geometric sequence and iteration theory of linear equations. The sufficient condition for the convergence of the algorithm is the same as that of incremental learning. The performance of two algorithms is compared, then a hyhrid one-by-one learning with incremental learning algorithm is proposed. The simulation results about hvo-dimension function approximation prove the hybrid algorithm has better performance in convergent speed and precision.

Control Engineering Practice, 1997

Neural identification and control techniques are well-suited to the problem of controlling robot dynamics. This paper describes the use of CMAC networks for the adaptive dynamic control of an orange-harvesting robot. Among the various neural-network p~,~ligms available, the CMAC model was chosen in this c~c because of its f~t convergence and on-line adaptation capability. The solution of this dynamic control problem with CMAC is an encouraging demonstration of "experience-based', as opposed to model-based, control techniques ,and is a good example of the use of on-line learning in adaptive neural control

Cerebellar Model Articulation Controller Neural Network is a computational model of cerebellum which acts as a lookup table. The advantages of CMAC are fast learning convergence, and capability of mapping nonlinear functions due to its local generalization of weight updating, single structure and easy processing. In the training phase, the disadvantage of some CMAC models is unstable phenomenon or slower convergence speed due to larger fixed or smaller fixed learning rate respectively. The present research deals with offering two solutions for this problem. The original idea of the present research is using changeable learning rate at each state of training phase in the CMAC model. The first algorithm deals with a new learning rate based on reviation of learning rate. The second algorithm deals with number of training iteration and performance learning, with respect to this fact that error is compatible with inverse training time. Simulation results show that this algorithms have faster convergence and better performance in comparison to conventional CMAC model in all training cycles .

IEEE Control Systems, 2000

T h e Cerebellar Model Articulation Controller (CMAC) neural network is capable of learning nonlinear functions extremely quickly due to the local nature of its weight updating. The rectangular shape of CMAC receptive field functions, however, produces d i s c o n t i n u o u s ( s t a i r c a s e ) f u n c t i o n approximations without inherent analytical derivatives. The ability to learn both functions and function derivatives is important for the development of many on-line adaptive filter, estimation, and control algorithms. It is shown that use of B-Spline receptive field functions in conjunction with more general CMAC weight addressing schemes allows higher-order CMAC neural networks to be developed that can learn both functions and function derivatives. This also allows novel hierarchical and multi-layer CMAC network architectures to be constructed that can be trained using standard error back-propagation learning techniques.

Systems, 2014

We survey some of the rich history of control over the past century with a focus on the major milestones in adaptive systems. We review classic methods and examples in adaptive linear systems for both control and observation/identification. The focus is on linear plants to facilitate understanding, but we also provide the tools necessary for many classes of nonlinear systems. We discuss practical issues encountered in making these systems stable and robust with respect to additive and multiplicative uncertainties. We discuss various perspectives on adaptive systems and their role in various fields. Finally, we present some of the ongoing research and expose problems in the field of adaptive control.

A control problem arising in typical fermentation process studies is solved here. Fermentation, i.e., microbial growth and substrate consumption, is described by a nonlinear differential system including a set of unknown parameters that may vary in time. The control objective is to get the state of the system to track the state of a given reference model despite the disturbances and system parameter uncertainties. The concentration of microbes is almost always impossible to determine on-line. The evolution of the main substrate is, however, measurable on-line. The system studied is controlled by varying the dilution rate of the concentrated substrate liquidfeed. Thefixedparameter control law that serves as a basisfor the adaptive method uses the calculated references of the specific growth rate and the dilution rate ofthe reference model. Adaptive state estimation is based on obtaining a stable estimatorfor the joint system of states and parameters via a Lyapunov technique. The structure of the adaptive controller is determined by the requirement to obtain stable reference model tracking. Application of Lyapunov's method gives a PI-type controller with adaptively adjustable coefficients. Given stability proofs are supported by some realistic simulation results. studied by Bastin and Dochain,' gradient-based estimation applied in Chamilothoris and Sevely,' modulated gain estimation by Dahhou3 and the extended Kalman filter utilized by many authors, for example Flaus,4 and Nihtila et al. ' On the process control side, depending on the control objective stated, different adaptive methods have been developed and applied. Predictive2,6 and pole place-ment3 applications are based on established techniques developed, for example by Clarke and Gawthrop' and by Astrom and Wittenmark.' Model reference adaptive control techniques developed by Landau' are also applicable although few applications in fermentation processes have been reported."