ANALYSIS by iaemedu


									                                           Research in Engineering
International Journal of Advanced in Engineering and Technology (IJARET)
International Journal of Advanced Research
and Technology (IJARET), ISSN 0976 – 6480(Print)                           IJARET
ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 2, Number 1, Jan - Feb (2011), © IAEME
ISSN 0976 – 6499(Online) Volume 2
Number 1, Jan - Feb (2011), pp. 01- 11                                       © IAEME

                       NEURO ADAPTIVE CONTROL
                                   Dr. Manish Doshi
                        Hemchandracharya North Gujarat University
                          Patan, E-Mail:

        Design of Intelligent system is very crucial and simple mathematical modeling is
not used for the analysis. Hence this paper represents the analysis of intelligent system
design by Neuro adaptive control. Here various methods like Neural networks for
identification, Series-Parallel Model, Supervised control, Inverse control are taken and at
last, Neuro fuzzy adaptive control is used to solve design of intelligent system.
Keywords: Adaptive Control, series parallel model, inverse control
        The adaptive control techniques described above assume the availability of an
explicit model for the system dynamics (as in the gain scheduling technique) or at least a
dynamic structure based on a linear experimental model determined through
identification (as in STR and MRAC). This may not be the case for a large class of
complex nonlinear systems characterized by poorly known dynamics and time-varying
parameters that may operate in ill-defined environments. Besides, conventional adaptive
control techniques lack the important feature of learning. This implies that an adaptive
control scheme cannot use the knowledge it has acquired in the past to tackle similar
situations in the present or in the future. In other words, while adaptive control
techniques have been used effectively for controlling a large class of systems with
predefined structure and slowly time-varying parameters, they nevertheless lack the
ability of learning and lack the ability for tackling the global control issues of nonlinear
systems. Making assumption of linear structures of processes is not always possible and
designers have to deal with the inherent nonlinear aspect of the systems dynamics. [1].

International Journal of Advanced Research in Engineering and Technology (IJARET)
ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 2, Number 1, Jan - Feb (2011), © IAEME

        As a potential remedy to some of these issues, designers have devised new
control approaches that avoid the explicit mathematical modeling of dynamic
processes. Some researchers have termed them as expert intelligent control approaches.
Some of these approaches, such as those based on fuzzy logic theory, permit for an
explicit modeling of the systems. Other approaches use well-defined learning algorithms
through which the system is implicitly modeled and adapts autonomously its controller
parameters so as to accommodate unpredictable changes in the system’s dynamics. Here,
the dynamical structure is not constrained to be linear, as is the case for most
conventional adaptive control techniques. The study of these approaches has constituted a
resurgent area of research in the last several years and several important results were
obtained. [2]
        The family of controllers using fuzzy logic theory are basically designed on the
premise that they take full advantage of the (linguistic) knowledge gained by designers
from past experiences. We tackle here another class of intelligent controllers, which are
based on neural modeling and learning. They are built using algorithms that allow for the
system to learn for itself from a set of collected training patterns. They have the
distinctive feature of learning and adjusting their parameters in response to unpredictable
changes in the dynamics or operating environment of the system. Their capability for
dealing with nonlinearities, for executing parallel computing tasks and for
tolerating a relatively large class of noises make them powerful tools for tackling
the identification and control aspect of systems characterized by highly nonlinear
behavior, time-varying parameters and possible operation within an unpredictable
        Given the fact that we deal here with dynamical models involving the states of the
model at different time steps, it is only natural to design a specialized structure of neural
networks that have the capability for “memorizing” earlier states of the system and for
accommodating feedback signals. This is in contrast with the conventional neural
networks (based on BPL) used mostly for system classification or function
approximation. In fact, despite their proven capabilities as universal approximators, a
limitation of standard feed forward neural networks, using backpropagation as the
learning mechanism, is their limitations for exclusively learning static input– output

International Journal of Advanced Research in Engineering and Technology (IJARET)
ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 2, Number 1, Jan - Feb (2011), © IAEME

mappings. While adept at generalizing pattern-matching problems in which the time
dimension is not significant, systems with a non-stationary dynamics cannot be strictly
modeled. This is the case of dynamic systems for which the representation is made
through time-dependent states. [4]
        One way of addressing this problem is to make the network dynamic, that is, by
providing it with a memory and feedbacks. A particular class of the recurrent structure is
the so-called recurrent time-delay neural networks. One way for accomplishing this is to
incorporate feedback connections from the output of the network to its input layer and
include time delays into the NN structure through their connections. As there are
propagation delays in natural neurobiological systems, the addition of these time delays
stems from theoretical, as well as practical, motivations. Time-Delay Neural Networks
(TDNNs) accomplish this by replicating the network across time. One can envision this
structure as a series of copies of the same network, each temporally displaced from the
previous one by one discrete time unit similar to the BPTT described earlier. The
resultant structures can be quite large, incorporating many connections, and novel
learning algorithms are employed to ensure rapid training of the network.[5],[6].
        To illustrate the idea let us presume that the dynamic system input–output
behavior of a nonlinear dynamic system is represented by the following equation:
        y(k 1) = f [ y(k), . . . , y(k n); u(k), . . . , u(k n)]
        where y(k 1) represents the output of network at ttime (k 1), and y(k), y(k
1), . . . , y(k n) are the delayed output states serving as part of the network input
along with the input signal u(k) and its delayed components: u(k                1), . . . , u(k
n). The network representation of such systems is depicted in Figure 1. Notice that
this network is recurrent (feedback signals) with time-delayed inputs. While several
algorithms have been proposed in the literature to deal with the training of time-delayed
recurrent networks, the dynamic backpropagation algorithm has been among the standard
techniques used for this purpose.

International Journal of Advanced Research in Engineering and Technology (IJARET)
ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 2, Number 1, Jan - Feb (2011), © IAEME

                       Figure 1 Time delayed recurrent neural network
2.1 Neural Networks for Identification
        As universal approximators, neural networks have been used in recent years
as identifiers for a relatively wide range of complex dynamic systems (linear and
nonlinear). Their capability for processing tasks in parallel and for tolerating noisy
versions of the input signals make them excellent candidates for system
identification. The process of systems identification, as mentioned previously, aims
to find a dynamical model that approximates within a predefined degree of accuracy
the actual plant dynamics when both systems are excited with the same signals. This
means that we require a minimization of the error e(k 1) between the predicted output
Np(k 1) and the actual out- put yp(k 1) of the system:

e(k + 1) = yp (k + 1) − yp (k + 1)
        Since this error depends on the parameters of the network, the solution
should provide the set of weights that would minimize it. Four major classes of
models encompass a wide range of nonlinear input–output model representations:
model1: yp(k +1) = ∑ayp(k −i) + g[u(k),......., y(k − m+1)]

mod el 2: yp(k +1) = f [ yp(k ),...., yp(k − n +1)] + ∑biu(k − i)]
                                                           i =0

model3: yp(k +1) = f [yp(k),...., yp(k −n+1)]+ g[u(k),......., y(k −m+1)]
mod el 4: yp(k +1) = f [ yp(k ),...., yp(k − n +1); u(k ),......., y(k − m +1)]
        In each of the models, the pair (u(k), yp(k)) represents the input–output pair
of the identified plant at sample k and m ≤ n. f and g are two smooth functions in

International Journal of Advanced Research in Engineering and Technology (IJARET)
ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 2, Number 1, Jan - Feb (2011), © IAEME

their arguments. However, among the four models, the fourth one has been used
most often given its sharing relevance to a wide majority of nonlinear dynamic
systems. The choice for the appropriate learning algorithm for the neural identifiers
(dynamic or static) depends mostly on whether the network takes as its inputs a
delayed version of its output, o r directly uses sample outputs from the plant itself.
The two schemes that have been used most frequently are the series-parallel
scheme and the parallel scheme. To illustrate the ideas, and if model four is taken
as the system model of choice, the two corresponding schemes are then given as:
2.2. Series-Parallel Model
        The future value pof the output N (k 1) in this model is expressed as:

yp(k +1) = NNI[ yp(k),...., yp(k − n +1)]; u(k),......., y(k − m +1)]
        where NNI stands for mapping provided by the neural network
identifier. This model is represented in Figure 2, and a careful inspection of its
structure shows that while it still requires a set of delayed signals for it inputs, it does
not involve feedback from the network output, which is the case for the second
model known as parallel.
2.3. Parallel Model
        The estimated future value of the output N (k 1) in the parallel model is

expressed as:

yp(k +1) = NNI[ yp(k),...., yp(k − n +1)]; u(k),......., y(k − m +1)]
        This model is illustrated in Figure 3 and uses the delayed recursions of the
established output as some of its input.
        One may notice here that given the particular structure of the series-
parallel model, which doesn’t involve recurrent states of the network output, the
standard BPL could be used for extracting the parameters of the network. This is,
however, not the case for the parallel structure given that the model includes a
feedback loop including nonlinear elements. As such BPL cannot be applied in this
case, and dynamic backpropagation would be the learning algorithm of choice.
        There have been a number of control schemes proposed in recent years
involving neural networks for identification and control. Following categories:

International Journal of Advanced Research in Engineering and Technology (IJARET)
ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 2, Number 1, Jan - Feb (2011), © IAEME

supervised (“teacher” or model-based) control, inverse model- based control, and
neuro-adaptive control are outlined next.

                  Figure 2 Series-parallel scheme for neural identification

                      Figure 3 Parallel scheme for neural identification

International Journal of Advanced Research in Engineering and Technology (IJARET)
ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 2, Number 1, Jan - Feb (2011), © IAEME

3.1 Supervised control
        The first category assumes that a controller is synthesized on the basis of the
knowledge acquired from the operation of the plant by a skilled operator (as shown
in Figure 4) or through well-tuned PID controllers. During the nominal operation of
the plant (at a particular operating condition) experimental data are collected from
sensor devices, and are later used as a training set for a neural network. Once the
training is carried out adequately, the process operator then becomes an upper-level
supervisor without the need for being part of the control loop of the process. This is
particularly useful in the case of hazardous environments or for improving the
automation level of the plant. The neural network could also play the role of a
gain interpolator of a set of PID controllers placed within the control loop of the
plant. Once trained, the neural network will act here as a gain feeder for the PID
controller, providing it with appropriate gains every time the operating conditions
of the system change. This is a good alternative for interpolating between the gains
instead of delivering them in a discrete way, as is the case of conventional gain
scheduling. One of the main issues, however, is that the training data collected are
usually corrupted with a large amount of noise and as such have to be filtered before
becoming useful as valid training data. This has the potential of making of the
neural network an expensive and possibly a time-consuming process. Another issue
pertains to difficulties in extracting the knowledge acquired by the experts into a
set of patterns t h at could be used for the network’s training.

                 Figure 4 Neural network acting as a supervisory controller

International Journal of Advanced Research in Engineering and Technology (IJARET)
ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 2, Number 1, Jan - Feb (2011), © IAEME

3.2 Inverse control
        The second category where neural networks could be implemented within the
control loop is known as inverse control or inverse plant modeling. Here, the
designer seeks t o build a neural network structure capable of mapping the input–
output behavior of an inverse controller. The inverse controller is by definition a
controller which, when applied to the plant, leads ideally to an overall transfer
function of unity. A schematic representation of this control scheme is shown in
Figure 4. Designing an inverse neural controller should be always done under the
assumption t h a t the process is minimum phase and causal. The main advantage of
implementing an inverse controller into a neural structure is

                Figure 4 Neural network as inverse model-based controller

International Journal of Advanced Research in Engineering and Technology (IJARET)
ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 2, Number 1, Jan - Feb (2011), © IAEME

that it would allow for a faster execution and tolerance to a range of noises. The
implementation of a neural network as an inverse controller, however, has been
hindered by several difficulties, including the induction of unwanted time delays
leading to possible discretization of processes that are originally continuous. This is
mostly due to the effect of noncausality of the inverse controller. Moreover, the
inverse controller cannot realistically match exactly the real plant, a fact that leads
to unpredictable errors in the controller design.
        The third category of neural controllers pertains to implementing two
dynamical neural networks within an adaptive control architecture, similar to the
MRAC one. One of the networks serves here as an identifier, while the second
serves as a controller. A large amount of research work has been dedicated to
controllers belonging to this category. Given the dynamic nature of the system
being identified and controlled, recurrent time-delayed neural networks have been
the tools of choice for this case. The main advantages of these neuro-adaptive
structures pertain to their capability in effectively tackling the nonlinear behavior
of the systems without compromising on their representation using linear
approximations (such as the ARMA model auto-regressive moving average). This
has not always been possible with the conventional schemes of adaptive control
systems as described in earlier sections. As in the identification section, the
controller here is handled using another            structure     of   a   time-delay     neural
network for which the output serves as input to the plant. The same output is
delayed in time and fed back to the network to serve as part of its input signals.
Now combining the two aspects (identification and control) within a well-defined
adaptive structure such as the one described in a previous section leads to the
representation of Figure 5. In this structure, which is very similar to the MRAC,
identification is carried out first, and the identified model is then compared with
the output of a reference model. The recorded error is then used to update the
control law through modifications of the neural controller weight. This process is
described in ample detail. Despite the fact that the structure is similar to the MRAC,
the current scheme developed here is known as an inverse neuro-adaptive control

International Journal of Advanced Research in Engineering and Technology (IJARET)
ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 2, Number 1, Jan - Feb (2011), © IAEME

scheme. This is mainly due to the fact that the error provided t o the neural
controller is n ot computed directly as the difference between the plant output
and the reference, but rather between the model identified and the reference.

                    Figure 5 Neural network as neuro-adaptive controller
        The motivation for the early development of neural networks stemmed from
the desire to mimic the functionality of the human brain. A neural network is an
intelligent data-driven modeling tool that is able to capture and represent complex and
non-linear input/output relationships. Neural networks are used in many important
applications, such as function approximation, pattern recognition and classification,
memory recall, prediction, optimization and noise-filtering. They are used in many
commercial products such as modems, image-processing and recognition systems, speech
recognition software, data mining, knowledge acquisition systems and medical
instrumentation, etc.
    1. Fausett, L., Fundamentals of Neural Networks, Prentice-Hall, Englewood Cliffs,
        NJ, 1994.
    2. Ham, F. and Kostanic, I., Principles of Neurocomputing for Science and
        Engineering, McGraw Hill, New York, NY, 2001.

International Journal of Advanced Research in Engineering and Technology (IJARET)
ISSN 0976 – 6480(Print), ISSN 0976 – 6499(Online) Volume 2, Number 1, Jan - Feb (2011), © IAEME

    3. Haykin, S. (1994) Neural Networks: A Comprehensive Foundation, Englewood
        Cliffs, NJ: McMillan College Publishing Company.
    4. Hopfield, J.J. and Tank, D.W., Computing with Neural Circuits: A Model,
        Science, Vol. 233, 625−633, 1986.
    5. Hopgood, A. (1993) Knowledge-based Systems for Engineers and Scientists,
        Boca Raton, Florida, CRC Press, 159–85.
    6. Jang, J. S., Sun, C. T., and Mizutani, E., Neuro-Fuzzy and Soft Computing,
        Prentice Hall, Englewood Cliffs, NJ, 1997.


To top