You are on page 1of 11

IJEIT Vol. 1 No.

1 Oct 2009

An Artificial Intelligence based Control for Micro Hydro Power Plants


Himani Goyal, M.Hanmandlu

Abstract-- Micro hydropower plants are emerging as a major renewable energy resource today as they do not encounter
the problems of population displacement and environmental problems associated with the large hydro power plants.
However, they require control systems to limit the huge variation in input flows expected in rivulets over which these are
established to produce a constant power supply. This paper proposes an electric servomotor as a governor for a micro
hydro power plant especially those plants that are operated in isolated mode. An advanced controller is developed
combining four control schemes for the control of the governor following the concept that the control action can be split
up into linear and non linear parts. The linear part of this controller contains an adaptive Fast Transversal Filter (FTF)
algorithm and normalized LMS (nLMS) algorithm. The non-linear part of the controller incorporates Fuzzy PI and a
neural network. The concept behind splitting the control action is reasoned out and the conditions for stability of the
controller are proved. The new controller has a superior performance compared to other control schemes.
Index Terms-- Hydro power plants, ANN, FTF, Fuzzy PI, nLMS, Non-linear systems, PI Controller, Servomotor,
Stochastic Load Disturbance

I INTRODUCTION connected. The current trend is therefore to use load


side regulation.[ 2,3,5,6]
In an electric power system, consumers require B. Electronic Load Controller for Water Turbines
uninterrupted power at rated frequency and voltage. Electronic load controllers govern the turbine speed by
To maintain these parameters within the prescribed adjusting the electrical load on the alternator. As lights
limits, controls are required on the system. Voltage is and electrical appliances are turned on and off, the
maintained by the control of excitation of the electronic controller varies the amount of power fed
generator and frequency is maintained by eliminating into a ‘ballast’ load. The load controller therefore
the mismatch between generation and load demand. maintains a constant electrical load on a generator in
Since frequency is an indicator of the energy balance spite of changing user loads. This permits the use of a
in the system, the problem of maintenance of turbine with no flow regulating devices and the
constant frequency is analyzed in this paper. A novel governor control system. Load controllers however
scheme is proposed for the speed control of hydro waste precious energy that can be used gainfully. Also
turbines. This scheme regulates the flow of water they do not carry out flow control implying that the
being fed to the turbine in accordance with the load mineral rich water is made to spill away which could
perturbations thereby maintaining the frequency of have been diverted at high head for irrigation purposes.
the system at the desired level. Henderson [39, 40, 41] describes development of a
microprocessor based electronic load governor for a
A. Conventional Governors micro hydroelectric power plants. The governor
Conventional Governor Systems can be maintains the speed of the set by adjusting an electrical
classified as mechanical-hydraulic governors, electro ballast load connected to the generator terminals, thus
hydraulic governors or mechanical types. Mechanical balancing the total electrical load torque with the
hydraulic governors are sophisticated devices which hydraulic input torque from the turbine. [8]
are generally used in large hydro power systems.
They require heavy maintenance and are expensive to C.Servomotor as a Governor
install, making their usage in micro hydro power
In the proposed control system an electric servomotor is
plants uneconomical. Electro hydraulic governors
used as a governor [1]. An electric servo motor is a
are complex devices needing precision design and are
expensive. Mechanical governors incorporate a precision electric motor whose function is to cause
massive fly ball arrangement and usually do not motion in the form of rotation or linear motion in
provide flow control. They require an elaborate set of proportion to a supplied electrical command signal.
complex guide vanes, inlet valves and jet deflectors. Type Zero servomechanism is used in the proposed
Hence conventional governing systems because of system. A feedback control system of Type Zero is
their cost and complexity are not ideally suited for generally referred to as a regulator system. Such
installing at the isolated areas that are not grid systems are designed primarily to maintain the
controlled variable as constant at a certain desired value

33
IJEIT Vol. 1 No. 1 Oct 2009

despite disturbances. Here the controlled variables II gives formulation of the state space model for the
are the frequency and the turbine power. The electric hydro plant. Some of the advanced control schemes are
servo motors are preferable for the control of micro discussed and a new controller is proposed based on the
hydro power systems as they have a simple design, concept of splitting up of control action in Section III.
require less maintenance and are less expensive than Performance of this new controller is compared with
conventional governors. other control schemes to illustrate its effectiveness in
Section IV. Conclusions are drawn in Section V. The
D. Literature Survey concept and the proof of stability find place in the
Glattfelder [30] has advocated use of a compensating Appendices.
element in addition to a speed regulator as a speed II. FORMULATION OF PLANT MODELS FOR MHP
governor for low head hydro units in an isolated grid. PLANT
Hagihara et al. [31] consider the effect of derivative
gain and other governor parameters on the stability of The approximate transfer function for the servo motor
a hydraulic turbine supplying an isolated load. based governor is considered for the analysis and is
Pereira et al. [32] propose the addition of eddy given by
current brake to help governor achieve greater degree
of control. Schniter et al. [33] make use of the G(s) = 1 1
adjustable blade angles to achieve maximum (1)
operating efficiency at a given load. Tano and (1 +sT1) (1+sT2)
sannomiya [34] have developed an integrated digital where, T1 = mechanical time constant, T2 =
control panel for control, protection and supervisory electrical time constant. In addition, unity gain is
functions. Malik et al. [35] recommend the use of applied as a feedback. A PI controller with the
microprocessor based governor for frequency following transfer function is superimposed on the
measurement and control and Thappar et al. [36] servomotor based governor:
present the distributed microprocessor based control G(s) = Kpl + Ki
for stable frequency and voltage output. Djukanovic s
et al. [37] propose the non linear multivariable Where Kpl = Proportional constant, Ki =
control using adaptive network based fuzzy inference Integral constant
system (ANFIS)

E. Need for a New Governing System

The new Governing System should be relatively


inexpensive, simple to operate and easy to maintain
by incorporating the flow control. Therefore, we will
explore a set of advanced control schemes for better
load frequency control.

In this work, we will discuss how the control action Fig. 1: Model of a Micro hydro power plant using
can be split up into linear and non linear components
as this will allow the selection of appropriate control Servomotor as a Governor
schemes for the micro hydro power plant. In the light
of this concept we will explore different control The block diagram of micro hydro power (MHP) plant
schemes consisting of combinations of linear and is shown in Fig.1. This plant can be reduced to a
nonlinear parts to come out with a scheme that meets simpler representation as in Fig. 2 by employing partial
the desired performance in terms of peak overshoot fractions.
and steady state error. We will also show that the
chosen scheme is stable. Thus this work will pave the
way for the design of a suitable controller for any
type of plant by an appropriate choice of control
components.

The organization of this paper is as follows: Section

34
IJEIT Vol. 1 No. 1 Oct 2009

where X,  and p are the state, control and


disturbance vectors respectively and [A], [B] and

[] are constant matrices of appropriate dimensions.


Using Eqns. (2)-(8), Eqn. (9) is rewritten as :

Fig. 2: Simplified representation using Partial


Fractions
The differential equations for the governor can be
written as:
d d
 X E 1  K i f  K p l  f
d t d t

(2)
d 1 1 1
XE 2  XE 2 [  X E 3   f   X E 1]
dt T2 T2 R

(3)
d XE3  1 XE3  1 XE2
dt T3 T3 (10)
(4)
III. ADVANCED
The change in power generated is given by:
PI controllers provide a good control action for micro
 P g  C [  X E 3]   P g 1 hydro power plants. However the control action can be
(5)
further improved using advanced control methods. In
The differential equations for the hydro turbine are as
these methods, the representation and adaptation of
follows: information are the key issues to reduce complexities
and to eliminate the heuristic procedures in process
d 1 D
Pg1  Pg1  [XE 3] control. Moreover, good transient and steady state
dt 0.5Tw1 0.5Tw1 responses for different operating points of the processes
(6) can be achieved. These advanced techniques include
the Fuzzy PI, the nLMS algorithm, the FTF algorithm
and a Neural Network (NN). Here NN incorporates an
d 1 Kp adaptive algorithm, which is (7)a combination of nLMS
f  f  [Pg1 C.XE3  PL] and gradient descent algorithms. Fuzzy techniques
dt Tp Tp drastically reduce the development time and cost for the
. In view of the above, Eqn. (2) becomes synthesis of nonlinear controller for dynamical systems.
d 1 Kp Triangular membership functions are generally preferred.
XE1  Kif  Kpl[ f  [Pg1  C.XE3 PL]] (8) in many applications
dt Tp Tp The nLMS algorithm is used
where the input signals are subject to widely fluctuating
power levels causing gradient noise amplification,
From Eqns. (3)-(8) , the parameters are determined which in turn affects the stability, convergence and
as C = -2, D = 3. The system dynamics described by steady-state properties of the LMS algorithm. The
the above set of differential equations appears in the advantage of nLMS in micro hydro power plant is that
state space form as: it adapts the gain to its optimal value, resulting in fast,
X  [ A ] X  [ B ]  [ ] p stable convergence. The FTF algorithm, a well-known
(9) tool in the field of signal processing, is also applied for
control. In FTF algorithm, projection techniques and

35
IJEIT Vol. 1 No. 1 Oct 2009

vector space methods are used to derive a fixed order


transversal least squares filter. A new controller is
now proposed using a combination of the above
mentioned techniques with a view to minimize the
peak overshoot and achieve the early settling time by
tapping the advantages of individual components.
The schematic of the proposed controller is shown in
Fig. 3.

Fig 4: The block diagram of the proposed Scheme of


control
The combined output of the controller is

1) u (k )  u1 (k )  u2 (k )  u3 (k )  u4 (k )
(11)
Fig. 3: The model of the proposed scheme where u1(k) is the output of NN, u2(k) is the output of
FTF, u3(k) is the output of Fuzzy *PID and u4(k) is the
A.Different control combinations output of n-LMS.
The new controller incorporates the advanced control
techniques to cater to the plant linearities and (a)Scheme of control employing FTF and Neural
nonlinearities. The underlying concept is that the Networks
non-linear part of the controller operates on non- A combination of FTF algorithm and a neural network
linear error signals to make them adhere to the set from [19-21] is shown in Fig. 5. The weights of the
point and the linear part of the controller tries to neural network are adjusted by n-LMS and gradient
maintain the linearity between the two points. This descent algorithms. [13]
concept advocated by Srivastava in her thesis [19] is
adapted here to justify the composition of the new
controller, and is enumerated in [42]. From this
appendix it is possible to choose appropriate linear
and nonlinear components of controller such that the
desired performance is achieved. [12]. Moreover, it
has been noticed that each technique retains its
advantage and contributes to the overall performance
of the controller. Keeping this in mind, our control
scheme shown in Fig. 4 is devised. [38] In this, Fig. 5: A Scheme of Control employing FTF and Neural Networks
nLMS and FTF constitute the linear part and Fuzzy
PI and NN constitute the non linear of the controller.
These four constituents are now discussed in detail 2) The output of the NN, denoted by u1(k) is
followed by the stability analysis of the controller
M 
given in Appendix A.

u1 (k )  ψ  wn (i )e(i )  b 
 i 1 
(12)
where M is the number of samples taken at a time,
e  r  y is the error, y is the actual output, r is the set
point. The somatic gain denoted by ψ is the advanced
feature that improves the performance of the neural
network. The NN used here is composed of only input

36
IJEIT Vol. 1 No. 1 Oct 2009

and output layers. The weights of the NN are K i is independent of the magnitude of the step
adjusted by the following equation [10]:
disturbance. [9,11,15,17,18]
V
wn (k  1)  wn (k )  wn (k )  
wn 3) The FTF controller part

(13) 4) The output of the FTF, denoted by u2(k) is

wn (k )  α ' e(k )  e(k )


u2 (k )  w f 1  r (k )  w f 2  r (k  1)
(15)
(14)
where w f 1 and w f 2 are the FTF weights. The
1 2  computation of these weights can be found from [16].
where V  e and α '  2 , with  and  being
2 e (b) Fuzzy PI Control

the step sizes used in  - LMS and gradient learning A schematic diagram of the Fuzzy PI controller is given
respectively. in Fig. 7.

Off line parameter optimization using ANN

Fig. 7 Scheme employing Fuzzy PI control

The output of a product-sum fuzzy controller is of the


form [22-26]:
u = A + PE
where E = ke e = ei , A is the input membership
function, e is the error , u is the output of fuzzy
controller, ke is the scaling factor and P is equivalent
proportional term [25]. Hence, the control input to the
plant can be approximated by
Fig. 6: Block diagram of MLP feed forward ANN u3(k) = α. [A + PE] + β∫ (A+PE).dt
= α.A+ αPE + βA t +βP∫ ke.e.dt
The training of parameters using Multi Layer = α.A+ αPE + βA t + βP k e ∫.e.dt
Perceptron (MLP) feed forward ANN (See Fig. 6) is = α.A+ αP k e.e + βA t + βP k e ∫edt
now discussed. Inputs to the ANN are different
values of regulation parameter R, water starting time Where α is the weight on PD type fuzzy controller and
Tw and nominal loading. Output of ANN is the β is the weight on PI type fuzzy controller. Here, the
fuzzy controller becomes a parameter cum time-varying
desired proportional gain K p and integral gain, K i . PI controller with its equivalent proportional control
Prior to conducting training, a set of input–output and integral control components being αPke and βPke
patterns is first prepared. The network is trained until respectively. Hence this Fuzzy PI controller behaves
a good agreement between predicted gain settings as the PI type fuzzy controller. In (15) the derivative of
and the actual gains is reached. Once the network is e for the fear of accentuating the stochastic disturbance
adequately trained, the network is again tested to is not used.
ensure that it can adequately predict the correct gain (c)Normalised LMS Control
settings for the inputs that are not included in the The weight update in α -LMS rule is given by
training set. The results show that predictions by Wk+1=Wk+α.Єk.Xk/|Xk|2
ANN are in good agreement with the training data. It (18)
is observed that parameter optimization of K p and The time index or adaptation cycle number is k, Wk+1 is
the next value of the weight vector, Wk is the present
value of the weight vector and Xk is the present input

37
IJEIT Vol. 1 No. 1 Oct 2009

Type max 1st 2nd 3rd J= (SE)


{ o ,  u } ts
u o u
a 7 7 0.99 0.427 0.353 1.38178
b 4 5 0.834 0.291 0.267 0.90322
c 6 6 1 0.423 0.317 1.34276
5
d 3 4 0.6929 0.2869 0.2545 0.68026
e 5 2 0.993 0.22 0.2449 1.10637
9
f 2 3 0.691 0.28055 0.22439 0.64025
g 1 1 0.69 0.2798 0.1655 0.58978
0

pattern vector. The present linear error Єk is the


difference between the desired response sk and the
linear output rk = WkT.Xk [14]
Hence
ΔЄk=Δ(sk-WkT.Xk)=-XkT.ΔWk Fig.8: Flowchart
(19)
Rewriting the α -LMS rule from Eqn. (19), we have
ΔWk=Wk+1–Wk= α.Єk.Xk / |Xk|2
(20)
In view of Eqn. (20), Eqn. (19) is written as
Δ Єk = - α . Єk .XkT.Xk / |Xk|2 = - α . Єk (21)
But in our case Δf = Δek; so the output u4(k)= WkT. ek
.
Stability and speed of convergence depend on the
parameter α. For input pattern vectors independent in
time, stability is ensured for most practical purposes
if 0 < α < 2. Choosing α greater than 1 results in
over corrections hence practical range of α is 0.1 < α t (sec)
< 1.0. The analysis of Lyapunov stability of this Fig.9 Comparison of different control
controller is provided in Appendix-A.
IV. COMPARISON OF PERFORMANCE As shown in F
The results of simulations demonstrate that in general
Exhaustive and comprehensive simulations are the application of the advanced control techniques
carried out on micro hydro power plants using the using the linear and non linear parts improves the
proposed controller designated as “g” as well as performance of the micro hydro power plants as
different combinations of the control schemes (a- f) compared to a simple PI controller. The proposed
with a view to come up with an efficient combination
controller “g” outperforms over all other controllers by
that can reduce the peak overshoot and ensure quick
achieving the least peak overshoot and the early settling
stabilization of the supply frequency. The flow chart
time as shown in Table 1 where it may be observed that
of the controller scheme useful for implementation
purpose is shown in Fig. 8. The data model employed all controllers (i.e., d, f, & g) having both linear and non
linear parts have less Squared Error (SE). In Table 1,
is given in Appendix-B. The plots of f vs. t for
different control schemes are shown in Fig. 9 to
we denote the under shoot and over shoot by u and
ascertain their relative performance. The composition 0 respectively and the settling time by ts . As is seen
of various schemes is as follows:
from this Table 1 here, controller “g” has the minimum
squared error whereas controller “a” has the maximum
squared error. Therefore we rate the controller as Best
in terms of there squared errors.

38
IJEIT Vol. 1 No. 1 Oct 2009

Table 1: Comparison of Control Combinations (Here could be to switch them in a particular


1: Best; 7: Worst) sequence of linear and nonlinear parts so as to
achieve the similar performance but at the
Table 1 indicates guidelines for the choice of control reduced power consumption and utility.
components. We will first try a linear component one
after another and see the performance. Which ever
is the best we will select that one. Next we repeat the APPENDIX -A
experiment with one non-linear component at a time.
We will analyze the performance. This will be Analysis of the Proposed Scheme using Lyapunov
continued with next set of linear components Stability Criterion
followed by the non-linear components until we get
the desired combination of control components. A stability analysis of the proposed scheme using
Lyapunov’s criterion [19-21] is now discussed here. Let
the Lyapunov function be selected as:
V.CONCLUSIONS
1
V  e2
The maintenance of desired power generation and 2
frequency of micro hydro power plants using flow Where e  r  y is the error; r is the set point and y is
control is the main theme of this paper. An electric
the plant output. Now,
servomotor is suggested as a governor for these
 
plants. A state space model of the plant is derived. dr 
e   y , As r0
A new scheme is proposed for the control of micro dt
hydro power plants. This scheme consists of dividing dV  de 
the controller action into two parts: Linear and Non-  V  e  ee
linear. The non-linear part of the controller dt dt
incorporates Fuzzy PI and NN. The linear part is The plant output y is taken as:
based on the FTF and nLMS algorithms. The concept y  g [u1  u 2  u3  u4]
behind splitting the controller is that any nonlinear
error signal is better represented by linear and where g is the non-linear function of plant, u1 is the
nonlinear parts. The conditions for the stability of the output of NN and u2 is the output from FTF, u3 is
controller are derived using the Lyapunov criterion.
The performance of the new controller and several output of Fuzzy PI and u4 is the output of n-LMS.
others with different linear and linear components is These are given by:
compared. The new controller is found to be superior
M 
to that of other control techniques in terms of u1 (k )  ψ  wn (i )e(i )  b ;
reducing the sum of the squared error. Thus this work  i 1 
makes an important contribution in terms of splitting
of control action, stability proof and providing
u 2 (k )  w f r ;
guidelines for the choice of new control schemes. We u3(k) = α.A+ αPke.e +βA t+ βPke ∫edt ,
have also implemented the proposed controller on
small hydro power plants and the results are found to u4(k) = WkT.ek
be equally good as that of MHP. where wn are the weights of NN and w f are the
There are reservations in some circles
weights of FTF. Now,
on the use of sophisticated control schemes
dy  g du1 g du2 g du3 g du4
for the control of micro hydro power plants. y   
dt u1 dt u2 dt u3 dt u4 dt
However, we justify this study on the
ground that analysis of different schemes    
and their combinations will lead to better gu1 u1gu2 u2gu3 u3gu4 u4
understanding of the control action and will (A.6)
pave the way to a practically viable scheme.
For example an alternative to the proposed
controller in which all the individual
schemes are simultaneously functioning

39
IJEIT Vol. 1 No. 1 Oct 2009

   T 
du1 ψ dwn ψ de  gu ψw wn e  gu wf r  gu ( A Pkee)  gu W k.e 
 u1  e wn e 1 n

T  0
2 3 4

dt wn dt e dt  1 gu ψewn  gu W k 
  1 4
 
 ψ wn wn e  ψ e e wn (A.7)   T 
 gu ψw wn e  gu wf r  gu ( A Pkee)  gu W k.e 
dw f e 0
1 n 2 3 4

du 2   T
 u2  r  wf r 
1 g ψ w  g W (A.8) 
dt dt  u e n u k
1 4 
du 3  y dy 
 u 3   A   Pk ee (Note:Taking  y x and  y)
dt x dt
(A.9)  w (t  1)  wn (t )
T T Again as wn  n  wn (t )
du4   
t (A.10)
u4 W k. ek  W k. ek
dt 2 y u1 2
 wn (t )   ' e   e   ' e   eyu1 u1w
u1 wn n

Using Eqns. (A.7)-(A.10), Eqn. (A. 4) is modified as


 
y  gu1ψwn wn e  gu1ψe ewn  gu2 wf r 
   wn (t )  e( ' e  y u1 u1 wn )
(A.11)
(From -LMS and gradient descent algorithms)
T T 
gu3 ( A Pkee)  gu4 (W k.ek  W k.ek )   w f (t
From Eqn. (A.4) we get
Similarly, wf 
   
 y  gu1 ψwn wn e  gu1 ψe y wn  gu2 w f r  where c is the gain vector (From FTF algorithm).
Using Eqns. (A.14)- (A.16) ,we have
 T T   
T

gu3 ( A   Pkee)  gu4 ( W k.ek - W k. y )  gu ψw ( ' e   eyu u1w )e  gu cer  gu ( A   Pkee)  gu
2
W k.e 
e 0
1 n 1 n 2 3 4

   1  gu ψe wn  gu W k 
since e k  y and e k  e  1 4 
T
  

 g u ψ w wn e  g u w f r  g u (  A   Pkee )  g u W .e  
T

 (gu ψw  ')e  ( yu u1w gu ψw )e  (gu cr  gu Pke  gu 
k 3 2
W k)e  gu  A
 y
1 n 2 3 4

 0
1 n 1 n 1 n 2 3 4 3

T
T

1  g u ψ e wn  g u W k .  1  gu ψe wn  gu W k 
1 4
 1 4 
(A.12)
In view of Eqn. (A.12), Eqn. (A.3) is changed to 
T

   3
[( gu ψw  ')e  ( yu u1w gu ψw )e  ( gu cr  g u  Pke  gu W
2
k )e  gu  A ]  0
V  e e  e y 1 n 1 n 1 n 2 3 4 3

   
T
 For β << 0, the product βA <<0 as A is a membership
  gu ψw wn e  gu wf r  gu ( A   Pkee)  gu W k.e  function that assumes a value between 0 and 1. Hence
V  e  
1 n 2 3 4

T for small values of β, Eqn. (A.18) becomes,


 1  gu ψe wn  gu W

 1 4
k
 T
[(gu1 ψwn  ')e2  ( yu1 u1w gu1 ψwn )e1  (gu2 cr  gu3  Pke  gu4 W k ) ]  0
   T  n

  gu1 ψwn wn e  gu2 wf r  gu3 ( A Pkee)  gu4 W k.e  Solution of the above equation is given by :
V  e T  (A.13)
  
T

 1 gu1 ψewn  gu4 W k  2


( yu u1 gu ψw )  ( yu u1 gu ψw )  4(gu ψw  ')(gu cr  gu Pke  gu W k)
1
w 1 n 1
w 1 n 1 n 2 3 4

Applying the condition for the stability in Eqn.


n n

e(1 , 2 ) 
 2(gu ψw  ')
V  0 , we have
1 n

(A.13),

40
IJEIT Vol. 1 No. 1 Oct 2009

where (1 ,  2 ) are the roots. So the range of error is: From Case1, we have
  T
min(1, 2 )  e  max(1, 2 ) . As our motive e  0, c  0  w f  0, P  0, W k  0 , So
is to minimize the error (e), we will discuss two that Eqn. (A.22) becomes
cases. yu u1 w  0 . In other words, we can rewrite the
1 n
Case 1:
It is possible when roots of the above equation tend above for minimizing the error as: yu u1w  0 .
1 n

to zero, as e  0 . Then, we have Therefore, we can conclude by using Eqn. (A.15) that
  T
T
2
 e  0,c  0  w f  0, P  0, W k  0,
( gu ψw  yu u1w )  ( gu ψw  yu u1w )  4( gu ψw  ')(gu cr  gu  Pke  gu W k)  0
1 n 1 n 1 n 1 n 1 n 2 3 4


 y u1 u 1 w n  0 , w n  0
(A.20)
Hence, It may be noted that all the conditions in Eqn. (A.23)

 T
tend to zero indicating that the convergence of the
( gu1 ψwn  ' )(gu2 cr  gu3 Pke  gu4 W k)  0 system is guaranteed. Thus these conditions are the
(A.21) necessary conditions for stability. It is now proved that
This gives the solution as this control combination assures stability. Next one can

  T explore its performance in terms of reducing the sum of


c  0  w f  0, P  0, W k  0 , which is the squared error defined by
true for FTF algorithm , also P  0, means that the
J   SE   e 2
change in proportional part tends to zero and if
 T which in turn reduces the peak overshoot and settling
W k  0 , i.e. , change in nLMS weight tends to
time. Here the desired trajectory is zero but one can
zero as change in error tends to zero.
possibly take it as the second order response with the
Case 2:
specified first overshoot and settling time. The
 2
  T
performance measure in (A.24) is used to compare the
(gu ψw  yu u1w ) 4(gu ψw ')(gu cr  gu Pke  gu W k)  0
new controller with other control schemes.
 
1 n 1 n 1 n 2 3 4


T APPENDIX –B
2
(gu ψw  yu u1w )  4(gu ψw  ')( gu cr  gu  Pke  gu W k)
1 n 1 n 1 n 2 3 4

 T
4 '.( g u2 cr  g u3  Pke  g u4 W k) Data for the Model
( yu1 u1w ) 2 
n
g u1 .ψ wn The following data is considered for constructing the
model.
 T 1.Total rated capacity : 50 kW
 '.( g u2 cr  g u3  Pke  g u4 W k)
 yu1 u1w 2 2.Normal operating Load : (A.22) 25 kW
n
g u1 .ψ wn
3.Inertia Constant H: 7.75 seconds (2<H<8)
4.Regulation R: 10 Hz / pu kW

Assumption: Load - frequency dependency is linear.


Nominal Load = 48 % = 0.48;  Pd = 3 % = 0.03.

41
IJEIT Vol. 1 No. 1 Oct 2009

[5] Harvey, A., Brown, A., Hettiarachi, P. and Inversin,


The damping parameter [4,7], D =  Pd/  f = 0.48 A., Micro Hydel Design Manual, A Guide to Small
25 Scale Water Power Schemes, Intermediate Technology
= 0.004 pu kW / Hz 60  50 Publications, 1993.
[6] IEEE Committee Report, Dynamic models for
Generator parameters are: steam and hydro turbines in power system studies,
IEEE Trans. on Power Apparatus and Systems, pp
Kp = 1 = 250 Hz / pu kW 1904-1915,1973,.
D [7] Kundur, P., Power system stability and Control,
Tata-McGraw Hill Co. 1221, Avenue of the Americas,
Tp= 2  H = 64.64 seconds New York, NY, 1994.
f0  D [8] Working Group on Prime Movers, Hydraulic
turbine and turbine control models for system dynamic
The open loop transfer function of a servomotor is studies, IEEE Tran, on Power Systems, 7, 1992, pp.
given by [27, 28, 29] 167-179.
[9] Kumpati S. Narendra and Kannan Parathasarathy,
G(s)H(s) = K n K aK g / K e “Identification and control of dynamic system using
neural network”, IEEE Trans. on Neural Networks,
Vol.
(1+T1,fS)(1+T
no.1, , March 1990, . pp.4-27
m s)
where, [10] Zi-Qin Wang, , Michael T. Manry, and Jeffery L.
Ka = net control field amperes per volt actuating error Schiano, “LMS Learning Algorithms: Misconceptions
signal, and New Results on Convergence,” IEEE Trans. on
Neural Network, vol. 11, No. 4, Jan. 2000.
Kg = No-load amplidyne terminal voltage per net [11]Madan M. Gupta and Dandina H. Rao, Neuro-
Control field current, Control Systems, Theory, Principles and Applications,
IEEE Press, May 1994.
Tf /Rf= Lf = Time constant of quadrature field [12] S. Thomas Alexander, Adaptive Signal Processing:
of amplidyne, seconds, Theory and Applications, Springer-Verlag New York,
1986.
Tm = JRa = Time constant of motor and [13]L.Ljung , M.Morf and D.D.Falconer, “Fast
KTKe load, seconds, Calculation of Gain Matrices for Recursive Estimation
Kc = Motor volts per radian per second of motor, Schemes,”Int. J. of Control,Vol.27,pp. 1-17,January
1978.
Kn = Voltage from tachometer per radian per second [14]D.D.Falconer and L.Ljung, “ Application of Fast
of motor. Kalman Estimation to Adaptive Equalization,” IEEE
Trans. On Communications, Vol. COM-26, pp-1439-
For our model, we choose the following values:- 1445, October 1978.
Kn =1; KaKg / Ke =1 ; Tf =0.001 seconds and [15]J.M.Zurada, , St. Paul, MN (Editor): Introduction to
Tm =0.01 seconds Artificial Neural Systems , West Publishing, 1992.
PI Controller parameters: Kpl = 0.056, Ki = -0.002 [16] J.M.Cioffi and T.Kailath, “Fast Recursive Least –
Squares Transversal Filters for Adaptive Filtering,
IEEE Trans. On Acous. Speech and Signal Processing”,
REFERENCES Vol. ASSP-32, pp-304-338, April 1984.
[17]M.L.Honig, “Recursive Fixed Order Covariance
[1]Andersson, P.M., and Fouad,A.A. , Power System Least Squares Algorithms”, Bell Sys. Tech. J.,
Control & Stability, ( IInd Edition, John Wiley & Vol.62,pp-2961-2992, December 1983.
Sons, Inc., 111, River Street, Hoboken, NJ., IEEE [18]O. Adetona, and L.H. Keel, “A New Method for the
Series on Power Engineering ,2003 ). Control of Discrete Nonlinear dynamic Systems Using
[2] Bhatti, T.S., Bansal, R.C. and Kothari, D.P., Neural Networks”, IEEE Trans. on Neural Networks,
Small hydro power plants, Dhanpat Rai & Co., New Vol. 11, no.1, pp.102-112, January 2000.
Delhi ,2004. [19]S.Srivastava, M.S. Singh and A.N. Jha,
[3] Carstens, J.R, Automatic Control Systems & “Identification of Non Linear Systems using FTF and
Components, Prentice –Hall, Inc., New Jersey.U.S.A, -LMS algorithms”, Proceedings of NSC, IIT
1990. Kharagpur, December 2003,INDIA.
[4] Elgerd, O.I. Electric energy systems theory: An [20]S.Srivastava, M.S. Singh , M.Hanmandlu and A.N.
Introduction , Tata McGraw –Hill, New Delhi, 1982. Jha,“Development of a New Controller for NonLinear

42
IJEIT Vol. 1 No. 1 Oct 2009

Systems ,” Communicated to Journal of Soft [34] Tano, H., Sannomiya, Y.,( February 13-22,
Computing in October 2003. 2001),‘Integrated Digital Control panel for
[21] S.Srivastava, Modeling and Control of Non- Hydroelectric Power Stations’, International Course on
Linear Systems Using Intelligent Tools, PhD Thesis, Planning, technology Selection and Implementation of
IIT Delhi, 2005. Small Hydro Power Project, A.H.E.C, IIT, Roorkee,
[22] Ghanayem, O. and Reznik L. “ Excitation India, pp. 461-496.
Control of a Synchronous generator Using an On- [35] Malik, O.P., Hope, G.S., Hancock, G.C., Zhaohui,
Linear Adaptive Fuzzy Logic Controller Structure “, Li., Luqing, Ye., Shouping,WEI.,( September 1991),
proceedings of the 6th IEEE International Conference ‘Frequency Measurement for use with a
on Fuzzy Systems, July 1-5, 1997, Barcelona, Spain, Microprocessor-Based Water Turbine Governor’, IEEE
IEEE neural Networks Council, Vol. 3,pp. 1493- Transactions on Energy Conversion, Vol. 6, No.3, pp.
1498. 361-366.
[23]Bernard Widrow, “30 Years of Adaptive Neural [36] Rakesh Thapar, Michael Ruane, David A.
Networks perceptron, Madaline, and Perreault,( February 13-22, 2001 ), ‘Microprocessor
Backpropagation”, Proceedings of the IEEE, Vol. 78, Controller for a Small Unattended Hydroelectric
no.9, September 1990, pp. 1415-1442. System’, International Course on Planning, technology
[24]Huang,T Hung-Yuan Chung and Jin-Jye Lin, “ A Selection and Implementation of Small Hydro Power
Fuzzy PID Controller Being Like Parameter varying Project, A.H.E.C, IIT, Roorkee, India.
PID”, 1999 IEEE International Fuzzy Systems [37] Djukanovic, M., Novicevic, M., Dobrijevic, D.J.,
Conference proceedings, August 22-25, 1999, Seoul, Babic, B., Dejan J.Sobajic, Yoh-Han Pao, (December
Korea, pp.I-269-276. 1995),‘Neural-Net Based Coordinated Stabilizing
[25]Wu Zhi Qioa, and Masaharu Mizumoto, “PID Control For the Exciter and Governor Loops of Low
type fuzzy controller and parameters adaptive head Hydropower Plants’, IEEE transactions on Energy
method”, Fuzzy Sets and Systems, 78 (1996) 23- Conversion, Vol.10, No.4, pp.760-767.
35.[26] [38]IEEE Guide for control of small hydroelectric
[26] Onur Karasakal, Yesil, Guzelkaya, Eksin, power plants, Std 1020-1988.
“Implementation of a New Self-Tuning Fuzzy PID [39] Henderson, D.S. “Recent Developments of an
Controller on PLC”, Turk J. Elec Engin, Vol. 13, No. Electronic Load Governor for Micro Hydroelectric
2, 2005, pp. 277-286. Generation”, Proceedings HIDROENERGIA 93,
[27]Brown,G.S and Campbell,D.P. “ Principles of Munich, Germany, 1993.
Servomechanism-Dynamics & Synthesis of Closed- [40] Henderson, D.S., 1992, A three phase electronic
Loop Control Systems”, John Wiley & Sons, New load governor for micro hydro generation, Thesis for
York, Chapmen & Hall, Limited,London,1958. the degree of Doctor of Philosophy, The University of
[28]Chestnut,H and Mayer, R.W., “Servomechanism Edinburgh.
& Regulating System Design”, Volume I, John Wiley [41] Henderson, D.S., 1992, “An Advanced Electronic
& Sons,Inc.New York,London, 1959. Load Governor for Control of Micro Hydroelectric
[29]Woodward,J.L. “Hydraulic-Turbine Transfer Power Generation”, IEEE Transactions on Energy
Function for use in Governing Studies”, proc. IEE, Conversion, Vol. 13, No.3, September 1998.
Vol. 115,pp. 424-426,1968. [42]Himani S. An Advanced Control of Micro Hydro
[30]Glattfelder, A.H..,Rettich,J., “Frequency control Power Plants, PhD Thesis, IIT, Delhi, India, February
for low-head hydro units in isolated networks”, 2007.
Water Power & Dam Construction, pp.42-46, March
1988.
[31] Hagihara, S., Yokota H.,& Goda, K., Isobe, K.,(
Nov./Dec, 1979 ), ‘Stability of a hydraulic turbine
Generating Unit Controlled By P.I.D. Governor’,
IEEE Transactions on Power Apparatus and Systems,
Vol. PAS-98, No.6, pp. 2294-2297.
[32] Pereira, L., (November 1981), ‘Induction
generators for small hydro plants’, Water Power &
Dam Construction, pp.30-34.
[33] Philip Schniter, Wozniak, L., (June 1995),
‘Efficiency based optimal control of Kaplan hydro
generators’, IEEE Transactions on Energy
Conversion, Vol.10, No.2, pp.348-353.

43

You might also like