1

CONTENTS
1. Abstract.4 2. Introduction..5 2.1 Low Frequency Oscillations In Power Systems..5 2.2 Fixed Parameter Controllers.......7 2.2.1 Conventional Stabilizers.8 2.2.2 Other Fixed Parameter Controllers..9 2.2.3 The Drawbacks Of Conventional Fixed Parameter Controllers10 2.3 Adaptive Controllers.10 2.4 Fuzzy Logic Controllers12 2.5 Artificial Neural Network 13 2.5.1 Artificial Neural Network as a Controller...14 2.5.2 Artificial Neural Network as a Parameter Identifier..15 2.6 Scope of Present Work.16 3. Power System Stabilizers(PSS).17 3.1 Power System Stability17

3.2 Power system stabilizer19 3.3 System investigated..21 3.4 Transfer function model of the power system stabilizer and the Design Considerations.22 3.4.1 Phase lead compensation...23 3.4.2 Stabilizing signal washout..23 3.4.3 Stabilizer gain.23 3.5 Modeling of a single machine system connected to infinite bus System...24 4. Artificial Neural Network(ANN)28 4.1 Neuron Model..29 4.2 Characteristics of artificial neural networks..32 4.3 Learning methods32 4.4 Backpropagation.36 4.4.1Feed forward network.38 4.4.2recurrent network39 4.4.3 Levenberg-Marquardt algorithm..40

4.5 Backpropagation learning algorithm.41 5. Mathematical modeling.45 6. Experimental investigations.49 7. Experimental results.60 8. Discussion of experimental results and experimental investigations..61 9. Conclusions62 10. References...64 11. Appendix..69

ABSTRACT

The problem of dynamic stability of power system has challenged power system engineers since over three decades now. The application of Power System Stabilizer (PSS) can help in damping out low frequency oscillations and improve the system stability. The traditional and till date the most popular solution to this problem is application of conventional power system stabilizer (CPSS). However, continual changes in the operating condition and network parameters result in corresponding change in system dynamics. This constantly changing nature of power system makes the design of CPSS a difficult task.

This project work presents a systematic approach for designing a self-tuning adaptive power system stabilizer (PSS) based on artificial neural network (ANN). An ANN is used for self-tuning the parameters of PSS e.g. stabilizing gain Kstab and time constant (T1) for Lead PSS in real-time. The inputs to the ANN are generator terminal active power (P) and reactive power (Q). Investigations are carried out to assess the dynamic performance of the system with selftuning PSS based on ANN (ST-ANNPSS) over a wide range of loading conditions.

2.INTRODUCTION
2.1 Low Frequency Oscillations in Power System:
Small oscillations in power systems were observed as far back as the early twenties of this century. The oscillations were described as hunting of synchronous machines. In a generator, the electro-mechanical coupling between the rotor and the rest of the system causes it to behave in a manner similar to a spring-mass-damper system which exhibits oscillatory behaviour following any disturbance from the equilibrium state.

Small oscillations were a matter of concern, but for several decades power system engineers remained preoccupied with transient stability. That is the stability of the system following large disturbances. Causes for such disturbances were easily identified and remedial measures were devised. In early sixties, most of the generators were getting interconnected and the automatic voltage regulators (AVRs) were more efficient. With bulk power transfer on long and weak transmission lines and application of high gain, fast acting AVRs, small oscillations of even lower frequencies were observed. These were described as Inter-Tie oscillations. Some times oscillations of the generators within the plant were also observed. These oscillations at slightly higher frequencies were termed as Intra-Plant oscillations.

The combined oscillatory behavior of the system encompassing the three modes of oscillations are popularly called the dynamic stability of the system. In more precise terms it is known as the small signal oscillatory stability of the system.

A power system is said to be small signal stable for a particular steady-state operating condition if, following any small disturbance, it reaches a steady state operating condition which is identical or close to the pre-disturbance operating condition."

The oscillations, which are typically in the frequency range of 0.2 to 3.0 Hz. might be excited by disturbances in the system or, in some cases, might even build up spontaneously. These oscillations limit the power transmission capability of a network and, sometimes, may even cause loss of synchronism and an eventual breakdown of the entire system. In practice, in addition to stability, the system is required to be well damped i.e. the oscillations, when excited, should die down within a reasonable amount of time.

Reduction in power transfer levels and AVR gains does curb the oscillations and is often resorted to during system emergencies. These are however not feasible solutions to the problem. The stability of the system, in principle, can be enhanced substantially by application of some form of close-loop feedback control. Over the years a considerable amount of effort has been extended in laboratory research and on-site studies for designing such controllers.

There are basically three following ways by which the stability of the system can be improved, (1) Using supplementary control signals in the generator excitation system. (2) Making use of fast valving technique in steam turbine. (3) Impedance Control-resistance breaking and application of the FACTS devices, etc.

The problem, when first encountered, was solved by fitting the generators with a feedback controller which sensed the rotor slip or change in terminal power of the generator and fed it back at the AVR reference input with proper phase lead and magnitude so as to generate an additional damping torque on the rotor [1]. This device came to be known as a Power System Stabilizer (PSS).

Damping power oscillations using supplementary controls through turbine, governor loop had limited success. With the advent fast valving technique, there is some renewed interest in this type of control [2].

There can also be other kinds of controls applied to the system for counteracting the oscillatory behaviour - for instance FACTS devices can be fitted with supplementary controllers which improve the system stability.

Power system stabilizers are now routinely used in the industry. However, the complex, constantly changing nature of power systems has severely restricted the efficacy of these devices.

2.2 Fixed Parameter Controllers:

Over the years, a number of techniques have been developed for designing PSSs and other damping controllers [3]. Some of these stabilizing methods have been briefly described in

this section. The main motivation for including this rather brief exposition of the existing techniques is to introduce the need for the application of robust control techniques in power systems. Some of references cited here include a more comprehensive coverage of the topic.

2.2.1 Conventional Stabilizers:

The earlier stabilizer designs were based on concepts derived from classical control theory [4-8]. Many such designs have been physically realized and widely used in actual systems. These controllers feedback suitably phase compensated signals derived from the power, speed and frequency of the operating generator either alone or in various combination as input signals so as to generate an additional rotor torque to damp out the low frequency oscillations. The gain and the required phase lead/lag of the stabilizers are `tuned' by using appropriate mathematical models, supplemented by a good understanding of the system operation.

The principles of operation of this controller are based on the concepts of damping and synchronizing torques within the generator. A comprehensive analysis of this torques has been dealt with by deMello and Concordia in their landmark paper in 1969 [1]. These controllers have been known to work quite well in the field and are extremely simple to implement. However, the tuning of these compensators continues to be a formidable task especially in large multi-machine systems with multiple oscillatory modes. Larsen and Swann, in their three part paper [6], describe in detail the general tuning procedure for this type of stabilizers.

PSS design using this method involves some amount of trial and error and experience on part of the designer. Further these controllers are tuned for particular operating conditions and

with change in operating conditions they require re-tuning. Robustness issues are also not adequately addressed in this classical setting.

2.2.2 Other Fixed Parameter Controllers:

There have also been numerous attempts at applying various other control strategies in particular modal control [9-11] and LQ optimal control [12, 13] techniques for designing damping controllers. These attempts exemplify the growing preference for algorithmic controller design methods as opposed to the classical intuitive ones. They call for a lesser amount of engineering judgment and experience on part of the designer. The ill-suited ness of the quadratic performance index used in LQR/LQG to the problem has motivated researchers to define alternative performance indices which aptly capture the magnitude of system damping [14, 15]. Such indices can be optimized using standard numerical optimization techniques [16].

These techniques have the advantage of being straight forward and algorithmic with little ambiguity in the recommended procedure. A few extensions of these methods tried to incorporate some robustness by optimizing some additional index such as eigen value sensitivities. Sensitivity minimization in this form, though, quite helpful as a means of providing robustness in the absence of better methods is essentially a qualitative approach and hence does not guarantee performance preservation in the face of modal inaccuracies [17].

10

2.2.3 The Drawbacks of Conventional Fixed Parameter Controllers:

The main drawback of the above controllers is their inherent lack of robustness. Power systems continually undergo changes in the load and generation patterns and in the transmission network. This results in an accompanying change in small signal dynamics of the system. The fixed parameter controllers, tuned for a particular operating condition, usually give good performance at that operating condition. Their performance, at other operating conditions, may at best be satisfactory, and may even become inadequate when extreme situations arise. However such stabilizers have been very useful in system that could be represented by single machine infinite bus models. In interconnected multi-machine systems the dynamic instability can manifest itself in the form of poorly damped oscillation of one particular unit with the rest of the system or a group, or a group of machines oscillating against another group of machines. Thus, a generating unit in a multi-machine environment often participates in both `local' and `inter-area' modes of oscillations simultaneously. The spectral and temporal distributions of these modes are largely determined by the rest of the system. As the operating conditions and system configuration are constantly changing in actual power system the performance of the fixed parameter stabilizers can not be always guaranteed.

2.3 Adaptive Controllers:

The problem of changing system dynamics due to changes in the operating conditions can be handled by the application of adaptive control [18, 19]. The power system can be continuously monitored and the controller parameters can be updated in real time to maintain specified performance inspite of changes in the system dynamics. All three standard methods of adaptive control listed below have been tried for designing power system stabilizers.

11

(a) Model reference adaptive control (MRAC) [20, 21] (b) Self tuning control (STC) [22-24] (c) Gain scheduling adaptive control (GSAC) [25]

In MRAC, the desired behavior of the closed loop system is incorporated in a reference model. With the plant and the reference model excited by the same input, the error between the plant output and the reference model output is used to modify the controller parameters, such that the plant is driven to match the behavior of the reference model.

In STC, at every sampling instant, the parameters of an assumed model for the plant are identified using some suitable algorithms, such as Recursive least squares (RLS) or Maximum likelihood estimator etc. The identified parameters are then used in control laws which could be based on various popular techniques such as pole-shifting, pole placement etc.

In GSAC, the gains of the controller are adjusted according to a variety of innovative control strategies depending upon the plant operating conditions and important system parameters. The gains could be computed either off-line or on-line. A few non standard adaptive control schemes have also been reported [26, 27] which do not fit into any of the above categories. These schemes have been shown to work quite well through simulations and laboratory experiments.

12

Adaptive controllers totally avoid the problem of tuning since that is taken care of by the adaptation algorithm. The trade off is the larger on-line computational requirement. The stabilizers are difficult to design and are also susceptible to problems like non-convergence of parameters and numerical instability. Due to these reasons practical implementation of adaptive stabilizers in actual plants has not been popular.

There have been numerous non conventional approaches including feedback linearization, variable structure or sliding mode control and, in more recent times, schemes involving neural networks, fuzzy systems and rule based systems [3] for designing stabilizers. any of these non-conventional approaches have been shown to work quite well in simulated power system models.

Some of the above approaches have also been applied for designing supplementary stabilizing controls for FACTS devices. Most of the modern control theoretical techniques use a black box model for the plant. Hence, identical procedures can be adopted for the design of power system stabilizers and other damping controllers.

2.4 Fuzzy Logic Controllers:

In recent years, Rule based [28, 29], Artificial Neural Network (ANN) based [30, 31] and Fuzzy Logic based (FLC) [32-36] controllers have been suggested for PSS design. These are

13

model free controllers i.e. precise mathematical model of the controlled system is not required. Here control strategy depends upon a set of rules which describe the behavior of the controller.

Here lies, both the strength and weakness of this design philosophy. FLC controllers are well-suited for PSS design as system and its interrelations are not precisely known as they keep constantly changing with changes in both system and operating conditions. However, as the design is rule and experience based, there can not be a unique design procedure.

2.5 Artificial Neural Network:

Two reasons are put forward for using ANN. First, since an ANN is based on parallel processing, it can provide extremely fast processing facility. The second reason for the high level of interest is the ability of ANN to realize complicated nonlinear mapping from the input space to the output space.

ANN has various applications in the power systems [37-56]. There is a overview and general appreciation of the basic concepts of ANNs [37], they also give an insight into how these networks can be employed to solve complex power system problems, particularly those where traditional approaches have difficulty in achieving the desired speed , accuracy and efficiency. But in several applications of power systems ANN focus is major role of voltage stability assessment [38-51] and Power system stabilizer design [52-56].

14

The ANN based PSS proposed in the literature may be classified into the following two categories. (a) In the first category of the ANN based PSS, the ANN is used for real-time tuning of the parameters of the conventional PSS (e.g. proportional and integral gain settings of the PSS [52]). The input vector to the ANN represents the current operating condition, while the output vector comprises the optimum parameters of the conventional PSS.

The ANN-tuned PSS can be regarded as a kind of self-tuning PSS. The main advantage of ANN-tuned PSS over self-tuning PSS is that the conventional self tuned PSS requires system identification which is not the case with ANN-tuned PSS.

(b) In the second category of the ANN based PSS, the ANN is designed to emulate the function of the PSS and directly computes the optimum stabilizing signal It may be noted that the number of training patterns required in the second category is very large, as compared to that in the first category [52, 54]. Moreover, the generation of training patterns in the first category is very straightforward as compared to those in the second category.

2.5.1 Artificial Neural Network as a Controller:

In the second category of the ANN based PSS as mentioned above, the ANN is designed to emulate the function of the PSS and directly computes the optimum stabilizing signal. [53-56]. In this case of ANN based PSS, ANN directly giving stabilizing signal (or control signal) to improve the damping characteristics of the system. So here ANN acts as a "controller". Several papers are presented on "Neural Networks as a controller" [53-56].

15

2.5.2 Artificial Neural Network as a Parameter Identifier:

In the first category of ANN based PSS, parameters of PSS are adapted in real-time i.e. the ANN is used for real-time tuning of the parameters of the conventional PSS. In this category ANN identifying the parameters of PSS. So here ANN acts as a parameter Identifier [52].

The observations of the method [52] are, y y The method merely considers lead controller. The PSS parameters computed by the using of the Eigen values of generator electro mechanical mode are fixed at the locations corresponding to nominal operating point. For any operating point these Eigen values doesnt change. y While in the training process error between desired PSS parameters and obtained PSS parameters is to be about 1*10-5within 1000 iterations.

Advantages of the present method: y PSS parameters computed using the Eigen values of generator electro mechanical mode are going to change depending upon the operating point.

Limitation of suggested method: y While in the training process error between the desired PSS parameters and obtained PSS parameters is comparatively large within same number of epochs .

16

2.6 Scope of Present Work:

The main objectives of the work presented are: 1. To present a systematic approach for designing a multilayer feed forward artificial neural network based self-tuning PSS (ST-ANNPSS). 2. To study the dynamic performance of the system with ST-ANNPSS and compare with that of conventional PSS. 3. To investigate the effect of variation of loading condition on dynamic performance of the system with ST-ANNPSS.

This work presents a systematic approach for designing a self-tuning power system stabilizer (PSS) based on artificial neural network (ANN). An ANN is used for self-tuning the parameters of PSS in real-time. The nodes in the input layer of the ANN receive generator terminal active power P, reactive power Q while the nodes in the output layer provide the optimum PSS parameters, (e.g. stabilizing gain Kstab, time constant (T1 ) in case of lead PSS or conventional PSS .

17

3.Power system stabilizers

3.1Power system stability: Power system stability denotes the ability of an electric power system, for a given initial operating condition, to regain a state of operating equilibrium after being subjected to a physical disturbance, with most system variables bounded so that system integrity is preserved. Integrity of the system is preserved when practically the entire power system remains intact with no tripping of generators or loads, except for those disconnected by isolation of the faulted elements or intentionally tripped to preserve the continuity of operation of the rest of the system. Stability is a condition of equilibrium between opposing forces; instability results when a disturbance leads to a sustained imbalance between the opposing forces. The power system is a highly nonlinear system that operates in a constantly changing environment; loads, generator outputs, topology, and key operating parameters change continually.When subjected to a transient disturbance, the stability of the system depends on the nature of the disturbance as well as the initial operating condition. The disturbance may be small or large. Small disturbances in the form of load changes occur continually, and the system adjusts to the changing conditions. The system must be able to operate satisfactorily under these conditions and successfully meet the load demand. It must also be able to survive numerous disturbances of a severe nature, such as a short-circuit on a transmission line or loss of a large generator. Following a transient disturbance, if the power system is stable, it will reach a new equilibrium state with practically the entire system intact; the actions of automatic controls and possibly

18

human operators will eventually restore the system to normal state. On the other hand, if the system is unstable, it will result in a run-away or run-down situation; for example, a progressive increase in angular separation of generator rotors, or a progressive decrease in bus voltages. An unstable system condition could lead to cascading outages and a shut-down of a major portion of the power system. The response of the power system to a disturbance may involve much of the equipment. For instance, a fault on a critical element followed by its isolation by protective relays will cause variations in power flows, network bus voltages, and machine rotor speeds; the voltage variations will actuate both generator and transmission network voltage regulators; the generator speed variations will actuate prime mover governors; and the voltage and frequency variations will affect the system loads to varying degrees depending on their individual characteristics. Further, devices used to protect individual equipment may respond to variations in system variables and thereby affect the power system performance. A typical modern power system is thus a very high-order multivariable process whose dynamic performance is influenced by a wide array of devices with different response rates and characteristics. Hence, instability in a power system may occur in many different ways depending on the system topology, operating mode, and the form of the disturbance. Traditionally, the stability problem has been one of maintaining synchronous operation. Since power systems rely on synchronous machines for generation of electrical power, a necessary condition for satisfactory system operation is that all synchronous machines remain in synchronism or, colloquially,in step. This aspect of stability is influenced by the dynamics of generator rotor angles and powerangle relationships.

19

Instability may also be encountered without the loss of synchronism. For example, a system consisting of a generator feeding an induction motor can become unstable due to collapse of load voltage. In this instance, it is the stability and control of voltage that is the issue, rather than the maintenance of synchronism. This type of instability can also occur in the case of loads covering an extensive area in a large system. In the event of a significant load = generation mismatch, generator and prime mover controls become important, as well as system controls and special protections. If not properly coordinated, it is possible for the system frequency to become unstable, and generating units and=or loads may ultimately be tripped possibly leading to a system blackout. This is another case where units may remain in synchronism (until tripped by such protections as underfrequency), but the system becomes unstable. Because of the high dimensionality and complexity of stability problems, it is essential to make simplifying assumptions and to analyze specific types of problems using the right degree of detail of system representation [63]

3.2 Power system stabilizer:

The distinction between local modes and inter area modes applies mainly for those systems which can be divided into distinct areas which are separated by long distances. For systems in which the generating stations are distributed uniformly over a geographic area, it could be difficult to distinguish between local and inter area modes from physical considerations. However a common observation is that the inter area modes have the lowest frequency and highest participation from the generators in the system spread over a wide geographic area.

20

The PSS are designed mainly to stabilize local and inter area modes. However, care must be taken to avoid unfavorable interaction with intra-plant modes or introduce new modes which can become unstable. Depending on the system configuration, the objective of PSS can differ. In western USA, PSS are mainly used to damp inter area modes without jeopardizing the stability of local modes. In other systems such as Ontario Hydro, the local modes were the major concern. In general, however, PSS must be designed to damp both types of modes. The procedures for tuning of PSS depends upon the types of application. If the local mode of oscillation is major concern (particularly for the case of a generating station transmitting power over long distances to a load center) the analysis of the problem can be simplified by considering the model of a single machine (the generating station is represented by an equivalent machine) connected to an infinite bus (SMIB). With a simplified machine model and the excitation system, the analysis can be carried out using the block diagram representation. The instability arises due to the negative damping torque caused by fast acting exciter under operating conditions that lead to <0. The objective of PSS is to introduce

additional damping torque without affecting the synchronizing torque [64].

21

3.3 System investigated:

Fig.1 BLOCK DIAGRAM OF A SINGLE MACHINE -INFINITE BUS SYSTEM WITH CONVENTIONAL PSS[65] A single machine-infinite bus (SMIB) system is considered for the present investigations. A machine connected to a large system through a transmission line may be reduced to a SMIB system, by using Thevenin' s equivalent of the transmission network external to the machine. Because of the relative size of the system to which the machine is supplying power, the dynamics associated with machine will cause virtually no change in the voltage and frequency of

22

the Thevenin's voltage (infinite bus voltage). The Thevenin equivalent impedance shall henceforth be referred to as equivalent impedance Conventional PSS comprising cascade connected lead networks with generator angular speed deviation as input signal has been considered shows the small perturbation transfer function block diagram of the SMIB system relating the pertinent variables of electrical torque, speed, angle, terminal voltage, field voltage and flux linkages. This linear model has been developed, by linearizing the nonlinear differential equations around a nominal operating point.

3.4 Transfer function model of the power system stabilizer and the design considerations:
The transfer function of a PSS is represented as:

where KSTAB is stabilizer gain, Tw is washout time constant and

; are time constants of

the lead-lag networks. An optimum stabilizer is obtained by a suitable selection of time constants ; and stabilizer gain .

Fig.2. conventional PSS

23

3.4.1 Phase lead compensation:

To damp rotor oscillations, the PSS must produce a component of electrical torque in phase with the rotor speed deviation. This requires phase-lead circuits to compensate the phase-lag between exciter input (i.e. PSS output) and the resulting electrical torque. The phase characteristic of the system depends on the system parameters and the operating condition. The required phase-lead for a given operating condition and system parameters can be achieved by selecting the appropriate value of time constants.[65]

3.4.2 Stabilizing signal washout:

The signal washout is a high-pass filter that prevents steady changes in the speed from modifying the field voltage. The value of the washout time constant Tw should be high enough to allow signals associated with oscillations in rotor speed to pass unchanged. From the viewpoint of the Washout function, the value of Tw is not critical and may be in the range of 1-20 s. For local mode oscillations in the range of 0.8-2.0 Hz, a washout time constant of about 1.5 s is satisfactory. From the viewpoint of low-frequency inter area oscillations, a washout time constant of 10 s or higher is desirable.[65]

3.4.3 Stabilizer gain: Ideally, the stabilizer gain should be set at a value corresponding to
optimum damping. However, this is often limited by other considerations. It is set to a value, which results in satisfactory damping of the critical modes without compromising the stability of the other modes, and which does not cause excessive amplification of stabilizer input signal noise[.65]

24

3.5 Modeling of a single machine system connected to infinite bus system:

Fig.3 A Single Machine System

consider a single machine system connected to a infinite bus system

Fig.4.Excitation system

25

Torque angle loop: By solving the following torque angle loop and rotor mechanical equations we get the equations for K1 and K2 values







Fig.5.Torque angle loop Representation of flux decay: By solving the flux decay blocks we get the values of K3 and K4

 -

Fig.6.Representation of flux decay

26

Representation of excitation system: By solving the block diagram we can get the values of K5 and K6

 

Fig.7.Excitation system block diagram



The coefficients K1 to K6 that are obtained from the block diagrams are termed as Heffron-Phillips constants. They are dependent on machine parameters and operating conditions Generally K1,K2,K3 and K6 are positive. K4 is also mostly positive except for cases when Re is high. K5 can be either positive or negative. K5 is usually negative for moderate to high external impedances and heavy loadings. For Re=0 the expressions for the constants K1 to K6 are simplified. As the armature resistance is already neglected, this refers to a lossless network on the stator side.

27

The expressions are given below:

It is to be noted that Heffron-Phillips constants can also be defined for any general network connected between the generator and infinite bus[64].

28

4. ARTIFICIAL NEURAL NETWORK

Artificial neural networks can be most adequately characterised as computational models with particular properties such as the ability to adapt or learn, to generalize, or to cluster or organise data, and which operation is based on parallel processing. However, many of the above mentioned properties can be attributed to existing (non-neural) models; the intriguing question is to which extent the neural approach proves to be better suited for certain applications than existing models. An artificial network consists of a pool of simple processing units which communicate by sending signals to each other over a large number of weighted connections. A set of major aspects of a parallel distributed model can be distinguished by Rumelhart and McClelland y y A set of processing units (neurons, cells) A state of activation for every unit, which equivalent to the output of the unit.

Connections between the units. Generally each connection is defined by a weight y y This determines the effect which the signal of unit j has on unit k. A propagation rule, which determines the effective input Inputs. y An activation function effective input y y An external input , which determines the new level of activation based on the of a unit from its external

and the current activation for each unit

A method for information gathering (the learning rule)

29

y y

An environment within which the system must operate, providing input signals and if Necessary error signals.[58]

4.1 Neuron Model:

One neuron cant do much on its own. Usually we will have many neurons labelled by indices k, i, j and activation flows between them via synapses with strengths wki, wij:

Fig.8. neuron model An elementary neuron with R inputs is shown below. Each input is weighted with an appropriate w. The sum of the weighted inputs and the bias forms the input to the transfer function f. Neurons can use any differentiable transfer function f to generate their output.

30

Multilayer networks often use the log-sigmoid transfer function logsig.

The function logsig generates outputs between 0 and 1 as the neuron's net input goes from negative to positive infinity.Alternatively, multilayer networks can use the tan-sigmoid transfer function tansig.

31

Occasionally, the linear transfer function purelin is used in back propagation networks.

If the last layer of a multilayer network has sigmoid neurons, then the outputs of the network are limited to a small range. If linear output neurons are used the network outputs can take on any value.[58]

NUMBER OF LAYERS: In a feed-forward network, the inputs perform no computation and their layer is therefore not counted. Thus a network with one input layer, n hidden layer, and one output layer is referred to as a network with two layers. This convention is widely though not yet universally used.

32

4.2 CHARACTERISTICS OF ARTIFICIAL NEURAL NETWORKS y The Neural Networks exhibit mapping capabilities, that is, they can map input patterns to their associated output patterns. y The Neural Networks learn by examples. Thus, Neural Network architecture can be Trained with known examples of a problem before they are tested for their Inference capability on unknown instances of the problem. They can, therefore, identify new objects previously untrained. y The Neural Networks are robust systems and are fault tolerant. They can, therefore, recall full patterns from incomplete, partial or noisy patterns. y The Neural Networks possess the capability to generalize. Thus, they can predict new outcomes from past trends. y The Neural Networks can process information in parallel, at high speeds, and in a distributed manner.[66]

4.3 LEARNING METHODS:

Learning of Artificial Neural Networks [58] The most significant property of a neural network is that it can learn from environment, and can improve its performance through learning. Learning is a process by which the free parameters of a neural network i.e. synaptic weights and thresholds are adapted through a continuous process of stimulation by the environment in which the network is embedded. The network becomes more knowledgeable about environment after each iteration of learning

33

process. There are three types of learning paradigms namely, supervised learning, reinforced learning and self-organized or unsupervised learning. In supervised learning, an external teacher, having the knowledge of the environment, represents a set of input-output examples for the neural network which may not have any prior knowledge about that environment. When the teacher and the neural network are both exposed to a training vector drawn from environment, by virtue of built-in knowledge, the teacher is able to provide the neural network with a desired response for that training vector. The network adjusts its weights and thresholds until the actual response of the network is very close to the desired response. Figure 2.8 shows a diagram of supervised learning.

Fig.9: Block diagram of supervised learning

In unsupervised or self-organized learning there is no external teacher or critic to oversee the learning process. The network representation and free parameters are optimized to become tuned to statistical regularities of the input data to develop the ability to form internal representation for encoding input data and thereby gather knowledge about the environment.

34

There are many different kinds of learning algorithm for example, error correction learning, Boltzmann learning, Thorndikes law of effect, Hebbian learning and competitive learning. In competitive learning the outputs of a neural network compete among themselves for being the one to be active whereas in Hebbian learning several output neurons may be active at the same time. Some other learning algorithms are: back propagation algorithm, conjugate gradient descent, Quasi-Newton, Levenberg-Marquardt, quick propagation, Delta-bar-Delta, and Kohonen training. Back propagation algorithm is the mostly used algorithm for feed-forward neural network. It is a supervised learning algorithm, which requires a set of training data with known input and output vector. It uses steepest gradient descent of error, which propagates backwards for updating the synaptic weights and thresholds. The advantage of this algorithm is the simplicity of calculation during weight updates. Although widely used, the back propagation algorithm suffers from slow rate of convergence and hence requires long training time for large network with large number of training patterns. However, some methods have been developed to overcome the slow rate of learning, for example, optimization of initial weights [59], adaptation of learning rate using delta-bar-delta learning rule [60], use of multiple activation functions [58]. Also adding a momentum factor, it can learn faster and can overcome local minima [61]. Conjugate gradient descent works by constructing a series of line searches across the error surface. It first works out the direction of steepest descent, just as back propagation would do. However, instead of taking a step proportional to a learning rate, conjugate gradient descent projects a straight line in that direction and then locates a minimum along this line, a process that is quite fast as it only involves searching in one dimension. Subsequently, further line searches are conducted. The directions of the line searches (the conjugate directions) are chosen to try to ensure that the directions that have already been minimized stay minimized. Quasi-Newton is the

35

most popular algorithm in nonlinear optimization, with a reputation for fast convergence. It works by exploiting the observation that, on a quadratic error surface, one can step directly to the minimum using the Newton step - a calculation involving the Hessian matrix. Main drawbacks of this algorithm are that the Hessian matrix is difficult and expensive to calculate and Newton step would be wrong if the error surface is non-quadratic. It requires a huge memory and therefore it is not advised to use it for large networks. Error-Correction Rules [57] In the supervised learning paradigm, the network is given a desired output for each input pattern. During the learning process, the actual output y generated by the network may not equal the desired output d. The basic principle of error-correction learning rules is to use the error signal (d -y) to modify the connection weights to gradually reduce this error. The perceptron learning rule is based on this error-correction principle. A perceptron consists of a single neuron with adjustable weights, w,, j = 1,2, . . . , n, and threshold u, as shown in Figure 2.2. Given an input vector x= (x1, x2,. . . , x j), the net input to the neuron is
n

v ! wjx j  u
j !1

.. 23

The output y of the perceptron is + 1 if v > 0, and 0 otherwise. In a two-class classification problem, the perceptron assigns an input pattern to one class if y = 1, and to the other class if y=0.

36

w
j !1

xj u ! 0
.1

The linear equation defines the decision boundary (a hyper plane in the n-dimensional input space) that halves the space.

Rosenblatt developed a learning procedure to determine the weights and threshold in a perceptron, given a set of training patterns. Note that learning occurs only when the perceptron makes an error. Rosenblatt proved that when training patterns are drawn from two linearly separable classes, the perceptron learning procedure converges after a finite number of iterations. This is the perceptron convergence theorem. In practice, you do not know whether the patterns are linearly separable. Many variations of this learning algorithm have been proposed in the literature. Other activation functions that lead to different learning characteristics can also be used. However, a single-layer perceptron can only separate linearly separable patterns as long as a monotonic activation function is used. The back-propagation learning algorithm is also based on the error-correction principle.

4.4 BACKPROPAGATION:
Backpropagation is the generalization of the Widrow-Hoff learning rule to multiple-layer networks and nonlinear differentiable transfer functions. Input vectors and the corresponding target vectors are used to train a network until it can approximate a function, associate input vectors with specific output vectors, or classify input vectors in an appropriate way. Networks

37

with biases, a sigmoid layer, and a linear output layer are capable of approximating any function with a finite number of discontinuities. Standard backpropagation is a gradient descent algorithm, as is the Widrow-Hoff learning rule, in which the network weights are moved along the negative of the gradient of the performance function. The term backpropagation refers to the manner in which the gradient is computed for nonlinear multilayer networks. There are a number of variations on the basic algorithm that are based on other standard optimization techniques, such as conjugate gradient and Newton methods. Properly trained backpropagation networks tend to give reasonable answers when presented with inputs that they have never seen. Typically, a new input leads to an output similar to the correct output for input vectors used in training that are similar to the new input being presented. This generalization property makes it possible to train a network on a representative set of input/target pairs and get good results without training the network on all possible input/output pairs. There are generally four steps in the training process: 1. Assemble the training data. 2. Create the network object. 3. Train the network. 4. Simulate the network response to new inputs.

38

In back propagation it is important to be able to calculate the derivatives of any transfer functions used. Each of the transfer functions above, logsig, tansig, and purelin, can be called to calculate its own derivative. The three transfer functions described here are the most commonly used transfer functions for back propagation, but other differentiable transfer functions can be created and used with back propagation if desired. 4.4.1 FEED FORWARD NETWORK where the data flow from input to output units is strictly feed-forward. The data processing can extend over multiple (layers of) units, but no feedback connections are present, that is, connections extending from outputs of units to inputs of units in the same layer or previous layers.A single-layer network of S logsig neurons having R inputs is shown below in full detail on the left and with a layer diagram on the right.

Feedforward networks often have one or more hidden layers of sigmoid neurons followed by an output layer of linear neurons. Multiple layers of neurons with nonlinear transfer functions allow

39

the network to learn nonlinear and linear relationships between input and output vectors. The linear output layer lets the network produce values outside the range -1 to +1.

On the other hand, to constrain the outputs of a network (such as between 0 and 1), then the output layer should use a sigmoid transfer function (such as logsig). As noted in Neuron Model and Network Architectures, for multiple-layer networks the number of layers determines the superscript on the weight matrices. The appropriate notation is used in the two-layer tansig/purelin network shown next.

This network can be used as a general function approximator. It can approximate any function with a finite number of discontinuities arbitrarily well, given sufficient neurons in the hidden layer. 4.4.2 RECURRENT NETWORKS that do contain feedback connections. Contrary to feedforward networks, the dynamical properties of the network are important. In some cases, the activation values of the units undergo a relaxation process such that the network will evolve to a

40

stable state in which these activations do not change anymore. In other applications, the change of the activation values of the output neurons are significant such that the dynamical behaviour constitutes the output of the network.

4.4.3 Levenberg-Marquardt algorithm

The Levenberg-Marquardt algorithm was designed to approach second-order training speed without having to compute the Hessian matrix. When the performance function has the form of a sum of squares (as is typical in training feed forward networks), then the Hessian matrix can be approximated as

.2 and the gradient can be computed as

.3 where J is the Jacobian matrix that contains first derivatives of the network errors with respect to the weights and biases, and e is a vector of network errors. The Jacobian matrix can be computed through a standard backpropagation technique that is much less complex than computing the Hessian matrix. The Levenberg-Marquardt algorithm uses this approximation to the Hessian matrix in the following Newton-like update:

41

.4

When the scalar When

is zero, this is just Newton's method, using the approximate Hessian matrix.

is large, this becomes gradient descent with a small step size. Newton's method is faster

and more accurate near an error minimum, so the aim is to shift toward Newton's method as quickly as possible. Thus, is decreased after each successful step (reduction in performance

function) and is increased only when a tentative step would increase the performance function. In this way, the performance function is always reduced at each iteration of the algorithm.

4.5 BACKPROPAGATION LEARNING ALGORITHM

There are many variations of the backpropagation algorithm. The simplest implementation of backpropagation learning updates the network weights and biases in the direction in which the performance function decreases most rapidly, the negative of the gradient. One iteration of this algorithm can be written[66] ..5 where is a vector of current weights and biases, is the current gradient,  is the

learning rate. There are two different ways in which this gradient descent algorithm can be implemented: incremental mode and batch mode. In incremental mode, the gradient is computed

42

and the weights are updated after each input is applied to the network. In batch mode, all the inputs are applied to the network before the weights are updated.[66]

 ALGORITHM: Step 1: Normalise the inputs and outputs with respect to their maximum values. It is proved that the neural networks work better if inputs and outputs lie between 0 and 1. For each training pair,assume there are l inputs given by {I} l*1 and n outputs {O}n*1 in a normalised form. Step 2:Assume the number in the hidden layer to lie between l<m<2l Step 3:[V] represents the weights of synapses connecting input neurons and hidden neurons and [W] represents weights of synapses connecting hidden neurons and output neurons. Initialise the weights to small random values usually from -1 to 1. For general problems, can be assumed as 1 and the threshold values can be taken as zero. = [random weights] = [random weights]  = 6

Step 4: For the training data, present one set of inputs and outputs. Present the pattern to the input layer
 as

inputs to the input layer. By using linear activation function, the output of the

input layer may be evaluated as  7

43

Step 5: Compute the inputs to the hidden layer by multiplying corresponding weights of synapses as  8

Step 6: Let the hidden layer units evaluate the output using the sigmoidal function as



............................9

Step 7: Compute the inputs to the output layer by multiplying corresponding weights of synapses as  ..10

Step 8: Let the output layer units evaluate the output using sigmoidal function as

..11

Step 9: Calculate the error and the difference between the network output and the desired output as for the ith training set as

..12

Step 10: Find {d} as



13

44

Step 11: Find [Y] matrix ase [Y]= Step 12: Find ..15 Step 13: Find {e}= [W] {d}  .16  Find [X] matrix as [X]= Step 14: Find = .17 ..14

Step 15: Find 18 ..19 Step 16: Find error rate as Error rate = ..20

Step 17: Repeat steps 4-16 until the convergence in the error rate is less than the tolerance value.

45

5. MATHEMATICAL MODELLING The system equations are non linear and have to be solved numerically. in solving these equations it is assumed that the system][ is at stable equilibrium point till time t=0 and disturbance occurs at t=0 or later. From the power flow calculations in the steady state,we get the real and reactive power(Pt and Qt),the voltage magnitude (Vt )and angle( ) at the generator terminals. Here is the angle w.r.t to the slack bus[64] Step1 : To find Heffron-Phillips constants :

  If Then

21

22

.23

Else

..24

   

.....25 ....26 27 28

46



...29



30



31



..32



33



.......34



..35

47

Step2 : To find matrices A, B and C.

A= [  (-

0 /(2h) /(2h) )/

wo 0 0 ((-

0 /(2h)) /( )) )/

0 0 (1/ ) -(1/ ) ]

0 (-



 



C=

Step3 : For finding eigen vectors  Selection of

[55]

..36 is to be done should be negative.

For system to be stable the real part of Assume If imaginary part of  

..37

48

Else 
   

..38

Now

+j (imaginary part of

.39

Step4 : To find values

[55]

40



.41

.42

49

6.Experimental investigations
From the mathematical modelling,for different P and Q values Heffron-Phillip constants(K1-K6) are calculated. Using state variable method the values for stabilizer gain (Kstab) and time constant (T1) are calculated . The following are the values obtained shown in table 1

P (p.u) 0.7 1 1 0.8 0.6 0.9

Q (p.u) -0.5 0 0.5 -0.2 -0.3 -0.4

K1 1.0308 1.1060 1.1718 1.0789 1.0701 0.9821

K2 1.3845 1.3288 1.2557 1.3659 1.3349 1.3816

K3 0.3600 0.3600 0.3600 0.3600 0.3600 0.3600

K4 1.7722 1.7009 1.6073 1.7484 1.7087 1.7685

K5 -0.1516 -0.1002 -0.0793 -0.0868 -0.2280 -0.1945

K6 0.1128 0.3608 0.4185 0.2701 0.2770 0.1460

Kstab 18.4242 36.9937 43.4259 27.7016 23.9444 23.5048

T1 0.1318 0.0534 0.0470 0.0667 0.0597 0.1115

Table 1: Pss parameters and Heffron-Phillip constants calculated through off line studies

By substituting the values of (K1-K6) and Kstab and T1 in the following figures and simulation is done and the obtained deviation plots are observed. Figure .10 shows the block diagram of SMIB without PSS and the figure 11. Shows the block diagram of SMIB with PSS and the plot 1, plot 2 show speed deviation waveform and rotor angle deviation waveform without PSS respectively. The plot 3, plot 4 show speed deviation waveform and rotor angle deviation waveform with PSS respectively

50

Figure 10 :block diagram of SMIB without PSS

Plot 1.

51

Plot 2.

Figure 11 :block diagram of SMIB with PSS

52

Plot 3.

Plot 4. Now an artificial neural network is introduced to make a conventional PSS to a self tuning ANNPSS. The figure 12. Shows how an ANN is connected to the PSS

53

Figure 12.

Figure 13. simulated block diagram of ST-ANNPSS

54

The following plots are obtained for different values of P and Q for conventional power system stabilizer(CPSS) and self-tuning power system stabilizer using artificial neural network(STANNPSS)

Plot.5

Plot.6

55

Plot.7

Plot.8

56

Plot.9

Plot.10

57

Plot.11

Plot.12

58

Performance plots of neural network for different number of neurons in hidden layer 30 neurons ; Time:27 secs ; Iterations:1000

Plot .13 40 neurons ; Time:11 secs ; Iterations:272

Plot.14

59

50 neurons ; Time:9 secs ; Iterations :197

Plot.15

60

7.Experiments results

Table.2 operating conditions and optimum PSS parameters computed through off-line studies

Table.3 PSS parameters computed using ANN for 40 neurons and 50 neurons

61

8.Discussions of experimental results and experimental investigations:

y When figure .10 is simulated plots.1 and plot .2 are observed and the speed deviation waveform and rotor angle deviation waveform and it can seen that system is not stable. y When figure 11 is simulated plots 3 and 4 are observed and speed,rotor angle deviation waveform reach steady state, meaning that system is stable with PSS y When figure13 is simulated and is compared with conventional PSS plots 5,6 are obtained for P=0.7,Q=-0.5 and we can observe that the waveform match with each other to great extent. y When P and Q values are changed the usage of ANN makes the system to settle with less number of oscillations compare to CPSS as shown in plots 7,8,9,10,11,12. y Performance of the ANN is observed for different number the ANN is observed for different number of neurons in the hidden layer and performance graphs are observed for 30,40,50 neurons . y y For 30 neurons the performance goal is not reached even for 1000 iterations For 40 neurons the goal is reached for 272 iterations in 11 seconds and the gain values are near to the off line studies y y For 50 neurons the goal is reached in 197 iterations is 9seconds Number of iterations per second for 40 neurons=24.72 And y Number of iterations per second for 50 neurons=21.88

So from above it can be seen that the suitable value for number of neurons is 40.

62

9.CONCLUSIONS
Chapter 3 describes the need of the power system stabilizers and the problem of the low frequency oscillations (0.2 Hz to 3.0 Hz), which limit the power transmission capability of a network and, sometimes, even cause a loss of synchronism and an eventual breakdown of the entire system. The application of Power System Stabilizer (PSS) can help in damping out these oscillations and improve the system stability. The literature survey shows the interest of adaptive PSS and in that way ANN is very helpful for robust design of PSS. In chapter 4, ANN over view , the error-correction rules and Back-propagation algorithm are given which are very helpful for design of ANNPSS . In chapter 6 an artificial neural network has been developed for the tuning of power system stabilizers. The ANN receives generator real power(P) and reactive power(Q), which characterize the loading condition of a generator, as its inputs and provides the desired PSS parameter settings as its output. In the training process, several input-output training patterns are first compiled .These training patterns are used to train the neural network and obtain the connection weights between neurons. Once trained, the ANN is capable of providing the PSS parameters in real-time based on on-line measured system operating point. In chapters 6 and 7, studies shows that dynamic performance with self-tuning PSS is virtually identical to that obtained with conventional PSS at nominal operating point. However, the dynamic performance of self-tuning ANNPSS is quite superior to that of conventional PSS for the loading condition different from the nominal. Investigations also reveal that the performance of self-tuning ANNPSS is quite robust to a wide variation in loading condition. Simulation results for a synchronous generator subject to a severe change in the operating

63

condition indicate that, when the parameter settings are updated in real-time by the ANN, the PSS can offer good dynamic performance over a wide range of operating conditions. On the other hand, a PSS with fixed parameter settings can only provide good damping effect under some particular operating points. Since the proposed ANN approach does not require model identification in deriving the PSS parameters, it is more efficient than the self-tuning controllers and is, therefore, more suitable for real-time applications.

64

10.REFERENCES [1] F. P. Demello and C. Concordia , Concepts of Synchronous Machine Stability as Effected by Excitation Control", IEEE Transactions on Power Apparatus and Systems, Vol. PAS-88, No. 4, pp. 316-329, April 1969. [2] A. Klofenstein, Experience with System Stabilizing Excitation Controls on the Generation of the Southern California Edition Company", IEEE Transactions on PAS, Vol. 90, No. 2, pp. 698-706, March/April 1971. [3] IEEE Working Group Annotated Bibliography on Power System Stability Controls 1986- 1994 ", IEEE Transactions on Power Systems, Vol. 88, No. 4, pp. 794-804, May 1996. [4] K. Bollinger, A. Laha, R. Hamilton and T. Harras, Power System Stabilizer Design Using RootLocus Methods", IEEE Transaction on PAS, Vol. 94, No. 5, pp. 1484-1488, Sep/Oct. 1975. [5] F.P. Demello, P.J. Nolan, T.F. Laskowski, and J.M. Undrill, Coordinated Application of Stabilizers in Multi-machine power systems", IEEE Transactions on PAS, Vol. 99, pp. 892-901, 1980. [6] R.V. Larsen and D.A. Swann, Applying Power System Stabilizers, I, II and III", IEEE Transactions on PAS, Vol. 100, No.6, pp. 3017-3046, June 1981. [7] J.S. Czuba, L.N. Hannett and J.R. Willis, Implementation of Power System Stabilizer at the Ludington Pumped Storage Plant", IEEE Transactions on Power Systems, Vol. PWRS-1, pp. 121128, Feb. 1986. [8] C.T. Tse, S.K. Tso, Approach to the Study of Small-perturbation Stability of Multi- machine Systems ", IEE Proceedings, Generation, Transmission and Distribution, Vol. 135, No. 5, pp. 396405, 1988. [9] C. Chen, Y. Hsu, Coordinated Synthesis of Multi-machine Power System Stabilizers Using an Efficient Decentralized Modal Control Algorithm", IEEE Transactions on Power Systems, Vol. 2, No. 3, pp. 543-551, Aug. 1987. [10] C.T. Tse, S.K. Tso, Refinement of Conventional PSS Design in Multi-machine System by Modal Analysis", IEEE Transactions on Power Systems, Vol. 8, No. 2, pp. 598-605, May 1993. [11] Yao-Nan Yu and Qing-Hua Li, Pole Placement Power System Stabilizers Design of an Unstable Nine-Machine System", IEEE Transactions on PAS, Vol. 5, No. 2, May 1990. [12] Y.N. Yu, K.Vongsuriya, and I.N. Wedman, Application of an Optimal Control Theory to a Power System", IEEE Transactions on PAS, Vol. 89, No. 1, pp 55-61, 1970. [13] Y. Lee and C. Wu , Damping of Power System Oscillations With Output Feedback and Strip Eigenvalue assignment ", IEEE Transactions on Power Systems, Vol. 10, No. 3, pp 1620-1626, Aug 1995.

65

[14] M.R. Khaldi, A.K. Sarkar, K.Y. Lee, and Y.M. Park , The Modal Performance Measure for Parameter Optimization of Power System Stabilizers", IEEE Transactions on Energy Conversion, Vol. 8, No. 4, pp 660-666, Dec 1993. [15] J.B. Simo, I. Kamwa, G. Trudel, and S. Tahan, Validation of a New Modal Performance Measure for Flexible Controllers Design", IEEE Transactions on Power Systems, Vol. 11, No. 2, pp 819826, May 1996. [16] A.J. Urdaneta, N.J. Bacalao, B. Feijoo, L. Flores, and R. Diaz, Tuning of Power System Stabilizers Using Optimization Techniques", IEEE Transactions on Power Systems, Vol. 6, No. 1, pp 127-134, Feb. 1991. [17] Jan Lunze, Robust Multivariable Feedback Control", Prentice Hall International, U.K., 1989. [18] D.A. Pierre, A Perspective on Adaptive Control of Power Systems", IEEE Transactions on PWRS, Vol. 2, No. 2, pp. 387-396, May 1987. [19] A. Ghosh, G. Ledwich, O.P. Malik and G.S. Hope, Power System Stabilizer Based on Adaptive Control Techniques", IEEE Transactions on PAS, Vol. 103, pp. 1983-1989, March/April 1984. [20] O.P. Malik, G.S. Hope and Ramanujan, Real Time Model Reference Adaptive Control of Synchronous Machine Excitation", IEEE PES Winter Meeting Paper, 178, 297-304, 1976. [21] A. Ghandakly and P. Idowu, Design of a Model Reference Adaptive Stabilizer for the Exciter and Governor Loops of Power Generators", IEEE Transactions on PAS, Vol. 5, pp. 887-893, March/April 1990. [22] Shi-Jie Cheng, O.P. Malik and G.S. Hope, Self Tuning Stabilizer for a Multi-machine Power System", IEE Proceedings, Part C, No. 4, pp. 176-185, 1986. [23] A. Chandra, O.P. Malik and G.S. Hope A self-Tuning Controller for the Control of Multi-mahine Power Systems", IEEE Transactions on PAS, Vol. 3, No. 3, pp. 1065-1071, Aug. 1988. [24] N.C. Pahalawaththa, G.S. Hope and O.P. Malik, Multivariable Self-Tuning Power System Stabilizer Simulation and Implementation Studies", IEEE Transactions on Energy Conversion, Vol. EC-6, No. 2, pp. 310-316, June 1991. [25] F. Ghosh and Indraneel Sen, Design and Performance of a Local Gain Scheduling Power System Stabilizer for Interconnected System", TENCON-91, International Conference on EC3-Energy, Computer, Communication and Control Systems, Vol. 1, pp. 355-359, Aug. 1991. [26] D.P. Sen Gupta, N.G. Narhari, I. Boyd and B.W. Hogg, An Adaptive Power System Stabilizer which Cancels the Negative Damping Torque of a Synchronous Generator", Proceedings of IEE, Part. C, pp. 109-117, 1985.

66

[27] G. Bandyopadhyay and S.S. Prabhu, A New Approach to Adaptive Power System Stabilizers ", Electrical Machines and Power System, 14, pp. 111-125, 1988. [28] T. Hiyama, Application of Rule-Based Stabilizing Controller to Electrical Power System", IEE Proceedings, Part C, Vol. 136, No. 3, pp. 175-181, 1989. [29] T. Hiyama, Rule-Based Stabilizer for Multi-Machine Power System", IEEE Transactions on PWRS, Vol. 5, No. 2, pp. 403-411, May 1990. [30] Y. Zhang, G.P. Chen, O.P. Malik, and G.S. Hope, An Artificial Neural Network Based Adaptive Power System Stabilizer", IEEE Transactions on Energy Conversion, Vol. 8, No. 1, pp. 71-77, 1993. [31] Y. Zhang, O.P. Malik, G.S. Hope and G.P. Chen, Application of an Inverse Input-Output Mapped ANN as a PSS", IEEE Transactions on Energy Conversion, Vol. 9, No. 3, pp. 433-441, 1994. [32] T. Hiyama and C.M. Lim, Application of Fuzzy Logic Control Scheme for Stability Enhancement of a Power System", IFAC International Symposium on Power Systems and Power Plant Control, Korea, pp. 313-316, Aug. 1989. [33] Y. Hsu, C. Cheng, Design of Fuzzy Power System Stabilizer for Multi-Machine Power Systems", IEE Proceedings, Part C, Vol. 137, No. 3, pp. 233-238, May 1990. [34] M.A.M. Hasan, O.P. Malik and G.S. Hope, \A fuzzy Logic Based Stabilizer for a Synchronous Machine", IEEE Transactions on Energy Conversion, Vol. 6, No. 3, pp. 407- 413, September 1991. [35] M. Braae and D.A. Rutherford, Selection of Parameters for a Fuzzy Logic Controller", Fuzzy Sets and Systems, Vol. 2, No. 3, pp. 185-199. [36] K.A. El-Metwally and O.P. Mallik, Parameter Tuning for a Fuzzy Logic Controller", Proceedings of the IFAC Twelfth World Congress on Automatic Control, Sydney, Australia, Vol. 2, pp. 581584, July 18-23, 1993. [37] Raj Aggarwal and Yonghua Song," Artificial Neural Networks in Power Systems", Part 1,2,3 Power Engineering Journal 1998. [38] A.A. El-Keib and X. Ma, Application of Artificial Neural Networks in Voltage Stability Assessment, IEEE Transactions on Power Systems, Vol. 10, No. 4, November 1995 [39] V. R. Dinavahi and S. C. Srivastava, ANN Based Voltage Stability Margin Prediction [40] A. Arunagiri B. Venkatesh Stella Morris Artificial Neural Network Approach in determining Voltage Stability in Power system Networks , Proceedings of the 9th International Conference on Neural Information Processing (ICONIP'OZ) , Vol. 5 [41] Saikat Chakrabarti and B. Jeyasurya On-line Voltage Stability Monitoring Using Artificial Neural Network IEEE Transactions on Power Systems, Vol. , No. , 2004

67

[42] Antnio C. Andrade ,F. P. Maciel Barbosa, J. N. Fidalgo and J. Rui Ferreira Voltage stability Assessment Using a New FSQV Method and Artificial Neural Networks IEEE MELECON 2006, May 16-19, Benalmdena (Mlaga), Spain. [43] Antonio C. Andrade, F. P. Maciel Barbosa, FSQV and Artificial Neural Networks to Voltage Stability Assessment. [44] S. Chauhan and M.P. Dave, Sensitivity based voltage instability alleviation using ANN IEEE Transactions on Electrical Power and Energy Systems 25 (2003) 651657 [45] H .P. Schmidt, Application of artificial neural networks to the dynamic analysis of the voltage stability problem, IEE proc.-gener.transm.distrib.Vol.144,No.4 , July 1997 [46] B. Jeyasurya, Artificial Neural Networks For on-line Voltage stability Assessment 0-7803-64201/00/$10.00 (c) 2000 IEEE [47] Y. Zhang and Z. Zhou, Online Voltage Stability Contingency Selection Using Improved RSI Method Based on ANN Solution 0-7803-7322-7/02/$17.00 2002 IEEE [48] Bansilall, D. Thukaram', and K. Harish Kashyap', Artificial Neural Network Application to Power System . Voltage Stability Improvement 0-7803-7651-x/03~17.000 2 003 lEEE [49] S. Kamalasadan, , A. K. Srivastava, and D. Thukaram,Novel Algorithm for Online voltage Stability Assessment Based on Feed Forward Neural Network 1-4244-0493- /06/$20.00 2006 IEEE. [50] Mohamed M. Hamada, Mohamed A. A. Wahab and Nasser G. A. Hemdan, Artificilal Neural Network Modeling Technique for Voltage Assessment of Radial Disribution Systems [51] S.Sahari, A.F.Abidin, and T.K.Abdul Rahman,Development of Artificial Neural network for Voltage Stability Monitoring,National Power and Energy conference(PECon)2003

Proceedings,Bangi, Malaysia. [52] Hsu Y-Y, Chen C-R. "Tuning of power system stabilizers using an artificial neural network". IEEE Trans Energy Conversion 1991 ; 6(4):612-9. [53] Zhang Y, Chen GP, Malik OP, Hope GS. An artificial neural network based adaptive power system stabilizer. IEEE Trans Energy Conversion 1993;8(1):71-7. [54] Zhang Y, Malik OP, Chen GP. Artificial neural network power system stabilizers in multi-machine power system environment. IEEE Trans Energy Conversion 1995;10(1):147-54. [55] Guan L, Cheng S, Zhou R., Artificial neural network power system stabilizer trained with an improved BP algorithm. IEE 143(2):135-141. Proceedings on Generation Transmission Distribution 1996;

68

[56] Park YM, Lee KY, A neural network based power system stabilizer using power flow characteristics. IEEE Trans Energy Conversion 1996;11(2):435-41. [57] Ani1 K. Jain ,Jianchang Mao and K.M. Mohiuddin "Artificial neural networks: A Tutorial"00189162/96/ march 1996 IEEE, pp. 31-44.
nd

[58] Simon Haykin, Neural Network: A Comprehensive Foundation, 2 Edition, IEEE Press, 1999. [59] Jim Y. F. Yam and Tommy W. S. Chow, Feed forward Coefficients Networks Training Speed Enhancement by Optimal Initialization of the Synaptic IEEE Transactions on Neural Networks, Volume 12, No.2, March 2001, pp.430-434. [60] R. Haque, N. Chowdhury, An Artificial Neural Network Based Transmission Loss Allocation For Bilateral Contracts 18th Annual Canadian Conference on Electrical and Computer Engineering CCECE05, May 1-4, 2005, pp. 2197-2201. [61] Phansalkar, V.V., Sastry, P.S., Analysis of the back-propagation algorithm with momentum, IEEE Transactions on Neural Networks, Volume: 5, Issue: 3, May 1994, pp. 505 506. [62] Ping Chang and Jeng-Shong Shih ,"The Application of Back Propagation Neural Network of Multichannel Piezoelectric Quartz Crystal Sensor for Mixed Organic Vapours" Tamkang Journal of Science and Engineering, Vol. 5, No. 4, 2002, pp. 209 217. [63] Kundur, P., Power System Stability and Control, McGraw-Hill, 1994
nd

[64] K.R.Padiyar.,Power system dynamics stability and control., 2 Edition.,BS publications.

[65] Ravi Segala, Avdhesh Sharmab, M.L. Kotharic A self-tuning power system stabilizer
based on artificial neural network Received 14 November 2001; revised 3 October 2003; accepted 7 November 2003 [66] S.Rajasekaran,G.A.Vijayalakshmi Pai,neural networks,fuzzylogic and genetic algorithms synthesis and applications

69

11.APPENDIX Data used for test system

f(Hz) D(pu) h(sec) Ra(pu) xd(pu) xq(pu) Xd'(pu) Xq'(pu) Tdo'(sec) Tqo'(sec)

60

0.0

5.0

0.0

1.6

1.55

0.32

0.5

6.0

0.81

Eb Xe(pu) KA(sec) TA(pu) Tr(sec) T3(sec) Tw(pu) T2 1 0.4 5.0 0.006 0.0 2.1 1.4 0.05

Nomenclature h Eb -- inertia constant -- infinite bus voltage

Tm, Te -- mechanical and electrical torques, respectively Xd -- direct axis reactance

70

Xq Xd' Xq' Xl Ra Efd D Tr Vref

-- quadrature axis reactance -- direct axis transient reactance -- quadrature axis transient reactance -- leakage reactance -- stator resistance per phase -- equivalent exciter voltage -- damping factor -- terminal voltage transducer time constant -- AVR reference signal

KA,TA -- AVR gain and time constant, respectively VS TW Kstab -- stabilizing signal -- washout time constant -- PSS gain

T1,T2 -- PSS time constants Xe -- equivalent resistance and reactance, respectively