BOND SERVOLINE DATA ACQUISITION UNIT

Bond Servoline Data Acquisition Unit

Bond Servoline Data Acquisition Unit

1. Introduction

Electric power systems are increasing in capacity and their distribution over long distances. Therefore, it is becoming difficult to supply high-quality electric power. Over the last decade, much attention has been paid to the transient stability of power systems. Controllers such as AVRs (automatic voltage regulators), PSSs (power system stabilizers), and GOVs (governors) for improving the transient stability of power systems have been equipped with synchronous generators. AVR controls excitation voltage of the synchronous generator to keep constant terminal voltage, PSS is the controller that damps power system oscillations, and GOV controls mechanical input power of the synchronous generator to keep constant angular speed.

When these controllers are designed, considerable investment of time and effort is involved in tuning the controller parameters because the parameters of these controllers are decided by a trial and error procedure. To overcome the disadvantage of trial-and-error tuning of controller parameters, a genetic algorithm has been proposed (Abuta, and Amada, 1992). However, since the parameters of these controllers are optimized by the genetic algorithm, which are fixed, their control performance is dependent on system parameters and operating conditions. In recent years a neural network that can be expressed as a nonlinear function by itself has been studied actively for stabilization controllers [2-5]. However, control performance using a conventional neural network largely depends on training data, since the training of the neural network is carried out before its application, and then the systems are controlled by using the generalization/mapping ability of neural network. Therefore, this method cannot realize effective control for variable system conditions.

In this paper to overcome this problem, we propose a recurrent neural network (RNN) stabilization controller. This method can maintain the transient stability of power systems for variations of system parameters and operating conditions, because the proposed RNN is an online adjustment type. The weights of the proposed recurrent neural network are adjusted online to keep electrical output power deviation to zero. Since this method only connects the RNN in parallel with the conventional controllers, it is easy to apply to the real systems (Hsu and Jeng, 1995). To demonstrate the effectiveness of the proposed method, computer simulations are carried out on a single-machine infinite-bus power system under a three-line to ground fault condition. The simulation results show the controller robustness and its effect on damping of power system oscillations under parameters and system operating condition variation.

2. Control Scheme

2.1. Power System Equations

We consider a single-machine infinite-bus power system. The schematic diagram of the power system is shown in Figure 1. Generated power is transmitted on parallel transmission lines. The synchronous generator is equipped with AVR, PSS, and GOV, block diagrams for which are shown in Figures 2, 3, and 4 respectively.

Fundamental equations of the power system are

where M is the inertia constant of generator, D is the damping constant of generator, d is the phase angle, Pm, Pe are the mechanical input and electrical output power, ?? = ?-?o is the speed ...