Adaptive Critic Based Neurocontrollers for Turbogenerators in a Multimachine Power System

(School of Electrical and Electronic Engineering, University of Natal Durban South Africa, completed in Decemeber 2001)

by

Ganesh Kumar Venayagamoorthy

Abstract

Power systems containing turbogenerators are large-scale nonlinear systems.  The conventional excitation and turbine controllers for the generators are designed by linear control theory based on a single-machine infinite bus (SMIB) power system model.  These SMIB power system mathematical models are linearized at specific operating points and then used to design excitation and turbine governor controllers, and power system stabilizers. The system parameters change with loading in a complex manner, resulting in different behavior at different operating points and the conventional controllers which stabilize the system under specific operating conditions, may no longer yield satisfactory results when there is a drastic change in the power system operating conditions and configurations.  Conservative designs are therefore traditionally used, particularly in multimachine systems, to attempt satisfactory control over the entire operating range of the power system.

However, nonlinear control theory can be applied to power system control, to improve system transient stability.  Instead of using an approximate linear model, as in the design of the conventional controllers, nonlinear models are used and nonlinear feedback linearization techniques are employed, thereby alleviating the operating point dependent nature of the linear designs.  However, nonlinear controllers have a more complicated structure and are difficult to implement compared to linear controllers.  In addition, feedback linearization methods require exact system parameters to cancel the inherent system nonlinearities, and this contributes further to the complexity of the stability analysis. However, the use of Artificial Neural Networks as identifiers and controllers offer a possibility to overcome this problem.

Adaptive critic designs (ACDs) are neural network designs capable of optimization over time, under conditions of noise and uncertainty. This family of ACDs bring new optimization techniques which combine concepts of reinforcement learning and approximate dynamic programming, thus making them powerful tools.  For a given series of control actions that must be taken sequentially, and not knowing the effect of these actions until the end of the sequence, it is impossible to design an optimal controller using the traditional supervised learning neural network.  The adaptive critic method determines optimal control laws for a system by successively adapting two ANNs, namely an action neural network (which dispenses the control signals) and a critic neural network (which ‘learns’ the desired performance index for some function associated with the performance index).  These two neural networks approximate the Hamilton-Jacobi-Bellman equation associated with optimal control theory.  The adaptation process starts with a non-optimal, arbitrarily chosen, control by the action network; the critic network then guides the action network towards the optimal solution at each successive adaptation.  During the adaptations, neither of the networks need any ‘information’ of an optimal trajectory, only the desired cost needs to be known.  Furthermore, this method determines optimal control policy for the entire range of initial conditions and needs no external training, unlike other neurocontrollers.

In this thesis, neural networks are used to identify turbogenerators in a multimachine power system and are then used in the design of neurocontrollers based on Adaptive Critic Designs both in simulation, and are implemented as well on a physical laboratory model.  The laboratory model is a three machine power system consisting of two micro-alternators and the infinite bus as the third machine. These micro-alternators have parameters equivalent to those of 1000 MW turbogenerators.

The design and practical laboratory implementation of nonlinear excitation and turbine neurocontrollers based on Dual Heuristic Programming (DHP) theory (a member of the adaptive critics family) for turbogenerators in a multimachine power system, to replace the conventional Automatic Voltage Regulators (AVRs) and turbine governors, is presented in this thesis.  With DHP, optimal neurocontrollers can be designed offline, avoiding the computational load of online learning and the issues of instability. The DHP excitation and turbine neurocontrollers is implemented on the two micro-alternators. The simulation and practical results show that both voltage regulation and system stability enhancement can be achieved with these proposed neurocontrollers, regardless of the changes in the system operating conditions and configurations.