A knowledge based hybrid approach to an expert system for power system control

Many of the engineering problems require a knowledge based hybrid approach. The term hybrid here indicates the approach in which the knowledge based symbolic processing programs are supported by the algorithmic analysis programs. The knowledge based hybrid approach is particularly necessary in the case of power system control problems where the expert knowledge needs to be supported by analyses such cis the power flow analysis and the sensitivity analysis.

In narrow problem domains, expert systems can provide high performance, equalling or even exceeding those of individual human experts. This thesis describes a knowledge based hybrid approach in building an expert system, to bring a power system from an alert or an emergency state to a normal or the best possible state under given constraints.

A prototype of the expert system with machine learning capability has also been implemented on Symbolics 3670 Lisp Machine, which demonstrates a promising performance. The prototype is referred as EPSY, abbreviation for ''Expert System for Power System Control," throughout in this thesis. The power system configuration chosen for implementaion is popularly known as "New England 39 Bus System." It is demonstrated that by integrating symbolic processing program with the power system analysis programs, the expert system exhibits a superior performance than can be obtained by using just the symbolic processing programs. Two power system analysis programs, the power flow analysis and the sensitivity analysis programs have been used. The inference and control for the problem solving system and the machine learning subsystem of this expert system uses the blackboard architecture, which has been found particularly suitable for our problem.

An important feature of this expert system is that while attempting to correct bus voltage violations, the expert system learns from its past failures during its problem solving process. An attempt to correct a bus voltage violation that fails will help the expert system to learn the reasons that lead to the failure, so that the expert system can correct its strategy in its next attempt.

All aspects of such an expert system have been covered with particular emphasis on the machine learning. Finally the recommended future research and the scope of such a design approach have been mentioned.