ARTIFICIAL INTELLIGENCE TECHNIQUES IN POWER SYSTEMS : DEVELOPMENT OF AN OPERATIONAL GUIDANCE EXPERT SYSTEM

Abstract

The growing complexity and sophistication of power systems has necessitated the development of intelligent tools to aid various planning and control operations. This thesis investigates the application of fuzzy logic and neural network techniques for demand forecasting and generation scheduling in an interconnected power system. The global objective is the development of intelligent strategies to achieve the required level of accuracy and optimality in the presence of uncertainty, multiple objectives and complex system interactions. Short term prediction of future load demand is crucial for various operation and planning activities in power systems. Improvement in accuracy of forecasts in these highly capital intensive systems results in substantial savings in operating cost, and increased reliability of power supply. A hybrid architecture has been developed for realizing cascadable fuzzy logic and neural network modules which are used as building blocks for constructing a load forecasting system. This hybrid approach aims to combine the good features of both to overcome the limitations of each. The strengths of this powerful technique lie in its ability to forecast accurately on weekdays, as well as, on weekends and public holidays. Furthermore, use of fuzzy logic effectively handles the load variations due to special events. It has been extensively tested on actual data obtained from the Public Utilities Board of Singapore for 24-hour ahead prediction based on forecast weather information. Very impressive results, with an average error of 0.6% on weekdays, 0.8% on Saturdays and 1.1% on SUndays and public holidays have been obtained. This approach avoids complex mathematical calculations and training on many years of data, and is simple to implement on a personal computer. Based on the advanced knowledge of load demand, optimum unit commitment and dispatch schedules are obtained using a fuzzy expert system. This unique comprehensive approach to generation scheduling recognizes the existence of several objectives, namely, economy, reliability, security, and emission, all of which are mutually exclusive. In this approach, the main safeguard against the incidence of shortages from reliability point of view is the provision of sufficient redundancy and slack capacity throughout the system to meet unexpected contingencies. The security of transmission systems has been ensured by having adequate thermal margins, and by taking preventive measures to avoid steady-state and dynamic unstability. Although it is not possible to identify low-probability high-risk events a priori, it is critical that generation plans be developed for coping with undesired changes. Finally, the adverse effects of electricity generation on the environment are kept to a minimum by favoring the operation of units that emit less pollutants. Since the objectives are in conflict with one another, it is necessary to accept certain trade-offs between them. An optimization framework suitable for rational decision making in the presence of these multiple conflicting objectives has been developed. The fuzzy multi-objective optimization technique described in this thesis has both the analytical rigor and inherent flexibility to prepare generation schedules that satisfy these objectives. This approach provides an explicit framework for analyzing system costs and including various constraints, and permits an efficient and fair allocation of the demand on various generating units. A pattern recognition technique is used for assessing the stability of the interconnected system at each load level. and for evaluating the security transfer limits between neighboring systems. Simulations carried out on a moderately sized interconnected system, containing 19 generating unit,; in the local system, show that this technique is effective in determining the optimal dispatch of generation units, ensuring an appropriate balance between generation and predicted load demand and at the same time improving the stability of the interconnected system. Comparison of results with a classical generation scheduling method shows that use of the proposed fuzzy approach, which includes a larger number of constraints and objectives, results in a much smaller computation time.