Study on Fast and Reliable Power System Operation Control Using Parallel and Cooperative Distributed Problem Solving

点击翻译按钮获取中文摘要

This research aims at reducing the inference time and making the solution more accurate and more reliable in power system control problem by using parallel decentralized processing techniques in knowledge engineering. The following results are obtained from this study :(a) A new type of overload refief control system is developed on a McIntosh computer using object-oriented knowledge representation. Performances of the proposed system taking into account parallel processing are tested by using centralized, distributed and hierarchical problem solving techniques respectively. Here, load flow calculation is also excuted in parallel, which leads to a large reduction of the total processing time required for problem solving in a large-scale power system. It is made clear from the comparison of dwe problem solving techniques that there exists the optimal number of subsystems into which the whole system is divided and that hierarchical problem solving system is superior to distributed proble … More m solving system. In addition, it should be noted that the communication time among local processors belonging to subsytems plays an important role in determining the decomposition pattern of the large-scale system.(b) An assisting system for fault section estimation is developed on a McIntosh computer using Common Lisp language. The proposed assisting system uses cause-effect network which is developed here and operating time information of actuated relays and tripped circuit breakers for inference. Taking into account parallel processing, the proposed system makes the very rapid and correct fault section estimation possible compared with the conventional rule-based production system. Furthermore, it makes the proposed system robust to taking necessary countermeasures against erroneous or missing information of relays and circuit breakers. In addition, the use of Petri-net instead of cause-effect network is also investigated to improve the parallelism of inference and good results are obtained. Less