CONTRIBUTIONS TO APPROXIMATE COMPUTATION OF POWER GENERATING SYSTEM RELIABILITY INDEXES (ENERGY, POWER ENGINEERING, ELECTRIC SYSTEM).

摘要

The purpose of the present work is to investigate several extensions of Esscher's large deviation method, especially to the problem of determining the reliability of interconnected systems where the two system loads are correlated.; Several alternative algorithms obtained from the simple extensions using saddlepoint approximation turn out to be as effective as the original large deviation method in evaluating the system reliability and production costs. With the use of the first-order of the bivariate tetrachoric series expansions for the bivariate normal distribution, the large deviation method is extended to approximate the loss-of-load probability indexes for interconnected systems. The numerical results indicate the accuracy of the bivariate version of the large deviation technique in providing accurate estimates for the generation reliability of interconnected systems.; Another topic under investigation is the development of the large deviation method in production costing context where multiple-block dispatching is considered. In this situation, different "blocks" of a given unit are not statistically independent. The previously given algorithm is therefore modified to account for this dependence. A numerical example is given and comparison of the large deviation results with benchmark values in this case indicates that the large deviation approach to the multiple-block dispatching also provides accurate estimates for the production costing indexes.; The present work also examines an enchancement to the computational performance of the large deviation method in the production costing framework. The enchancement uses a mixture of normal distribution to approximate the load duration curve and, when applied with the large deviation method, the modified scheme results in an approximation that is computationally more efficient than the present large deviation method.; Numerical results comparing the large deviation results with the benchmark values are provided for the topics under investigation. It is found that the bivariate large deviation approximation is robust and effective and it represents a viable computational procedure for determining the generation reliability functions in the two-area problem. The large deviation approach to production costing problem is also found to be quite effective when a mixture of normal approximation is used as well as for the multiple-block dispatching situation. (Abstract shortened with permission of author.)