Power System Short-term Reliability Evaluations Based On Cross Entropy Theory

The extensive and deep construction of smart grid all over the world stimulates rapidly evolving technologies congruent with sorts of new demands and environments. A variety of concomitant uncertainties resulting from the new operating environments come to the fore. Reliability evaluation is a well-established theory and powerful to conjugate catastrophes or/and helpful to prevent systems from blackouts. In complex power system short-term reliability evaluations, catastrophes within a short interval ahead are commonly expected to occur with small probabilities. Consequently, classical Monte Carlo methods fail to disseminate in engineering due to their inefficiency to sample rare events. This dissertation, based on cross-entropy theory gaining ground from the information theory, contributes to power system short-term reliability evaluations by proposing three revised versions of classical Monte Carlo methods. The three methods, in allusion to the two typical classical simulation mechanisms labeled as "non-sequential" and "sequential", includes:(1) Non-sequential importance sampling method applicable for discrete multi-state power systems.(2) Three-stage sequential importance sampling method applicable for homogenous Markov systems.(3) Adaptive sequential importance sampling method applicable for non-homogenous Markov systems.The three methods share the same theoretical base that is the importance sampling technique which is a branch of the variance reduction technique, and meanwhile utilize pre-sampling samples to search for the optimal distorted state probabilities used for importance sampling. The premier difference adhered to the three methods resides in the optimization models of distorted parameters related with the considered problems. Through case studies carried out on test systems established by modifying IEEE-RTS79and Roy Billinton Reliability Test System, the accuracy of the proposed methods was proven and high efficiency over classical Monte Carlo methods was suggested.In addition, merits of the proposed methods are explored by exemplifying an application of the non-sequential importance sampling method to well-being evaluation of response capability for wind-integrated generation system. The well-being framework is a useful tool for power system spinning reserve adequacy evaluations, especially in tackling with a system integrated with wind power characterized by the intermittency and fluctuation. In this dissertation, a new architecture with revised well-being indices is proposed to adapt for response capability evaluation of wind-integrated generation systems. In this new architecture, four mutually exclusive states are constructed in replace of the three states in the conventional well-being framework and the associated computation methodologies are given. Case studies are carried out on several test systems modified from the RTS-79, and merits of the proposed architecture are revealed.