Avaliação da confiabilidade preventiva de sistemas elétricos de grande porte utilizando redes neurais

Abstract

The well-being analysis was recently proposed as a new framework to measure the degree of adequacy of power systems, which has as the main objective the incorporation of deterministic criteria into the reliability analysis process. The conceptual basis for this framework is obtained through the classification of the system states into three groups: healthy, marginal and at risk. For the identification of these operation states, the system is submitted to a deterministic criterion. In composite generation and transmission systems, the identification of a healthy or marginal state becomes much more complex than that one used, for example, in generation systems. Any deterministic criterion to be used must consider a list of contingencies. In principle, for each considered operation state, it is necessary to carry out a number of additional performance analyses equal to the number of elements in the list. Moreover, these adequacy analyses involve load flow runs with corrective measure optimizations. Therefore, the major difficulty found in assessing well-being indices consists of conciliating the deterministic criterion and the combinatorial nature of the problem. In this thesis, the evaluation of well-being indices for bulk composite generation and transmission power systems is focused. For this purpose, the following techniques are considered: non-sequential Monte Carlo simulation with a new test function; interior point method for optimal power flow with reduced constraints; probabilistic equivalent network; and incorporation of artificial neural nets in the classification of the operation states. These techniques can provide considerable reductions in the computational cost demanded during the classification of states. In order to verify the proposed concepts and models, the developed methodology is applied to several test systems, including a configuration of the Brazilian power network.