Modelagem analítica da operação cronológica de reservatórios no despacho probabilístico de geração

Abstract

Probabilistic production costing models evaluate the expected unit operating costs of dispatching generation resources to meet load requirements along the planning period. These models should describe in detail the actual operation of the generation system and should estimate costs under plausible alternative scenarios. The objective of evaluating the variables of the problem under alternative scenarios is to take into account uncertanties of the generation system, as equipment outages, load variation, inflow sequences, etc. The probabilistic production costing problem can be efficiently solved by a method proposed by Baleriaux. The Baleriaux scheme decomposes the production costing problem into generation reliability evaluation subproblems. In turn, the reliability problems reduce to the comparison between the distribution of available capacity and system demand, which is carried out by discrete convolution. One major limitation of the Baleriaux approach is related to the loss of chronological information. In order to carry out the analytical convolutions, it is assumed that the system states are independent of each other, i.e. system operation is represented as “snapshots". As a consequence, the operation of hydroelectric units, which is essentially chronological, has to be simplified. While this representation may be reasonably adequate for thermal-dominated systems, it is less acceptable for systems with a large component of hydro generation. The representation of hydro units in the Baleriaux scheme has been a topic of great interest. However, the approximations proposed until now has led to underestimated operating costs. One alternative is to use the Monte Carlo simulation method, which allows a better representation of hydrothermal systems. Unfortenately, the associated computacional effort may be very high. This thesis describes an extension of the Baleriaux methodology, able to handle the chronological aspects of reservoir operation. It is shown that Baleriaux's idea can be generalized for stochastic problems which can be represented as linear network flows. In this way, the problem is decomposed into a sequence of analytical convolutions of reservoir storage levels, demand, inflows and generation capacity at each stage. The extended Baleriaux scheme can also produce marginal cost information. The application ofthe methodology is illustrated in case studies with utility-derived systems. To verify the accuracy and efficiency of the proposed scheme, the sarne study was repeated with a Monte-Carlo simulation algorithm. Results were identical, but running times with the Monte-Carlo algorithm were higher by more than arder of magnitude.