Reducing computational costs for reliability indicators of electric power generating systems

In this paper, a modelling of load curves through mixtures of normal distributions is presented. We also include a procedure for reducing computation times when obtaining power system reliability indicators. Some results are shown for the Spanish power generating system where it is clear that the approximation of fortnight demand curves by a mixture of two normals leads to values of the LOLE index similar to those obtained through the original demand curves. The computation time reduction procedure we propose replaces a more conventional algorithm when using Esscherʹs large deviation technique. It is based on the application of random field models to the prediction of deterministic functions and it leads to a reduction of about 75% in the total computing time for the LOLE index with no significant loss of precision. We also present a comparison of this approximation with a direct convolution and with the application of the fast Fourier transform.