Modelling and identification of electrical load systems

Abstract

IV ABSTRACT In this thesis, as a model for the hourly elec trical load in Istanbul, a vector-valued autoregressive time series with 24 state variables and with meteorolo gical inputs is constructed. The model is parameterized using a priori statistical information and physical con siderations. Its extensions for short-term load fore casting are indicated. Parameters of the model are estimated via the method of maximum likelihood and the Daviden-Flecher-PowelT minimization algorithm. Estimated -V-3-lues _of the parameiers are- tabulated with their _s tan - dard errors. Behavior of the residuals is presented as a demonstration for the validation of the model. A method for studying time variations in the parameters is suggested. Generalities and remarks concerning the identification problem are extrapolated beyond this specific example.