LSPEST—Subroutine Package for Identification and Parameter Estimation of Linear Dynamical Models

Abstract

In this paper a Fortran-IV subroutine package and programme system for identification and parameter estimation of linear dynamical models is disscused. Both continuous and discrete models are considered and the programmes are divided into two groups, according to the type of the model. Least squares method is the theoretical basis of the programme system. For parameter estimation of linear discrete models one of the following methods can be used: ordinary least squares method, generalized least squares, instrumental variables, least squares with correction and correlation method with least squares, which provide unbiased estimates of the parameters. The programme system provides determination of the model structure, choosing the optimal one at which the 108S function is minimized. Continuous model parameter estimation is based on a modification of the generalized functions method. Users can include programmes based on their own modification of the least squares method and realize one of the following operations: data preprocessing and statistical characteristics calculations, printing and plotting all intermediate and final results, calculating impulse and step response of the optimal model, diagnostic test of the optimal model for adequacy. LSPEST consists of more than 50 modules and is organized in a way similar to the IBM scientific subroutine package and can be used on computers such as IBM 360/370 and ES, operating under DOS or OS. The realized, programme system unites all the modules by using overlay structure, which provide better use of the computer and helps the user in realization of various binds of computing schemes.