ABSTRACT

The Moore-Penrose algorithm provides a powerful method for fitting data to models of known linear terms with unknown coefficients. In practical applications, however, experimental data rarely fit theoretical models perfectly. The errors between the model and the data become very important, especially if the results of the model are used in further calculations (in which case, error propagation occurs). This presentation discusses the Moore-Penrose least squares method, outlines a geometrical interpretation of its proof, and presents MATLAB results of errors in polynomial models with varying degrees of Gaussian distributed “noise.”