Towards Estimation in the Sense of the Least Sum of Absolute Errors
暂无分享,去创建一个
Abstract This paper presents an attempt to introduce an estimation algorithm that can lead to estimates of coefficients of a parametric model, close to the values detennined by the Least Sum of Absolute errors of the model output. Examples of LSA-estimation application presented here show specific and interesting features. This estimation is quite insensitive to instantaneous, even very powerful and non-symmetric disturbances. A theorem for the derivation the LSA-estimation algorithm, as a special modification of the weighted Least Sum of Square estimation is presented. Application of the LSA-algorithm and a comparison with LSS-estimation results is presented on examples of static and dynamic models of linear plants.
[1] R. Fisher,et al. On the Mathematical Foundations of Theoretical Statistics , 1922 .
[2] M. S. Bartlett,et al. Some probability limit theorems with statistical applications , 1953, Mathematical Proceedings of the Cambridge Philosophical Society.
[3] Petre Stoica,et al. On the convergence of pseudo-linear regression algorithms , 1985 .