ANALYSIS AND DATA PROCESSING SYSTEMS
Print ISSN: 2782-2001
Online ISSN: 2782-215X
To solve the problem of effective identification of regression models of multifactor systems, as a rule, they resort to using the concept of optimal experiment design. The synthesis of experimental designs involves the use of an a priori chosen optimality criterion. Quite a few criteria have been proposed. Most often, criteria are used that are associated with the accuracy of estimating the parameters of regression models. We can name such well-known criteria as: the D-optimality criterion, the A-optimality, criterion and the E-optimality criterion. It should be noted that most of the theoretical and applied research is associated with the use of the D-optimality criterion. It is noted in the paper that often plans built according to the A-optimality criterion show good performance for a number of other optimality criteria. At the same time, the criterion itself characterizes an average variance of estimates of the parameters of the regression model and, for A-optimal designs the dispersion ellipsoid has the smallest overall dimensions. The use of the D-optimality criterion makes it possible to obtain an ellipsoid of dispersion of parameter estimates of the smallest volume, which does not exclude the possibility of obtaining an ellipsoid elongated along one or more principal axes. The paper proposes and describes two algorithms for the synthesis of discrete A-optimal designs. The first of them is based on the concept of the consistent completion of the experiment design to the required volume developed by the author. It can be successfully used in a situation where the researcher needs to increase the number of experiments to achieve the required accuracy of the resulting model. The second algorithm, which makes it possible to build plans for a given number of observations, consists of iterations in which points are added and removed from the plan according to certain rules.
Popov Alexander A.
1. Nalimov V.V., Golikova T.I. Logicheskie osnovaniya planirovaniya eksperimenta [Logical foundations for design an experiment]. Moscow, Metallurgiya Publ., 1981. 151 p.
2. Golikova T.I., Panchenko L.A., Fridman M.Z. Katalog planov vtorogo poryadka. Ch. 2 [Catalog of plans of the second order. Pt. 2]. Moscow, MSU Publ., 1974. 384 p.
3. Golikova T.I., Panchenko L.A. Sistematizatsiya planov dlya otsenki polinomial'nykh modelei vtorogo poryadka [Systematization of designs for evaluating second order polynomial models]. Planirovanie optimal'nykh eksperimentov [Design of optimal experiments]. Moscow, MSU Publ., 1975, pp. 106–149.
4. Kiefer J. Optimal design: Variation in structure and performance under change of criterion. Biomertrica, 1975, vol. 62, no. 2, pp. 277–288.
5. Dubova I.S. Fedorov V.V. Tablitsy optimal'nykh planov II (Nasyshchennye D-optimal'nye plany na kube) [Tables of Optimal Designs II (Saturated D-optimal designs on a cube)]. Moscow, MSU Publ., 1972. 42 p.
6. Fedorov V.V. Teoriya optimal'nogo planirovaniya eksperimenta [Theory of optimal experiment design]. Moscow, Nauka Publ., 1971. 312 p.
7. Denisov V.I., Popov A.A. A-, E-optimal'nye i ortogonal'nye plany regressionnykh eksperimentov dlya polinomial'nykh modelei [A-, E-optimal and orthogonal regression experiment designs for polynomial models]. Akademiya nauk SSSR, Nauchnyi sovet po kompleksnoi probleme “Kibernetika”. Moscow, 1976. 44 p.
8. Popov A.A. Optimal'noe planirovanie eksperimenta v zadachakh strukturnoi i parametricheskoi identifikatsii modelei mnogofaktornykh sistem [The optimal experiment planning in problems of structural and parametric identification of multifactor systems models: a monograph]. Novosibirsk, NSTU Publ., 2013. 296 p.
9. Wilkinson J.H. Algebraicheskaya problema sobstvennykh znachenii [Algebraic eigenvalue problem]. Moscow, Nauka Publ., 1970. 564 p.
10. Popov A.A. Posledovatel'nye skhemy postroeniya optimal'nykh planov eksperimenta [The sequential schemes constructing of the optimal experiment designs]. Sbornik nauchnykh trudov Novosibirskogo gosudarstvennogo tekhnicheskogo universiteta = Transaction of scientific papers of the Novosibirsk state technical university, 1995, no. 1, pp. 39–44.
11. Popov A.A. Posledovatel'nye skhemy sinteza optimal'nykh planov eksperimenta [The sequential schemes of synthesis of the optimum plans of experiment]. Doklady Akademii nauk vysshei shkoly Rossiiskoi Federatsii = Proceedings of the Russian higher school Academy of sciences, 2008, no. 1 (10), pp. 45–55.
12. Bellman R. Introduction to matrix analysis. New York, 1960 (Russ. ed.: Bellman R. Vvedenie v teoriyu matrits. Moscow, Nauka Publ., 1969. 367 p.).
13. Rao C.R. Linear statistical inference and its applications. New York, Wiley, 1965 (Russ. ed.: Rao S.R. Lineinye statisticheskie metody i ikh primeneniya. Moscow, Nauka Publ., 1968. 548 p.).
14. Searle S.R. Matrix algebra useful for statistics. 2nd ed. Hoboken, NJ, Wiley, 2017. 479 p.
15. Mitchell T.J. An algorithm for the construction of D-optimal experimental designs. Technometrics, 1974, vol. 16, no. 2, pp. 203–210.
Popov A.A. Algoritmy postroeniya diskretnykh A-optimal'nykh planov eksperimentov pri aktivnoi identifikatsii regressionnykh modelei mnogofaktornykh sistem [Algorithms for constructing discrete A-optimal experiment designs in active identification of regression models of multifactor systems]. Sistemy analiza i obrabotki dannykh = Analysis and Data Processing Systems, 2022, no. 2 (86), pp. 39–54. DOI: 10.17212/2782-2001-2022-2-39-54.
