2023 № 1 Системний аналіз, управління та інформаційні технології
Постійне посилання колекціїhttps://repository.kpi.kharkov.ua/handle/KhPI-Press/67240
Переглянути
1 результатів
Результати пошуку
Документ Efficiency substantiation for a synthetical method of constructing a multivariate polynomial regression given by a redundant representation(Національний технічний університет "Харківський політехнічний інститут", 2023) Pavlov, Alexander Anatolievich; Holovchenko, Maxim Nikolaevich; Drozd, Valeriia ValeriivnaIn recent years, the authors in their publications have developed two different approaches to the construction of a multivariate polynomial (in particular, linear) regressions given by a redundant representation. The first approach allowed us to reduce estimation of coefficients for nonlinear terms of a multivariate polynomial regression to construction of a sequence of univariate polynomial regressions and solution of corresponding nondegenerate systems of linear equations. The second approach was implemented using an example of a multivariate linear regression given by a redundant representation and led to the creation of a method the authors called a modified group method of data handling (GMDH), as it is a modification of the well-known heuristic self-organization method of GMDH (the author of GMDH is an Academician of the National Academy of Sciences of Ukraine O. G. Ivakhnenko). The modification takes into account that giving a multivariate linear regression by redundant representation allows for construction of a set of partial representations, one of which has the structure of the desired regression, to use not a multilevel selection algorithm, but an efficient algorithm for splitting the coefficients of the multivariate linear regression into two classes. As in the classic GMDH, the solution is found using a test sequence of data. This method is easily extended to the case of a multivariate polynomial regression since the unknown coefficients appear in the multivariate polynomial regression in a linear way. Each of the two approaches has its advantages and disadvantages. The obvious next step is to combine both approaches into one. This has led to the creation of a synthetic method that implements the advantages of both approaches, partially compensating for their disadvantages. This paper presents the aggregated algorithmic structure of the synthetic method, the theoretical properties of partial cases and, as a result, the justification of its overall efficiency.