Identification of nonlinearity in actual data with the use of simplified test
AbstractThe problem of nonlinearity identification in experimental data is considered with application of appropriate statistical tests. An analysis of known statistical nonlinearity test, which is based on Fisher relation, is presented; and a new simplified test, which can be used in conditions of incomplete experimental or statistical data, is proposed. The empirical statistical nonlinearity criterion is computed on the basis of existence of a link between the values of respective cumulative sum and standard deviation, which is calculated by experimental data. It was empirically established that there exists a close link between the proposed and existing tests in the sense of similarity of final testing results. To find the critical values of statistics that are necessary for the usage of the simplified test, the appropriate computational experiments have been fulfilled. It has also been established that the proposed simplified test can be used successfully in conditions of complete and incomplete experimental data. The practical application of the different tests to actual data proved the similarity of results obtained with various approaches.
Zaks B. Statisticheskoye otsenivaniye. — M.: Statistika, 1976. — 598 s.
Kuz’min V.N. Novyy statisticheskiy metod identifikatsii nelineynosti na osnove empiricheskikh dannykh // Konferentsiya po komp’yuternomu analizu dannykh i modelirovanii. — Minsk, 1998. — T. 1. — S. 159–164.
Bidyuk P.I., Baklan I.V., Gasanov А.S. Sistemnyy podkhod k analizu adekvatnosti nelineynykh modeley vremennykh ryadov // Kibernetika i sistemnyy analiz. — 2003. — № 3. — S. 147–158.
Babak O.V., Gasanov А.S., Mel’nikov S.V. Primeneniye metoda general’noy obobshchennoy peremennoy pri identifikatsii nelineynykh ob"yektov // Visn. KNU ym. T.H. Shevchenko. Ser. fiz.-mat. nauk. — 2002. — № 5. — S. 9–12.
Rao M.J.N. Filtering and Control of Macroeconomic Systems. — Amsterdam: North Holland, 1987. — 360 p.