Neural modeling often doesn't guarantee performance of the principle of a community – the neural model trained on one data set can be not adequate when giving on its entrance of data from other set. Therefore when using neural modeling procedure of testing of the received results by means of the method of ridge regression based on the theory of regularization incorrectly of objectives is necessary. The being of the offered method of adaptation of a neural network with fixing of shifts (ABNS) is as follows: 1. Instead of a two-layer neural network for adaptation the single-layer neural network more fully answering to use of a method of characteristic points as which the weighed sums of separate groups of signs get out is recommended. 2. For elimination of a problem of the ambiguity caused by a traditional choice of casual entry conditions, initial values of scales and shifts of neurons get out equal to zero. 3. For methodological unity of the solution of a straight line and the return problem of examination, on weight and shift of a neural network the following restrictions are programmatically imposed: the weight [0, 1], and shifts forcibly rely equal to zero by an adaptation speed parameter choice. 4. Results of neural modeling can often be doubtful owing to violation of the principle of a community and check of its observance requires obligatory testing of the received results, for example, by means of a method of ridge regression. As appears from the presented results, in all cases it is necessary to use the offered methods of consecutive and group adaptation with fixing of shifts of neurons, as thus there is a possibility of restoration of initial regression model. When fixing zero shifts of neurons their found weight gain values from the range [0, 1] that provides methodological unity of the solution of a straight line and return problem of examination.

composite index of quality, neural modeling, method of adaptation a neural network with fixing of shifts

1. Khaykin S. Neironnye seti [Neural networks]. Moscow, Saint-Petersburg, Kiev: Williams, 2006. 668 p. (In Russ.)

2. Medvedev V.S., Potemkin V.G. Neironnye seti. MATLAB 6 [Neural networks. MATLAB 6]. Moscow: DIALOG-MIFI, 2002. 496 p. (In Russ.)

3. Osovskii S. Neironnye seti dlya obrabotki informatsii [Neural networks for information processing]. Moscow: Finansy i statistika, 2004. 344 p. (In Russ.)

4. Dreyper N., Smith G. Prikladnoi regressionnyi analiz. Mnozhestvennaya regressiya Applied regression analysis. Multiple regression]. Moscow: Dialektika, 2007. 912 p. (In Russ.)

5. Esbensen K. Analiz mnogomernykh dannykh. Izbrannye glavy [Analysis of multidimensional data. Elected heads]. Chernogolovka, Izd-vo IPKhF RAN, 2005. 160 p. (In Russ.)

6. Bukharin S.V., Mel’nikov A.V. Klasterno ierarkhicheskie metody ekspertizy ekonomicheskikh ob”ektov [Cluster-hierarchical methods of examination of economic objects]. Voronezh, Nauchnaya kniga, 2012. 276 p. (In Russ.)

7. Bukharin S.V., Mel’nikov A.V., Navoev V.V. Forecasting of values of the generalized indicator of quality of volume announcers on the basis of neural modeling. Vestnik VI MVD Rossii. [Bulletin of Voronezh institute of the ministry Internal affairs of Russia], 2014, no. 2, pp. 165-174. (In Russ.)