V.N. Klyachkin - Dr. Sci. (Eng.), professor of chair «Applied mathematics and computer science» Ulyanovsk State Technical University. E-mail: v_kl@mail.ru
D.S. Bubyr - post-graduate student of chair «Applied mathematics and computer science», Ulyanovsk State Technical University. E-mail: lbubir91@mail.ru

Piecewise linear regression with breakpoint on the mean response is proposed to be used for state object forecasting. Quality of models is estimated on mean relative forecast error for the test sample. Reducing the value of this criterion is possible through the use of stepwise regression, accounting autoregression and other methods. Results of the carried out research have shown, that at forecasting of a condition of technical object it is expedient to choose a forecasting model from set of various types, using possible methods of estimation and different volumes of samples (in a considered example various updating of piecewise-linear regress have appeared the best variant).