The problem of predicting the technical state of complex systems is considered. Regularization is necessary by attracting additional information, which can be obtained either by expanding the space of technical condition parameters, or by expanding the space of parameter estimates, including those that characterize the structure of a number of dynamics by identifying the degree of irregularity in the process of changing the technical condition to solve this class of incorrect problems. This task is solving in two stages. In the first stage solves the problem of synthesis of model of spatial extrapolation, which can be reduced the task of building predictive models in space estimates of the parameters of the technical state of complex dynamic objects, on recovering an unknown vector function on vector value argument. At the second stage, if the partition of the space into classes in is not fully defined general, then the partition is refined by using regularizers. The developed prediction algorithm in a neural network logical basis combines the advantage of the traditional algorithm for calculating estimates, namely universality, and the advantage of neural networks as high performance with the required reliability of the forecast. The proposed algorithm can be realize in software and mathematical support for predicting the technical condition of complex dynamic objects, such as aircraft.