No abstract

No abstract

This article presents a general approach to increasing the accuracy of measurements and pattern recognition. A criterion for informativeness of an ensemble of signals is derived for the general case of an arbitrary signal distribution. An algorithm for determining the values of parameters from measurement signals is based on the probabilistic distance of measured points from calibration points.At the present time, the accuracy of measurements of the composition and properties of substances (or increasing the reliability of control and pattern recognition) is generally increased by increasing the number of signals used and by using increasingly complicated algorithms for processing them. In this case it is impossible to avoid complicating the measurement information system, increasing its cost, and decreasing the reliability of operation relative to sudden failures; this approach also complicates calibration. It was shown in [1][2][3] that this technical contradiction can be overcome, i.e., accuracy can be increased while simultaneously simplifying the measurement information system, by choosing an informative ensemble of signals and using new algorithms for processing them. The use of tightly correlated normally distributed signals makes it possible to reduce the probability of error in recognizing close values of controlled parameters by factors of tens or thousands even when using only two or three signals, which cannot be achieved with any of the classical methods for increasing accuracy [4]. However, in many practical cases, the scattering laws for random signals with constant values of controlled parameters are not normal laws. Below we will present a general approach to increasing accuracy that can be used with arbitrary distributions for signals (random variables). The new approach is based on successive solution of five basic problems: description of the scattering of signals, which are treated as random variables; justification for informativeness criteria for ensembles of correlated signals; selection of a set of maximally informative ensembles of correlated signals; calibration of a multidimensional measurement information system; and calibration of algorithms for determining the unknown value of a parameter from measurement signals.

No abstract

No abstract