Y.P. Mukha, M.G. Skvortsov
For application neural nets as a measuring means to detention states of complex system on integral parameter essential to design the metrological description a measuring process in neural nets.
In the foundation metrological description of neural net are a analize of measuring errors based by presentation of measuring process as a succession of input value.
The complexity to detention states of modern objects to require of application adequate level complexity formalism of presentation measuring transformation.
The article describes the category operation metrological description of transformation measuring in neuron, in multi layers neural nets and transformation function neural nets (analog digit transformation, normalization, minimum measuring channel).
The metrological description of hypothetical, accepted and realized algorithm of measuring neural nets process is given.
Decomposition of complex error to permit define deposit kind of measuring channel for complex error and changing of error each of channels during changing number of neurons.
The category operation metrological description to permit tradition metrological to detention of measuring neural nets process, to work type of errors of transformation measuring out of detail (in neurons, layers of neurons and measuring channels), to analyze influence the structure of neural nets for errors measuring, to design of systems neural nets measuring with the set metrological characteristics.