In [1-4], the main problems of creating a real-time multi-detector neutron spectrometer are considered and simulation modeling of such a spectrometer is carried out in order to determine the optimal composition of the detection unit of such a spectrometer.
 This article states the imperfection of the metrological and methodological support of not only spectrometric, but also radiometric measurements of arbitrary neutron fluxes. To increase the reliability of the results of neutron measurements, it is proposed to verify the multi-detector real-time neutron spectrometer being developed in reference neutron fields with different and reliably known shapes of energy spectra. To create such reference neutron fields, a neutron testing and verification complex was developed, the prototype of which was studied experimentally. But taking into account the limited variety of forms of the spectra of the reference neutron fields created by this complex, it is proposed to train the neural network embedded in the neutron spectrometer on a wider set of base spectra, which includes, in addition to the spectra of the reference fields created on the test and calibration complex, reliably known spectra of neutron fluxes found in the literature. It is shown that in this case the energy errors of the spectrometer when measuring neutron fluxes with a variety of spectrum shapes will be practically eliminated, even when the spectrometer is verified in reference fields with a narrower variety of spectrum shapes.