There are situations in the thin-walled steel girders with box intersection, in which the internal wall of compression flange is elastically restrained in the webs of section and along its length occurs the change of normal stresses. Such a wall was modelled, as a bilateral elastically restrained internal plate variably loaded at the length. Explicit formulation of neural formula on buckling coefficient of internal plate at non-linear distribution of longitudinal stress was discussed in this paper. A few structures of neural networks (NN) were envisaged in order to obtain the best of the numerical effectiveness of the final formula. The results noted from the literature were used for the evaluation of neural prediction of coefficient.
