For the processing of Irkutsk Incoherent Scatter Radar data we consider the prob- lem of non-linear discrepancy functional minimization in long experimental data series processing when measured process is described by integro-differential equation. Considering the non-linear inverse problem for functions with two continuous derivatives determined on compact we present three main techniques parts: descent algorithm, genetic algorithm and solution regularization. The central part of the technique is the descent algorithm that works on n-dimensional net as- sociated with some regular polytope. The regularization of solution is carried out by means of transition to Sobolev space. For deleting of ambiguities due to set of sub-optimal solution we use adaptation scheme on the base of genetic algorithms.