A controller for electromechanical drive systems according to the structure of an output neuron is proposed. Unlike the synthesis of a traditional neural network of a neurocontroller, when finding weight coefficients, that require multiple iterative computer calculations, for the proposed controller is excluded. It’s determined by the derived analytical relations. Compared to a modal controller, for which it is necessary to measure a number of the electric drive coordinates, including those that are difficult to measure, in the proposed one it is enough to measuring only one output coordinate. It was reached through the use of inverse finite difference method and clean physical concepts. For a linear system, one output neuron is enough, for a non-linear system, their number is equal to the number of sections linearizing the non-linearity. The method was illustrated on example of 2-mass electromechanical system with frictional load.