In this paper is presented the implementation of a VSC (voltage source converter), to be used in HVDC (high voltage direct current) systems. The control is carried out of active and reactive power flow for the VSC, by an inverse neural control, based on multi layer perceptron. The non linearity nature presents learning problems in the Neural Network like poor identification for the VSC, when the neural control is implemented in real time. This paper proposes a design for an inverse neural control of the active and reactive power flow for the VSC converter, and it solves the learning problem through real time control for the VSC. Four neural controllers are used, which are training and implemented in parallel, activated just one in time, depending on the work region in the converter. The system is controlled in a -100 W to 100 W range of active power and -30 VAR to 30 VAR in reactive power.