Adaptive neural schemes for the control of a shunt active power filter

In this paper, two Adaline-based schemes for identifying the harmonic currents generated by nonlinear loads in power distribution systems are presented. Each neural method is based on formalized expressions of the active and reactive instantaneous powers. In both methods, the expressions of the instantaneous powers are learned on-line by Adalines. The amplitude of each harmonic term is thus individually estimated with the Adalines weights. This allows to calculate the compensating currents of an active power filter in real-time. The efficiency of the active power filter using the two proposed neural schemes is demonstrated in terms of total harmonic distortion and power factor with experimental results under various operating conditions. The performance of the neural schemes are favorably compared to other methods.