ANN‐Based Alternative Controllers for Three‐Phase Four‐Wire Grid‐Connected NPC Inverters

Abstract

ABSTRACT The synchronously rotating reference frame (SRRF)‐based proportional‐integral (PI) control technique has been though studied in many different inverter applications, a well‐designed and clearly presented application of this technique for the three‐phase, three‐level, three‐leg and four‐wire (3P3L3L4W) grid‐connected (GC) neutral point clamped (NPC) inverter has not been found. Therefore, in this study, firstly, the 3P3L3L4W GC NPC inverter is controlled with the SRRF‐based PI controller. Then, to achieve an optimal artificial neural network (ANN) controller in terms of computational burden and control performance, two different ANN controllers, named ANN‐1 and ANN‐3, are designed with data obtained from the PI controller. The control objectives of the NPC inverter are carried out by a single ANN in ANN‐1 and by three independent ANNs in ANN‐3. The training results for ANN‐1 and ANN‐3 are approximately the same, but their computational burdens are quite different. Because ANN‐3 consists of three ANNs with minimum complexity, it has much less computational burden than ANN‐1. Their control performances are compared by using the MATLAB/Simulink, and presented for constant current reference, sudden changes in current reference, current reference with white Gaussian noise, sudden changes in DC source voltage, grid voltage imbalance, sag and swell, and different line filter parameters.