Deadbeat Control of PWM Inverter with Repetitive Disturbance Prediction

Abstrac-A new approach for deadbeat control is presented, in which the estimation of the disturbance is carried out with a repetitive predictor instead of a disturbance observer. Taking advantage of repetitiveness of the loads, it gives good performance at lower sampling frequency. Simulations and experiments confmed the advantages. I I II I L INTRODUCTION In recent years, much research has been carried out on digital feedback control of PWM inverters. Among them, the deadbeat(DB) control is distinguished by fast response. In early approaches[l,2] of the DB control, resistive load was assumed, but the system performance deteriorates when the load type changes. This situation can be improved if repetitive control is applied to compensate the cyclic disturbance[6], but designing a good repetitive controller is never a simple task. Another solution is taking the load current into account[2,3]. Unfortunately, this can not be accomplished by just sampling the load current because the control law need to be carried out a sampling period ahead, otherwise the maximum possible pulsewidth will be significantly reduced[l-3]. So the disturbance (i.e. load current) as well as the state variables must be predicted, and the prediction must be precise enough to ensure close regulation of the instantaneous output voltage. In recent effort of disturbance observer based DB control[3], the system performance was improved by using a state observer and a disturbance observer for this job. The performance of the state observer is satisfactory due to a deterministic and accurate state model. However, the assumed disturbance model is rather imperfect when considering the variety of load types. There is steady-state estimation error for the disturbance observer derived form this model, and the error increases quickly when harmonic contents in the load current gets heavy, typically with nonlinear loads. The estimation error can be unacceptable if the sampling fiequency of the load current is not high enough. In other words, this kind of system depends on high sampling Mg.1 Plant model