Double-Vector Model-Free Predictive Current Control Method for Voltage Source Inverters With Sampling Noise Suppression

Model-free predictive current control (MFPCC) methods based on look-up tables (LUTs) have been widely applied in voltage source inverters (VSIs) due to their simple implementation and excellent robustness. However, the performance of the VSIs is affected by the voltage vector used, accuracy of current gradients, and sampling noise. In order to improve these, a new double-vector MFPCC method is proposed in this article. First, a synthetic voltage vector composed of double vectors is applied in each control period to reduce current ripples, where the durations of the double vectors are calculated using predicted current gradients. Next, an extended LUT is designed by using the current gradients of the applied synthetic voltage vectors to update all current gradients of eight basic voltage vectors, so as to increase the accuracy of current gradients. Then, the effect of the sampling noise on MFPCC is analyzed in detail and suppressed by a sliding mode observer for the VSIs. Finally, simulation and experimental results have verified the effectiveness of the proposed method.

I. Introduction

Model predictive control (MPC) methods have been widely applied in voltage source inverters (VSIs) due to their simple implementation, fast response, and multivariable control capability [1], [2], [3]. In order to meet the grid connection requirements, the output current of VSIs should have small ripples and low total harmonic distortion (THD). However, the current performance (mainly refers to the THD and the current ripples) of the existing MPC strategies depends on the accuracy of model parameters. The mismatched parameters can lead to select the nonoptimal voltage vectors and cause the prediction errors, giving rise to current ripples and THDs [4], [5], [6]. In order to overcome this drawback in MPC, some researchers develop an online observer for determining model parameters [7], [8], [9]. The prediction errors caused by inaccurate parameters are evaluated and compensated by utilizing disturbance observers [10], [11]. These are effective but bring about other problems, such as the complexity of system control and the increase of computational burdens.