The traditional finite control set model predictive control (FCS-MPC) outputs a single switching state in a control cycle. When the sampling frequency is low， the control accuracy is poor， and the unfixed switching frequency makes it difficult to design the filter. Three vector model predictive current control (TV-MPCC) improves the control accuracy and keeps the switching frequency fixed. In order to improve the dynamic response speed and anti-interference ability of the static var generator (SVG) and reduce the harmonic content of the input current on the network side， a three vector model predictive direct power control (TV-MPDPC) algorithm based on the Luenberger observer is proposed. The inner loop of the control algorithm adopts three vector predictive direct power control， which has better control accuracy than FCS-MPC and better dynamic response performance than TV-MPCC. At the same time， aiming at the high dependence of the control strategy on the parameters， especially on the parameters of the grid connected inductor， the Luenberger observer is proposed to compensate the disturbance caused by the parameter mismatch. The simulation results show that the algorithm is effective and feasible.