Abstract: Model predictive control is applied to the PWM rectifier to reduce the pulse vibration in the direct power control and improve the dynamic response speed. However, the prediction of the state quantity of the future time in the traditional model predictive power control (MPDPC) depends only on the model. The parameter variation of the model is very sensitive. The measurement of voltage sensor and the influence of grid side harmonics limit the power prediction. In order to realize the real-time identification of the parameters of the rectification side and improve the overall prediction accuracy, achieve precise control of power. Based on the introduction of adaptive neural network voltage observer, a voltage-sensorless model predictive power control of PWM rectifier based on adaptive neural network observation (ANMPDPC) was proposed. By constructing an adaptive voltage observer including an adaptive neural network identifier and an adaptive neural network filter, to realize the influence of voltage higher harmonics on the grid side voltage estimation, and the voltage observer and power two. The combination of step prediction further reduces the power pulse vibration and improves the response speed and control accuracy of the system. The simulation and experimental results show that the proposed improved strategy not only achieves the model predictive control under the no-voltage sensor, but also effectively suppresses the influence of the high-frequency interference of the grid side harmonics and the parameter variation on the prediction accuracy.