With the large number of distributed power supplies, the harmonic impedance measurement of the microgrid system is more dynamic and uncertain. The active measurement of the harmonic impedance of the system is an important basis for harmonic analysis and filter design. When analyzing the stability of the microgrid system, the acquisition of the dq-axis harmonic impedance parameters is crucial. In this paper, the dq-axis harmonic impedance of the Norton equivalent circuit of the microgrid system is theoretically deduced, and the harmonic impedance model of the RL system is established. In order to improve the prediction accuracy, this paper introduces the prediction model of BP neural network, and introduces its core algorithm-gradient descent principle, and establishes a prediction model based on the dq-axis system harmonic impedance. Then in order to obtain the irrelevant injecting harmonic current vector to solve the system harmonic impedance matrix, the article introduces the rotating current phasor method and actively obtains the system dq-axis harmonic impedance. Finally, the simulation and prediction of the dq-axis harmonic impedance of the microgrid system are carried out. The simulation results prove the effectiveness of the proposed strategy.