To address the serious battery voltage fluctuation of the UAV power system, which leads to the large system noise and the low accuracy of identification results, this study proposes a kind of UAV power system parameters identification method based on the reverse predicted-extended Kalman filter. Firstly, the voltage-drop noise model is considered into the noise identification by establishing the extended parameter matrix. Secondly, the reverse predicted Kalman filter algorithm is proposed. An innovation square of threshold value is set. The ratio of the original predicted innovation square and the reverse predicted innovation square ratio is calculated, which adjusts the process noise by comparing the predicted innovation ratio with threshold size. In this way, the estimation model of correction is realized. Experimental results show that the average error of the RP-EKF algorithm is 39. 22 rpm, the root mean square error is 55. 85 rpm, and the mean relative bias is 0. 85% . Compared with the least square algorithm and the Kalman filter algorithm, the average error index values of the identification results using the proposed method of this study is improved by 41. 51% and 22. 26% , respectively. The root mean square error is improved by 49. 63% and 13. 0% , and the mean relative bias was improved by 41. 7% and 22. 7% . Results show that the proposed algorithm has higher identification accuracy than the traditional identification methods.