Abstract: This paper establishes a single-station mobile direction-finding positioning model to address the significant positioning errors caused by abnormalities， fluctuations， and missing direction-finding information in mobile direction finding. It proposes a radiation source positioning method based on long short term memory and Kalman filtering （LSTM-KF）. The tracking technology is applied in reverse to smooth， complete， and suppress the random fluctuations of the direction-finding values， thereby achieving high-precision electromagnetic radiation source positioning. First， the proposed algorithm utilizes an extended short term memory network based on the state information learned earlier， the current input information， and the network state （LSTM） algorithm to obtain the current time of arrival angle （direction of arrival， DOA） prediction value， and employs this prediction to address the missing values. Then， using the obtained angle prediction and measurement values， it adds fitting to ensure the continuity of direction changes， applying the Kalman filter to smooth the data， and correct and update to obtain the optimal DOA estimation for the current time. Finally， it uses the direction-finding intersection positioning algorithm to achieve positioning. In an actual test scenario， we utilize a vertical fixed-wing UAV platform equipped with an antenna array to gather data. Combined with simulation experiments， the results demonstrate that the positioning accuracy of the proposed algorithm is significantly improved compared with that prior to data processing， thereby verifying the effectiveness of the proposed algorithm.