In order to solve the problem of estimating the longitudinal speed caused by wheel slippage of multi-axle distributed electric vehicles under off-road conditions, a longitudinal vehicle speed fusion estimation algorithm is proposed. Based on triple exponential smoothing and a long short-term memory network, the wheel speed is predicted, and an interactive multi-model method is used to fuse the output prediction results. In this paper, a noise covariance matrix adjusting method based on the trend of wheel speed change is adopted, which the actively adjusted Kalman filter (AAKF) for speed estimation is realized. The acceleration signal is not used ensuring the availability of the AAKF when the inertial measurement unit fails. The AAKF vehicle speed estimation results are fused to ensure the reliability of the algorithm. The simulation and real vehicle test results verify the accuracy of the proposed algorithm.