<p>This paper addresses model inaccuracy arising from strong disturbances and time-varying parameters on open-pit mine roads and proposes a trajectory-tracking control method with model-mismatch compensation. First, a lateral yaw and tracking error dynamics model is formulated and augmented with a mismatch state. Second, building on a multi-scale Model Parameter Ratio (MPR) theory, an adaptive Kalman filter is designed to independently scale lateral and longitudinal process noise as well as measurement noise, improving estimation accuracy via innovation whitening. Finally, the estimated mismatch is fed forward into the MPC prediction model, creating a predict-and-cancel mechanism for persistent disturbances and reducing reliance on nominal model accuracy. Simulation experiments show that, on a critical road segment, the proposed approach reduces the maximum lateral absolute error by 28.24% and the maximum positive speed error by 28.07% relative to a baseline nominal model; under a 35&#xa0;km/h crosswind, the corresponding improvements are 19.97% and 31.85%. These results demonstrate superior accuracy and robustness under complex conditions, supporting safe and efficient operation of autonomous mining trucks.</p>