To solve the problems of limited data and insufficient data utilization during the service period of high-reliability equipment， a reliability assessment method based on an improved Sage-Husa adaptive unscented Kalman filter（UKF） is proposed. This method integrates the distribution parameters of the measured data reflecting performance degradation with predicted distribution parameters. A statistical model is established for performance degradation data， and a degradation failure model is constructed based on the optimal estimates of the distribution parameters. The least squares support vector machine （LS-SVM） is utilized to predict the distribution parameters， serving to characterize the data trend information. The predicted data is set as the state variable and the measured distribution parameters as observation variables. Based on the nonlinear Wiener process， the state and observation equations of the UKF model are established to obtain the optimal estimates of the distribution parameters. To reduce the impact of the measurement noise covariance matrix on the filtering accuracy， the improved Sage-Husa algorithm is introduced to optimize these matrices， enhancing the accuracy and stability of the optimal estimates. Exemplary results demonstrate that the proposed method increases the information content of the data available for reliability assessment. The reliability evaluation results obtained for the equipment are in good agreement with the actual engineering results， validating the rationality and effectiveness of the method.