Abstract: Ultra-precision aerostatic guideways are key components for achieving nanometer-level motion. The reliability of controller parameter tuning in simulation depends on the accuracy of the open-loop model, as model deviations can render the tuned parameters inapplicable to the actual system. To address this issue, a hybrid predictive modeling method with residual network-based correction along with a controller parameter tuning strategy is proposed. A linear model is identified through an open-loop frequency sweep test, and a residual network is constructed to capture its unmodeled dynamics. Bayesian optimization is then performed based on the hybrid predictive model for controller parameter tuning. The results show that the hybrid prediction modeling based on residual networks has improved the accuracy of the open-loop model. The fitting degree under swept-frequency testing has increased from 64.00% of the linear model to 75.05%, and the deviation of the step performance indicators between the simulation and the actual response is only 5.59%. The parameter tuning based on the hybrid prediction model reduced the overshoot of the step response from 33.03% to 10.84%, significantly improving the dynamic performance of the air-float shaft.