Sequential ensemble optimization based on general surrogate model prediction variance and its application on engine acceleration schedule design