Lowering the sound transmission loss of impedance-matching structures: Data-driven optimization assisted with a priori knowledge

•An approach introducing a priori knowledge into data-driven structural optimization of impedance-matching structure is proposed.•The sound transmission loss is decreased by 30% following the one-step optimization without additional iterations.•The optimal structural pattern exhibits the feature of impedance jumping which contradicts those in the training data and impedance-gradient structures commonly employed in the literature.•The underlying mechanisms are revealed with the transfer matrix method, guiding the future design of impedance-matching structure at low frequencies.