Purpose <p>The impedance boundary of the acoustic field greatly affects the sound pressure distribution in the acoustic cavity. An innovative model for the identification of acoustic impedance boundary in an enclosed cavity is established in this study.</p> Method <p>The discontinuity caused by general acoustic boundary impedance can be eliminated by using boundary-smoothed Fourier series method to construct the system acoustic pressure function. By refining the sub-step of the acoustic impedance value, the cavity frequency relationship corresponding to each boundary impedance value is calculated. Then multiple sets of corresponding data of acoustic impedance and natural frequencies are collected. The BP (backward propagation) neural network establishes the pattern between natural frequency and acoustic impedance boundary. The acoustic impedance is output when the measured natural frequencies are input into the BP neural network.</p> Results and discussions <p>It is advanced for using machine learning to identify the boundary acoustic impedance. The proposed method accumulates enough data for neural network training and is convenient for handling different boundary impedances. Adequate acoustic impedance and natural frequencies corresponding data sets ensure the identification effect of the neural network with noise, simultaneously mitigating the issue of poor training performances caused by insufficient data. The effectiveness and practicability of the proposed model for the identification of acoustic impedance of enclosed cavities are validated by comparing the results predicted by the model with the data from experiments. This provides a new method for accurate prediction of cavity acoustic fields.</p>