A method for diagnosing AC series arc faults based on the Inception module and BiLSTM (bidirectional long short-term memory) was proposed to address the challenge of identifying small current changes caused by arc faults in aviation cables. First, features of the raw current data were extracted by calculating the discrete sum of squares of the autocorrelation coefficient, Shannon entropy, and wavelet energy entropy. These features are then combined to form a new feature matrix, enhancing the original data's feature representation. Subsequently, the Inception-BiLSTM network learns from the feature matrix and ultimately completes the arc fault diagnosis. To validate the diagnostic performance of the model in practical environments, a series of experiments were conducted, including vibration tests, stress tests, and wet cable tests, based on an aviation cable arc fault simulation platform, with the experimental data being integrated as detection samples. The experimental results show that the proposed method achieves a high accuracy rate of 99.69% in identifying arc faults.