Digital predistortion is an effective technical method to improve the nonlinearization of power amplifier. Based on the fractional-order memory polynomial model and the traditional bilinear polynomial model, an improved fractional-order bilinear polynomial model predistortion method for indirect learning structure is proposed. The model uses the indirect learning structure and Recursive Least Square (RLS) adaptive identification algorithm to simulate the adaptive predistortion system. The experimental results show that, compared with the traditional bilinear polynomial model, the Normalized Mean Square Error (NMSE) is improved by 2.6 dB with 34.3% reduction of the number of model coefficients and the improvement of Adjacent Channel Power Ratio (ACPR) of 5 dB. Therefore, the model can effectively compensate the distortion of the power amplifier.