In order to solve the problems of high volume fraction silicon carbide particle reinforced aluminum matrix composite (SiC_p/Al) machining difficulty and poor surface quality, the longitudinal torsional ultrasonic vibration assisted milling composite process was proposed. Taking ultrasonic amplitude, cutting speed, feed per tooth and cutting depth as variables, a four-factor and five-level orthogonal experimental study was designed. By using response surface method and artificial neural network, the prediction models of cutting force, cutting temperature and surface roughness are established, the interaction effect of two indexes among the four parameter variables is analyzed, and the accuracy of the prediction models is compared and verified. Finally, the multi-objective parameters of cutting force, cutting temperature and surface roughness are optimized by genetic algorithm. The results show that both the response surface method and the artificial neural network have better predictive ability, but the artificial neural network is more accurate. The optimal parameter combination optimized by genetic algorithm is ultrasonic amplitude A=1.84 μm, cutting speed v_c=20 m/min, feed per tooth f_z=0.015 mm/z, cutting depth a_p=0.8 mm. After verification experiment, it is found that the optimal parameter can effectively reduce the cutting force, cutting temperature and surface roughness, and the values are F_t=7.23 N, T=40.18 ℃, R_a=2.4673 μm, respectively. And the prediction errors were 6.91%, 6.53% and 2.53%, respectively, which proved the accuracy of the prediction model and the effectiveness of the optimization parameters.