Cavitation-induced loading oscillations are mainly responsible for the pump vibration, the excitation source identification of which is critical for the control of pump vibration. The water-tunnel experiments and numerical simulations are conducted to study the pump cavitation under the wake flow perturbation of inlet guide vanes. Meanwhile, an excitation source identification method based on second-order statistics is proposed, which connects the blade loading distribution and the overall thrust considering physical causality. Combined with the mode decomposition, the loading distribution of pump cavitation with a dominant energy contribution can be effectively revealed. The results showed that the temporal-spatial characteristics of inflow incidence angle largely regulate the development and distribution of blade cavitation. The correlation modes with amplitude can represent the energy contribution. The normalized correlation modes can represent the similarity between variables. For the cavitation condition, it is found that the whole part from the blade middle to the hub contributes much to the thrust oscillations through excitation source identification based on the second-order statistics, which is dominated by low frequency. However, only the cavitation-induced loading oscillations can be extracted by first-order statistics.