The bellows pump (BsP) has been widely used for ultraclean fluid transportation in the semiconductor manufacturing process (SCMP). To ensure stable pump output and a high yield rate, maintaining the efficiency and safety of the entire SCMP system is crucial. Thus, flow pulsation compensation is vital. Advanced prediction of pump flow provides key information for timely pulsation compensation and effective monitoring of working conditions. However, the ultraclean requirement brings great challenge for it. This article proposes a flowrate prediction method based on multisensor information fusion processed by a developed neural network, combining the derived bellow deformation and the instantaneous flowrate. Taking advantage of the nonferromagnetic characteristic of the BsP component composition, the in situ bellow deformation can be contactlessly derived by solving an inverse magnetic problem about the marked rectangular permanent magnet (rPM). Experimental systems have been constructed for both inverse magnetic solution and flowrate prediction validations. The experimental results show that the absolute rPM marker localization errors range in [0.094, 0.409] mm with a mean value of 0.207 mm and [0.094, 0.409] rad with a mean value of 0.046 rad; the absolute errors of the predicted flowrate range in [0.02, 0.09] L/min for a flowrate of [3.12, 5.14] L/min. As a typical application, the predicted flowrate enables flowrate fluctuations control in the BsP, which will directly influence the wafer cleaning yield and residues of particles. Simulations based on Simulink were carried out, utilizing the experimentally derived predicted flowrate. The results show that the maximum flowrate fluctuation can be reduced from 3.0 to 0.3 L/min for a flowrate of 3.1 L/min, addressing the application of the proposed flowrate prediction method.