The tokamak magnet power supply (TMPS) is an important device for achieving plasma displacement balance control (PDBC) in nuclear fusion, but practical application of nuclear fusion is seriously affected by the continuous increasing demand of TMPS capacity. In order to reduce power capacity and realize capacity optimization of TMPS, grey prediction is used to predict the plasma vertical displacement (PVD) signal, and the TMPS reference signal is obtained in advance based on the predicted PVD value. When a small PVD error occurs, only a small current is needed from TMPS to control the displaced plasma back to the equilibrium position. To further improve the accuracy of the PVD prediction, old information in the original sequence of the grey prediction is transformed and weakened to enhance the weight of new information. A simplified Simpson formula is used to reconstruct background value to improve the prediction accuracy of the grey prediction model. The PVD predicted value error is repeatedly checked to ensure grey prediction model prediction accuracy, and when the PVD prediction error is accepted, the PVD prediction of the next moment is carried out. The PDBC model is established, and a reference signal obtained in advance is provided to TMPS to ensure the TMPS output current tracking reference signal, achieving excitation of the load coil to realize PDBC. After analysis and verification, PVD prediction is effectively achieved by improved grey prediction, and in the process of achieving PDBC, compared with TMPS actual engineering capacity, the designed TMPS capacity optimization method can significantly reduce the maximum output current and terminal voltage of TMPS, achieving TMPS capacity optimization.