The invention discloses an optimization method for the commutation failure prediction control of a DC control protection system. According to the technical scheme of the invention, the zero-sequence voltage 3U0 signal of a commutation bus at an inverter station is input into a fast Fourier converter, so that a fundamental component in the signal is separated out. After that, the fundamental component is compared with a preset setting value. When the fundamental component is smaller than the setting value, 1 is outputted. Otherwise, 0 is outputted. Meanwhile, an action signal Z_DET is inputted into a monostable multivibrator, and a corresponding time setting value is set. In this way, the action signal Z_DET is maintained at a high level, and then is outputted through a delay assembly in the delayed manner. The above two signals are together outputted to a two-way input selector through an AND gate to serve as an action criterion Ctrl. As a result, a single-phase fault and the disturbance of a magnetizing inrush current can be effectively distinguished from each other. When the single-phase fault occurs on the alternating-current side, the control logic is reliable in action. When the magnetizing inrush current occurs on the alternating-current side, the control logic can effectively recognize and inhibit the repeated fluctuation of the direct-current power.