The dust removing efficiency of electrostatic precipitators (ESPs) greatly depends on the voltage output of power supplies, and the fault recovery strategy when flashover fault happens plays an important role to ensure the efficiency. In classical recovery methods, such as three-line method and the two-line method, the output voltage of ESP power supplies has to operate in low value areas for a long time or to be given a limited up bond. Thus, the average output voltage is low. Moreover, for approach such as the two-line method, the adaptive capacity to environment is also poor when the electric field changes. For overcoming above deficiencies, this paper presents a new fault recovery strategy based on support vector regression (SVR). The strategy firstly predicts the voltage at which the soon-coming spark is likely to occur by analyzing near fault voltages. And then on this basis, it optimizes the process and parameters of two-line method to strengthen its adaptive capacity to environmental change and increase the average output voltage. Four groups of comparative experiments are made in the paper. Among all the experiments, the proposed strategy not only has the adaptability of the three-line method to the environmental change, but also yields a higher average output voltage than the traditional methods. Thus, the superiority of the proposed method is validated.