Abstract: Lightning strikes on overhead insulated lines in distribution networks frequently lead to disconnection incidents, posing significant risks to the safe operation of power grids. To enable accurate and reliable early warnings of lightning-induced disconnections and reduce their occurrence, a novel early warning method based on threshold judgment and support vector machines(SVM) is proposed. First, an experimental platform simulating arc erosion on conductors is constructed to establish the time window for disconnection warnings and verify their feasibility. Utilizing measured overvoltage data, the transient overvoltage period is identified using a wavelet sliding-window energy approach. Key overvoltage features are then extracted through time-domain analysis and wavelet transform. By integrating threshold judgment with SVM, overvoltage events associated with lightning flashover and arcing are accurately identified, facilitating the detection of disconnection risks. The results indicate that single-phase-to-ground small-phase overvoltage is the most critical factor in predicting lightning-induced disconnections. The proposed method achieves a prediction accuracy of 97.6%, with conductor breakage occurring approximately 40 min after exposure to small current arcing. These findings offer valuable insights into identifying potential lightning-induced breakage risks, enabling timely preventive measures to mitigate disconnection accidents.