The measurement accuracy of electronic voltage transformer (EVT) is closely related to the security and economy of the power grid. In order to accurately predict the error of EVT during the long?term operation, a combined error prediction method considering multiple features is proposed. This method selects parameters with strong correlation to EVT errors through correlation analysis as feature quantities. It utilizes a fused attention mechanism LSTM model and an SVR model to predict the errors of the transformer separately. The obtained prediction results are then combined to generate the final prediction result. The real?time operational data of a certain substation is simulated and analyzed. The results indicate that the proposed method can effectively predict the error variation information of the EVT over a certain period of time and has certain reference value for the timely prediction of EVT errors in substations and scheduling of measurement performance maintenance.