Abstract: This paper presents a method for predicting the errors of electronic voltage transformers(EVT) based on the subtraction-averaging-based optimization(SABO) algorithm and extreme learning machine(ELM). First, the fundamental principles of ELM and the SABO algorithm are introduced, followed by a detailed explanation of the parameter optimization process for the SABO-optimized ELM network. Next, actual EVT error data from a substation in Jiangsu is preprocessed for analysis. Finally, the proposed model is compared with common predictive models, including convolutional long short-term memory(CNN-LSTM) and convolutional gated recurrent unit(CNN-GRU), to evaluate their performance. The results indicate that the SABO-optimized ELM network provides the most accurate predictions for EVT errors, demonstrating superior evaluation metrics and offering a reliable method for accurately assessing EVT error states.