Corrosion problems can easily lead to safety hazards and even serious safety accidents in refining and chemical equipment, so it is of great significance to develop rapid and accurate corrosion state prediction technology for refining and chemical equipment. Most previous studies have made corrosion prediction based on ideal experimental data and virtual generated data, but have ignored the actual situation. This paper collected 621 sets of corrosion detection data for acid water vapor extraction equipment and adopted the synthetic minority over-sampling technique with edited nearest neighbors (SMOTEENN) algorithm to solve imbalances of the original data. A prediction model for predicting the equipment corrosion state (including corrosion type and corrosion degree) based on a genetic algorithm-based K-nearest neighbors (GA-KNN) algorithm was established. The results show that in terms of dataset balancing, the SMOTEENN algorithm can effectively balance the dataset and improve the model’s ability to recognize the corrosion state of the equipment. In terms of equipment corrosion state prediction, the KNN model optimized by a genetic algorithm has better prediction effect, with the prediction accuracy of equipment corrosion type and corrosion degree reaching 0.9933 and 0.9812, respectively. The model can realize a comprehensive diagnosis of equipment corrosion, and provides theoretical guidance for corrosion monitoring and the maintenance of acid water vapor extraction equipment.