Abstract: In response to the difficulties in predicting ionospheric irregularities and the low accuracy and tendency to fall into local optima of a single random forest (RF) model in prediction, a RC-WO-RF ionospheric irregularities regression prediction model was constructed by combining the Wild Horse Optimizer (WHO) algorithm with Residual Compensation (RC). Using observation data from the Hong Kong HKWS station from March 1, 2020 to June 30, 2024, the standard deviation of the total electron content change rate (ROTI) was calculated, and a series of background ionospheric parameters related to ionospheric irregularities were selected as input features. The results indicate that, ROTI, Cosine phase daily variation factor and geomagnetic activity are crucial for ionospheric irregularities; The root mean square error of the RC-WO-RF model is less than 0.1 TECU/min, and it has excellent response capability and prediction accuracy for sudden geomagnetic storm events; The average relative accuracy of the RC-WO-RF model in short-term forecasting 30 minutes in advance is 90.67%, which is 8.16% higher than the WHO-RF model and 11.2% higher than the single RF model. The prediction performance of the combined model is significantly better than that of the single RF model and the WHO-RF model.