Aiming at the problems of poor adaptability and low accuracy of traditional runoff prediction methods，a runoff prediction method based on chaos recognition SVM combination model is proposed.Taking Dianxia Reservoir as an example，a chaotic recognition support vector machine combination prediction model is constructed.This method is compared and analyzed with three models，i.e.the maximum Lyapunov exponent chaos prediction model，ANN，and AR models，to test the application effect of the combination model.The results showed that the average relative error evaluation indicator of the corresponding four models were 12.3%，14.6%，17.8% and 21.2%；The coefficient of certainty is 0.85，0.53，0.59 and 0.72，and the pass rate is 90.1%，74.8%，68.9% and 62.6%.Therefore，the SVM combination prediction model based on chaos recognition has the highest accuracy and credibility in predicting reservoir runoff，and the prediction effect is better than other methods.The research results can provide theoretical basis for predicting the runoff of the underground reservoir in the store.