Abstract: In order to predict the surrounding rock stability in coal roadway scientifically and accurately, a new prediction method combined self-adaptive different evolution （JＡDＥ） with extreme learning machine （ＥＬＭ） was proposed. Considering the advantages of ELM which has quick training speed, good generalization performance and accessing to the global optimal solution easily, JADE is used to optimize the input layer weight matrix and hidden layer bias of ELM so that the random error could be reduced, and the JADE-ELM prediction model is built. Case analysis is made by using measured data of Huozhou Coal Mine area, and the prediction result is compared with ELM, BP and RBF. The results show that the average accuracy of JADE-ELM model is 97.85% which is 4.05%, 17.85% and 22.85% higher than that of ELM, BP and RBF, it proves that the JADE-ELM model has a higher accuracy and can be more accurate to predict the stability of coal roadway.