Abstract: To timely obtain the thickness of the roadway surrounding rock loose circle and take measures to ensure the safety of the surrounding rock，support vector machine (SVM) based on maximum interval algorithm is adopted to make the prediction.Considering that the performance of SVM largely depends on the choice of parameters，the improved artificial fish algorithm to optimize the parameters of SVM is put forward，in order to obtain better accuracy.This method firstly added the idea of layering and crossover mutation to the basic artificial fish algorithm，then optimized the parameters and adopted the SVM to regress and predict for the test data.By improvement of artificial fish behavior and the parameters，the search space is expanded and the stability of global optimization is increased.Problems existing in the artificial fish late optimization such as slow speed are solved.The prediction on monitoring data of a roadway surrounding rock loose circle thickness showed that the model of roadway surrounding rock loose circle thickness has a higher prediction accuracy，which can shorten the search time，increase convergence speed and improve its generalization performance.This research can play a guide role in monitoring the safety of surrounding rock of roadway.

Key words: Layered artificial fish algorithm, Support vector machine, Surrounding rock, Loose circle thickness