The invention discloses a cylinder jacket-piston ring running-in state prediction method. According to the method, sample data is obtained through cylinder jacket-piston ring running-in testing, and aBP neural network based on an improved variable-gradient SCG algorithm is trained and tested; and next, condition parameters needing to be predicted are input, and then corresponding running-in stateinformation parameters can be obtained. The prediction method is high in speed, high in precision, high in reliability and high in feasibility.