Abstract: In view of the difficulty in establishing accurate mathematical model of wheel rail wear and in evaluating, predicting and quantitatively calculating wheel rail wear under various working conditions, this paper proposed a tread wear prediction method based on SQPSO optimized DELM model(SQPSO-DELM). First of all, the derivative characteristics were introduced into the learning machine, and a derivative learning machine model (DELM) was proposed. Then, the sequential quadratic programming (SQP) and quantum particle swarm optimization (QPSO) algorithm were introduced to optimize the parameters of DELM. Through SQPSO-DELM prediction model, the maximum wear of wheel tread under different test parameters of vehicle dynamics model simulation and the actual measured value of wear degree of on-site train tread were trained and predicted. The results showed that the performance parameters of SQPSO-DELM prediction model were better than LSSVM, ELM, PSO-ELM and QPSO-ELM, which can better reflect the influence of different parameters on wheel tread wear value.