<p>Wheel wear is inevitable during rail vehicle operation due to wheel-rail contact. Increased wear causes the wheel profile to deviate from the original standard profile. Accurate prediction of wheel profile evolution is crucial for monitoring wheel health and ensuring vehicle performance. In this paper, a profile evolution prediction model based on particle swarm optimization Levenberg-Marquardt back propagation (PSO-LMBP) neural network is proposed. The prediction model realizes the control of wheel profile evolution using the operation mileage and the wear position as inputs, and the wear amount as output. It is a new attempt to apply neural network model within the field of wheel profile evolution prediction. To improve accuracy, wear data are intervalized and smoothed using locally weighted regression; the LMBP neural network’s key parameters are optimized using the PSO algorithm. Comparative results show that the PSO-LMBP model achieves higher accuracy than traditional LMBP and BP models. Through applying the model to the prediction of wheel wear for vehicles with longer operation mileage and new vehicles, the long-term prediction performance and generalization performance of the model are verified. Finally, the wheel-rail contact geometry and dynamics performance of the vehicle were evaluated by establishing a wheel-rail contact model and a vehicle dynamics model using measured and predicted profiles. The simulation results show that the prediction model has high accuracy.</p>