The particle swarm optimisation algorithm was used to optimise probabilistic neural network for the metro rail corrugation, and a particle probabilistic neural network (PPNN) algorithm was proposed. The proposed algorithm randomly initialized the smoothing factor of the probabilistic neural network within a certain range of values. To ensure the global search capability and the computational efficiency of the proposed algorithm, the concave function decreasing inertia weights were chosen to achieve the update iteration of the smoothing factor, resulting in the optimal solution of the smoothing factor with the highest classification accuracy. To illustrate the effectiveness of the proposed algorithm, field tests were conducted on the roughness of the rails and the interior noise, the features of interior noise associated with rail corrugation were extracted, the effect of the population size and the number of evolutions of the algorithm on the accuracy of rail corrugation recognition was analysed, and the recognition results of different intelligent classification algorithms were compared. Results showed that the interior noise associated with rail corrugation was characterised by A-weighted sound pressure levels at 315, 400, 500, 630, 800 and 1000 Hz centre frequencies. The advantages of the proposed algorithm over mainstream intelligent classification algorithms such as decision tree, Gaussian naive Bayes, support vector machine and K-nearest neighbor are significant, with an accuracy of 98.582% for rail corrugation recognition.