Abstract: A dynamic neighborhood particle sware pattern was set on the basis of serial numbers and amount of neighboring particles. The optimum performances of six test functions were analyzed by the Taguchi DOE. The better change pattrens of the PSO parameters, such as inertia weight, number of the particle neighbors and acceleration coefficients, were selected. An optimization model of dynamic neighborhood particle swarm was built, which may have a wider adaptability and lower computation complexity. The penalty parameters and kernel function evaluation parameters of support vector machine were optimized with the proposed model. Compared with that of the BP ANN and SPSO-based SVM models, DNPSO-based support vector machine shows a better characteristics identification ability and higher robustness in the engine fault diagnosis.