The process of water quality assessment is multivariable, nonlinear and uncertain. The traditional particle swarm optimization training neural network water quality evaluation model has slow convergence speed and poor generalization performance. In order to overcome the shortcomings, this paper proposes a new model to use dynamic multigroup particle swarm optimization algorithm to train support vector machine, in which the DMPSO is used to optimize the parameters of the fuzzy neural network model. The model combines the search performance of PSO algorithm, the efficiency and robustness of SVM, which can improve the generalization ability of the model. Through the simulation experiment on the hydrological data, the results show that the relative error of this model is 2.74%, which is much lower than the 4.21% relative error of the traditional particle swarm optimization. It is proved that the efficiency and accuracy of the model are improved, and is suitable for the daily water quality evaluation.