Abstract: Precise prediction of water quality trend is of vital importance for water resources management. Commonly used data-driving models cannot reflect the physical characteristics of research objective. In view of this, a neural network coupled with dynamic equation is proposed in this paper, and the method to couple dynamic equation into model iteration is also given. A numerical case and a practical case are used to demonstrate the difference between network model with mechanism priori-knowledge and traditional network model. The results of fitting degree and calculation error indicate that the coupled priori-knowledge is able to improve calculation accuracy and enhance non-linear fitting. The proposed model is applicable and rational in water quality prediction. Sample size is the basis of neural network model application, and coupling mechanism priori knowledge under the circumstance of fixed sample size is an efficient approach to improving prediction accuracy.

Key words: water quality prediction, neural network model, dynamic equation, mechanism priori knowledge, Mackey-Glass chaotic system