A universally applicable hybrid modelling method is proposed for nonlinear industrial processes that combine the a priori process knowledge with a data-driven model. This method constructs a unified framework for the modelling process by integrating a data-driven modelling technique, sampling detection technique, constraint optimization problem, and an evolutionary algorithm. In the modelling process, a swarm intelligence algorithm is used to optimize the model parameters under the circumstances of satisfying the constraints of a priori knowledge. By adding the constraints of process a priori knowledge, more information can be obtained about the actual process and the over-fitting problem can be avoided to some extent, especially when modelling a system with a small quantity of samples. In order to show the effectiveness of the method proposed in this paper, two general data-driven models, the polynomial regression model and radial basis function network model, are used as case studies. Moreover, a function simulation experiment is designed to test effectiveness, and applied to estimate average particle size of ZrO2-TiO2 composite colloidal sols.