间歇过程的优化控制往往依赖于过程精确的数学模型,快速反应的市场要求使得数据驱动的建模方法被应用到了间歇过程的建模中。但常规的数据驱动建模方法在模型结构中没有考虑间歇过程具有重复性的特性,只是简单地将间歇过程作为一般的非线性结构进行处理。针对该问题,本文提出一种新颖的间歇过程时变神经模糊模型,将时间轴和批次轴的信息统一在二维集成模型的框架下,对间歇过程的输入输出数据按照三维信息进行处理,使模型参数变为时间的函数,从而按照批次轴方向进行学习,合理地应用了间歇过程在批次轴方向上的重复性信息。因此,通过引入时变权重的概念,使模型结构中蕴含了间歇过程重复性的特性。在此基础上,提出一种基于迭代学习和Lyapunov方法的参数学习算法,避免了传统学习算法中学习参数采用试凑法的缺点,并对模型的收敛性进行了严格的数学分析,得出模型的学习参数在批次轴方向上渐进收敛的结论。最后,将本文提出的时变神经模糊模型应用到一典型间歇过程的建模研究中,与传统的神经模糊模型进行了对比,仿真研究表明,本文提出的时变神经模糊模型具有较好的非线性逼近和自学习能力,能够反应间歇过程的二维模型特性,为间歇过程的建模研究提供了一条新的途径。 The optimization control of batch process often depends on the mathematical model of process precision. The market demand of rapid response makes the data-driven modeling method applied to the modeling of batch process. However, the conventional data-driven modeling method does not consider the repetitive nature of the batch process in the model structure, but simply treats the batch process as a general nonlinear structure. In order to solve this problem, this paper presents a novel time-varying neurological fuzzy model of batch process, which integrates the information of time axis and batch axis under the framework of two-dimensional integrated model, processes the input and output data of batch process according to three-dimensional information, The model parameters are transformed into functions of time to learn in terms of the batch axis direction and the repetitive information of the batch process in the batch axis direction is reasonably applied. Therefore, by introducing the concept of time-varying weight, the model structure contains the characteristics of repetitive process. On this basis, a parameter learning algorithm based on iterative learning and Lyapunov method is proposed, which avoids the shortcomings of the traditional learning algorithm using trial and error method and rigorous mathematical analysis of the convergence of the model, draw the model The learning parameters converge progressively in the direction of the batch axis. Finally, the time-varying neural fuzzy model proposed in this paper is applied to the modeling of a typical batch process, and compared with the traditional neuro-fuzzy model. The simulation results show that the proposed time-varying neural fuzzy model has good non-performance Linear approximation and self-learning ability, which can reflect the characteristics of two-dimensional model of batch process, provide a new approach for the modeling of batch process.