For the problem of difficulty in modeling of time-varying nonliner system, an approach to identification and prediction of system based on dynamics cluster multi-dimensional Taylor network model is proposed. Firstly, the method of construction of multi-dimensional Taylor network model is discussed, whereby the dynamics cluster is defined. Then, the self-reconfiguration algorithm of the dynamics cluster multi-dimensional Taylor network is given. Finally, an example is given to illustrate the method of the dynamics cluster multi-dimensional Taylor network how to work in the practical applications, and the results show the effectiveness of the proposed method.