A multi-layered dynamic neural group method for characteristic patterns identification and prediction of complex event series

In this paper, a new method based on multi-layered dynamic neural group network for analyzing event series is introduced. By the embedded multiple parallel structures, the new method can identify the character patterns contained in the event series. Then, a selective evaluation strategy is applied to integrate the different pattern clusters and predict the event in the next step. The aim is to generate the complex dynamic behaviors about the controlled system. The fundamental concepts and framework of this method are explained in detail. The effectiveness of our approach is demonstrated on the Internet-based telerobot soccer system by simulation experiments. The results are compared to those based on static neural group network. It is showed that, the telerobot can produce the predictive behaviors with high accuracy under the control of multi-layered dynamic neural group network. The proposed method could properly increase the local-autonomy of telerobot and maintain the stability of system. The conclusions and future work are described in the end.