Utilizing coupled reservoir computing to predict network dynamics and structure

As an efficient type of recurrent neural network, reservoir computing has been widely applied in time series prediction. However, there remains a scarcity of methods specifically addressing the application of reservoir computing to network dynamics prediction. This paper proposes a coupled reservoir computing method, which integrates network structure into the output layer of reservoir computing, demonstrating significant application potential. On the one hand, it accurately predicts dynamical behaviors by leveraging known network structures, providing a powerful tool for analyzing network dynamics. On the other hand, even with unknown network structures, the method can leverage Granger causality test to effectively predict network topology, overcoming ordinary RC limitations. The results of numerical experiments clearly demonstrate that our method not only offers significant advantages in network dynamics prediction but also enables highly accurate reconstruction of the network structure.