Topology detection of complex networks with hidden variables and stochastic perturbations

Complex networks have found widespread real-world applications. One of the key problems in research of complex networks is topology identification, which is concerned with deciding the interaction patterns from observed dynamical time series. This presents a very challenging problem, especially in the absence of the knowledge of nodal dynamics and in the presence of system noise. In this paper a simple and yet efficient approach is proposed for topology identification of complex networks in such challenging scenarios. The main idea behind the proposed approach is to use piecewise partial Granger causality, which measures the directed connections of nonlinear time series influenced by hidden variables. The effectiveness of the proposed approach in relation to network parameters is demonstrated by a commonly-used testing network.