Abstract
Study of dynamical systems using partial state observation is an importantproblem due to its applicability to many real-world systems. We address theproblem by studying an echo state network (ESN) framework with partial stateinput with partial or full state output. Application to the Lorenz system andChua's oscillator (both numerically simulated and experimental systems)demonstrate the effectiveness of our method. We show that the ESN, as anautonomous dynamical system, is capable of making short-term predictions up toa few Lyapunov times. However, the prediction horizon has high variabilitydepending on the initial condition-an aspect that we explore in detail usingthe distribution of the prediction horizon. Further, using a variety ofstatistical metrics to compare the long-term dynamics of the ESN predictionswith numerically simulated or experimental dynamics and observed similarresults, we show that the ESN can effectively learn the system's dynamics evenwhen trained with noisy numerical or experimental datasets. Thus, wedemonstrate the potential of ESNs to serve as cheap surrogate models forsimulating the dynamics of systems where complete observations are unavailable.