Quantum Reservoir Computing for chaotic time series and extreme events forecasting

Abstract

Recently, it has been demonstrated that quantum reservoir computing (QRC) is an effective tool for time-series prediction. In this work, QRC is applied for multivariate, non-linear and chaotic time series predictions using the IBM Quantum platform. We compare the results of QRC predictions with its classical counterpart, analysing reservoir dynamics and sizes. Specifically, this comprises time-series prediction, reservoir memory capacity and long-term statistics of the chaotic systems. The analysed chaotic systems are, 3-dimensional Lorenz-63, 10 and 20 dimensional Lorenz-96, and 9-dimensional chaotic shear flow model for extreme events. We show that in comparison to the classical reservoir, fewer hyper-parameters are needed by the quantum reservoir, highlighting the robustness of the method. Furthermore, we have compared ensembles of networks between the quantum and classical schemes, with optimal hyper-parameters chosen through grid search and Bayesian optimization. We find that entanglement in the quantum reservoir enriches reservoir dynamics and prediction capabilities. More specifically, one can reduce the size of the quantum reservoir by comparison to its classical counterpart while preserving accuracy of the output. Finally, we analyse the challenges and opportunities of QRC for recurrent time series predictions on Noisy Intermediate Scale Quantum computing (NISQ)-era quantum processors. This work confirms and highlights the potential of QRC for chaotic time series predictions on near term quantum computers.