Unsupervised neural network for dynamics control under chaotic regime

Abstract. Some natural phenomena have strong impacts on society, implying the loss of human lives and a huge amount of financial costs. Therefore, the possibility of controlling severe natural phenomena is of practical interest, preserving lives and mitigating costs. The simulation of dynamical systems could help us to predict severe events, allowing the deployment of control actions to avoid or to mitigate the severity associated with the predicted phenomenon. Similar to the numerical experiments of Miyoshi and Sun (2022), here, ensemble predictions are employed to drive the control action in a chaotic dynamical system. The Observing Systems Simulation Experiment (OSSE) is performed using the Cellular Neural Network (CeNN) as the data assimilation operator with the Lorenz-63 system. The proposed CeNN-based assimilation achieved control performance comparable to or better than a perturbed observation ensemble Kalman filter, particularly for the more frequent assimilation interval.