Predicting Stellar Mass Accretion: An Optimized Echo State Network Approach in Time Series Modeling

Modeling the dynamics of formation and evolution protostellar disks as well history stellar mass accretion typically involve numerical solution complex systems coupled differential equations. The resulting protostars is known to be highly episodic due recurrent instabilities also exhibits short timescale flickering. By leveraging strong predictive abilities neural networks, we extract some critical temporal experienced during including periods instability. Particularly, utilize a novel form Echo-State Neural Network (ESN), which has been shown efficiently deal with data having inherent nonlinearity. We introduce use Optimized-ESN (Opt-ESN) make model-independent time series forecasting rate in disks. apply network multiple hydrodynamic simulations different initial conditions exhibiting variety demonstrate predictability Opt-ESN model. model trained on simulation $\sim 1-2$ Myr, achieves predictions low normalized mean square error ($\sim 10^{-5}$ $10^{-3}$) for forecasts ranging between 100 3800 yr. This result shows promise application machine learning based models time-domain astronomy.