Novel Approach Based Minimization of Geometric Action for Predicting Rare and Extreme Events in Non-Equilibrium Systems.

Identifying and quantifying unexpected events in non-equilibrium systems is critical work that is necessary for systems managers to make well-informed decisions, particularly when forecasting rare and extreme events. In this paper neural networks are integrated to increase the predictive capacity of information theory. Two information theory techniques, “Information Length (IL) and Information Flow (IF)”, are being examined for their sensitivity to rapid changes. To simulate the first occurrence of extreme and rare events, we utilize a nonautonomous Kramer model to introduce a perturbation. we introduced a Dynamic Osprey Long Short-Term Memory (DOLSTM) for predicting rare and extreme events in non-equilibrium systems. Our results show that IL performs better than IF in accurately forecasting unexpected occurrences when combined with a neural network. This study highlights a novel integration between information theory & neural networks, giving an effective strategy for forecasting rare & extreme events in non-equilibrium environments. An effective methodology for identifying and forecasting the behavior of dynamic systems is established by combining information-length diagnostics with neural network predictions, especially in situations involving rare and extreme events. This novel method illustrates that the theory of information and neural networks can be used to provide robust predictions for dynamic systems, when encountering rare and extreme events. [ABSTRACT FROM AUTHOR]