New SDE Framework Outperforms Traditional Models for Marine Engine Health Forecasting

A new arXiv paper from Y. Harsha Vardhana Reddy and Soumyendu Raha addresses the challenge of forecasting marine engine parameters under diverse operational loads and environmental stressors. Traditional time-series models like Vector Autoregressive Integrated Moving Average (VARIMA) often fail due to fixed-window linear assumptions that can't capture the stochasticity and transient dynamics of complex systems. To solve this, the researchers developed a dual-layered estimation approach: first, an adaptive lookback mechanism dynamically adjusts the learning window based on instantaneous drift magnitude, ensuring responsiveness during non-stationary regimes. Second, a multi-particle ensemble is evolved via Euler-Maruyama discretization, where each particle trajectory represents a stochastic realization of the system state.

The refined ensemble mean uses a Girsanov transform to mitigate 'noise-chasing' by assigning higher probabilistic weights to particles aligned with the physical drift. Theoretical evaluation and empirical benchmarking across 10 pages, 9 figures, and 4 tables show the adaptive SDE framework significantly outperforms classical baselines in multi-step prediction stability and computational efficiency. The model offers a scalable, 'grey-box' solution for real-time risk quantification in systems characterized by high-frequency volatility and non-linear transitions, directly improving operational reliability of marine propulsion systems.