A method for synthesizing index policies to ensure the survivability of a mobile platform-based information system

Abstract

This paper proposes a method for synthesizing index policies to ensure the survivability of a mobile platform-based information system operating under intermittent connectivity and incomplete observability. The method aims to maintain the probability of violations within specified budgets by combining event-based index ranking with calibrated risk management and dual-loop parameter tuning. The fast loop adaptively adjusts the weight of the risk component of the index based on the magnitude of recorded excesses, while the slow loop periodically recalibrates feature normalization, base weights, and relevance factors, thereby enhancing robustness to non-stationarity and small sample sizes. Probabilistic constraints are interpreted through upper confidence bounds on a short sliding window, enabling controlled risk without rigid assumptions regarding disturbance distributions and without reliance on continuous telemetry or frequent global coordination. To maintain decision consistency, a computationally lightweight acceptability check and event logging are employed, facilitating independent reproducibility. The study presents the index rule formulation—including feature normalization, weight structure, and risk testing—along with a risk correction protocol for sampling per decision step and a procedure for detecting mode changes and safe reconfiguration. The computational complexity of the method is amortized linear-logarithmic with respect to the number of active objects and step capacity, ensuring suitability for real-time operation. Verification was performed on event logs featuring intermittent connectivity and surge load scenarios. The results demonstrate the convergence of the violation frequency upper bound to specified budgets, the stabilization of the risk component weight without fluctuations, and predictable service metric dynamics. A step-by-step ablation study highlighted the contribution of each component: disabling the fast loop delays the achievement of target risk levels, while the absence of the slow loop locks the risk component at an excessive weight, degrading the operational balance. Compared to heuristic rules lacking explicit risk control and model-predictive approaches sensitive to telemetry quality, the proposed method ensures the survivability of the information system through controllable probabilistic guarantees, event scalability, and reproducible verification procedures.