Joint Optimization of Mission Abort and Learning Strategies in Cloud Computing Environments

When operating in dynamic uncertain environments, safety-critical systems are often threatened by unknown stochastic shock damages, where traditional risk control strategies with known model parameters struggle to address the coupled challenges of parameter uncertainty and dynamic risks. This paper proposes an adaptive first learning and then optimize framework based on Bayesian inference, which enhances system survivability by jointly optimizing learning duration and risk-sensitive termination decisions. First, a learning duration is set and Bayesian parameter learning strategy is developed to address the unknown shock arrival rate, enabling online learning of shock rates through observed number of shocks. Second, the mission abort decision is optimized based on the updated parameter distribution. Furthermore, a multi-objective optimization model that integrates observation costs, failure losses, and termination penalties is established to determine the optimal observation time. The quantitative trade-off between information acquisition accuracy and risk exposure duration is studied. Theoretical analysis demonstrates that the proposed adaptive strategy dynamically balances the exploration-exploitation dilemma through the synergy of Bayesian sequential updating and risk-sensitive decision-making. This study provides a novel methodological framework for reliable operation of safety-critical systems such as drones and industrial robots in unknown environments.