A simulation-based Prediction Framework for stochastic System Dynamic Risk Management.

We propose a simulation-based prediction framework which can quantify the prediction uncertainty of system future response and further guide operational decisions for complex stochastic systems. Specifically, by exploring the underlying generative process of real-world data streams, we first develop a nonparametric input model which can capture the important properties, including non-stationarity, skewness, componentwise and time dependence. It can improve the prediction accuracy, and the posterior predictive distribution can quantify the prediction uncertainty accounting for both input and stochastic uncertainties. Then, we propose the simulation-based prediction framework which can efficiently search for the optimal operational decisions hedging against the prediction uncertainty and minimizing the expected cost occurring in the planning horizon. The empirical study demonstrates that our approach has promising performance.