Value-of-information-based optimization of grey-box models for computational risk assessment of cyber-physical systems

The complexity of Cyber-Physical Systems (CPSs) requires that their safety be evaluated by Computational Risk Assessment (CRA). The very large computational challenge can be addressed by Grey-Box Models (GBMs), which integrate first-principles White-Box Models (WBMs) and data-driven Black-Box Models (BBMs): different WBMs or BBMs can be used to model each CPS subsystem, which leads to a combinatorial number of possible GBM alternatives that come with different trade-offs between computational burden and physical fidelity. In this work, we propose a Value-of-Information (VoI)-based procedure to identify the best GBM for CPS CRA. We show the procedure by application to an Integrated Power and Telecommunication (IP&TLC) infrastructure: VoI-driven GBM alternatives are evaluated and those that reduce the computational burden while keeping accurate in computing the risk metrics of interest are identified.