Information Theory-Based Sensor Placement for Data Reconciliation with Uncertain Models
Brijesh Kumar, Mani Bhushan · Industrial & Engineering Chemistry Research · 2025
Data reconciliation adjusts noisy measurements from plants using a known process model. Traditional methods for the data reconciliation problem assume a perfect model. This assumption does not hold in many situations. This work proposes a Symmetric Kullback–Leibler Divergence (SKLD)-based Sensor Placement Design (SPD) for data reconciliation under model uncertainty. The uncertainties are modeled as bounded or stochastic uncertainties for two scenarios: uncertainty only during design or during both design and operation. The resulting integer programming SPD problems are reformulated as computationally tractable problem formulations using semidefinite programming. The approach is demonstrated on two case studies: (i) Go Yang water distribution network and (ii) Steam metering network, which show that neglecting model uncertainty results in suboptimal placements and reduced accuracy of estimates.