Researchers Validate Encrypted Model Predictive Control for Secure Cloud Deployment
Polynomial approximation unlocks fully homomorphic encrypted MPC with proven stability.
Researchers from Slovakia have experimentally validated an encrypted model predictive control (MPC) approach that addresses critical gaps in secure process control. Traditional MPC relies on solving optimization problems in real time, but outsourcing that computation to cloud platforms risks exposing sensitive system data and controller coefficients. The team's breakthrough uses a polynomial approximation of the explicit control law, which can be evaluated entirely within a fully homomorphic encryption (FHE) scheme. This allows the controller to run on untrusted third-party hardware without ever decrypting the underlying data or model parameters.
The method was tested on a laboratory-scale setup, confirming that the encrypted controller preserves both closed-loop stability and recursive feasibility—properties that previous encrypted MPC implementations failed to guarantee. The polynomial approach avoids the computational overhead of directly solving encrypted optimization problems, making it practical for real-time applications. This work bridges the gap between theoretical FHE-based control and real-world deployment, paving the way for secure, cloud-native automation in critical infrastructure, manufacturing, and autonomous systems.
- Polynomial approximation of the optimal control law enables encrypted evaluation within fully homomorphic encryption (FHE).
- Lab-scale experiments validated closed-loop stability and recursive feasibility for the first time in encrypted MPC.
- Both process data and controller coefficients remain encrypted on third-party or cloud platforms, ensuring end-to-end privacy.
Why It Matters
Encrypted MPC enables secure, cloud-based control of critical infrastructure without exposing sensitive operational data.