Privacy Preserving Approximated Optimal Control of Pasteurization Unit Using Homomorphic Encryption

Homomorphic encryption (HE) enables computations to be carried out over
encrypted data, ensuring data privacy during processing. This feature is lucrative in various
fields, including general process control and network control systems. However, there are some
limitations to the applicability of modern noisy HE schemes, especially for computationally more
challenging control techniques, such as Model Predictive Control. One of the solutions to this
problem is to use homomorphically encrypted neural networks trained to mimic the behavior of
MPC. This paper showcases such a neural network and demonstrates its capabilities for data
privacy preserving control of the laboratory pasteurization unit.