Privacy-Preserving Distributed Optimising Control of Water Supply

Johansen, Simon Smith Melgaard ; Lauridsen, Lau ; Ørum, Fie

4. semester

Electronic Systems, Master

2024

2024-05-29

86

Society depends on critical infrastructure that, due to internet connectivity, faces increasing risk from cyberattacks. This thesis presents a method for designing distributed, optimization‑based controllers that keep local cost functions and constraints private under passive eavesdropping and defend against specific active attacks. The approach distributes a Model Predictive Control (MPC) problem—a controller that looks ahead by solving an optimization problem—using the Alternating Direction Method of Multipliers (ADMM), a technique that splits a large problem into smaller, coordinated subproblems. Privacy is achieved with Secure Multi‑Party Computation (SMC), which lets multiple parties compute a joint result without revealing their private inputs. We emulate a Water Distribution Network in a laboratory and implement PI (proportional‑integral) controllers so the network can be modeled without detailed pump, valve, and pipe dynamics. Future commands to the PI controllers are obtained by solving a non‑convex MPC problem (a difficult optimization with multiple possible solutions) whose cost function represents the system’s electricity bill, subject to system constraints. The optimization is solved in a distributed manner, and communication between stakeholders is protected with Shamir’s Secret Sharing, which splits data into shares that only reveal information when enough shares are combined. The privacy‑preserving controller reaches minima with the same cost as a standard optimizing controller; the trade‑off is longer computation time and more communication.

Samfundet er afhængigt af kritisk infrastruktur, som på grund af internetforbindelser er mere udsat for cyberangreb. Denne afhandling præsenterer en metode til at designe distribuerede, optimeringsbaserede regulatorer, der holder lokale omkostningsfunktioner og begrænsninger private ved passiv aflytning og beskytter mod bestemte aktive angreb. Metoden deler et Model Predictive Control (MPC) problem – en styring, der ser frem i tiden ved at løse et optimeringsproblem – ved hjælp af Alternating Direction Method of Multipliers (ADMM), en teknik til at opdele et stort problem i mindre dele, der kan koordineres. Privatliv sikres med Secure Multi‑Party Computation (SMC), hvor flere parter beregner et fælles resultat uden at afsløre deres input. Vi emulerer et vanddistributionsnet i et laboratorium og implementerer PI‑regulatorer (proportional‑integral), så netværket kan modelleres uden detaljer om pumpe‑, ventil‑ og rørdynamik. Fremtidige styresignaler til PI‑regulatorerne findes ved at optimere et ikke‑konvekst MPC‑problem (et vanskeligt optimeringsproblem med flere mulige løsninger), hvor omkostningsfunktionen svarer til systemets elregning, under systembegrænsninger. Optimeringen udføres distribueret, og kommunikationen mellem interessenter beskyttes med Shamirs Secret Sharing, som fordeler data i andele, der først kan genskabes, når tilstrækkeligt mange andele samles. Den privatlivsbevarende regulator opnår minima med samme omkostning som en almindelig optimerende regulator; afvejningen er længere beregningstid og øget kommunikation.

[This apstract has been rewritten with the help of AI based on the project's original abstract]

ADMM ; MPC ; SMPC ; Shamirs Secret sharing ; Secret sharing ; Consensus ; consensus ADMM ; Alternating Direction Method of Multipliers ; Secure Multiparty Computation ; Multiparty Computation ; Model Predictive Control ; Predictive control ; Optimising Control ; Water Supply ; SWIL ; Water Distribution Network ; Cyber Security ; Privacy preservation ; Distributed Control

Download PDF
View record in AAU Student Projects