Clustered Firefly Algorithm for Global Parameter Estimation in Cyber-Physical Systems: Applications in Wastewater Treatment and Residential Heating
Author
Gebauer, Hannes
Term
4. term
Education
Publication year
2025
Submitted on
2025-06-11
Pages
83
Abstract
Dette speciale introducerer en ny variant af Firefly Algorithm, kaldet Clustered Firefly Algorithm (cFA), der gør parameterestimering i dynamiske modeller til modelprædiktiv regulering (MPC) hurtigere og mere effektiv. Metoden kombinerer den naturinspirerede Firefly Algorithm (en sværmbaseret optimeringsmetode) med k-means-klyngedannelse (en måde at gruppere kandidatløsninger på). Denne kobling reducerer beregningsindsatsen og bevarer en stærk global søgeevne - det vil sige evnen til at udforske bredt og undgå at sidde fast i dårlige lokale løsninger. På standardiserede optimeringsbenchmarks konvergerer cFA hurtigere og mere stabilt til løsninger af høj kvalitet end den oprindelige Firefly Algorithm og andre grundmetoder. Tilgangen afprøves også på to virkelighedsnære cases: parameterestimering i en teoretisk spildevandsmodel (ASM1) og i en termisk bygningsmodel til styring af en varmepumpe. Resultaterne tyder på, at cFA håndterer komplekse, multimodale søgelandskaber effektivt. Trods de lovende resultater anbefales yderligere eksperimentel validering for at vurdere cFA's anvendelighed i forskellige MPC-scenarier.
This thesis introduces a new variant of the Firefly Algorithm, called the Clustered Firefly Algorithm (cFA), to make parameter estimation in dynamic models for Model Predictive Control (MPC) faster and more efficient. The method combines the nature-inspired Firefly Algorithm (a swarm-based optimization method) with k-means clustering (a way to group candidate solutions). This combination reduces computational effort while preserving strong global search ability - meaning it can still explore widely and avoid getting stuck in poor local optima. On standard benchmark optimization problems, cFA converges faster and more consistently to high-quality solutions than the original Firefly Algorithm and other baseline methods. The approach is also tested on two real-world motivated cases: parameter estimation for a theoretical wastewater treatment plant model (ASM1) and a thermal building model used to control a heat pump. The results indicate that cFA handles complex, multimodal search landscapes effectively. Despite these promising findings, further experimental validation is recommended to assess cFA's applicability across diverse MPC scenarios.
[This summary has been rewritten with the help of AI based on the project's original abstract]
Keywords
Documents