A comparative study of multi-agent control approaches for optimization of central cooling systems without significant storage

Abstract

This paper presents the application of a multi-agent control methodology to large chiller plants. The approach, originally conceived to automate controller design and reduce engineering costs in building energy systems, consisted of a multi-agent simulation framework with distributed consensus-based optimization algorithms. To adapt the approach to this scenario, agents representing physical components of a cooling plant were developed and incorporated in the framework along with two alternative optimization methods: centralized, parallel optimization with a genetic algorithm (GA), and a combination of the GA with a quasi-newton method to handle non-linear equality constraints associated to physical component behavior. An existing cooling plant was utilized as case study to simulate the performance of the methods under different operating conditions. The results demonstrated the difficulty of the consensus-based algorithms to find optimal solutions. The GAs, on the other hand, showed that significant energy savings can be achieved through the implementation of multi-agent control with algorithms capable of handling non-convex objective functions and a combination of discrete and continuous variables, which are characteristic of central cooling systems.