Conference Year



cooling plants, central cooling, optimization, multi-agent control, distributed control


This paper presents an application of a multi-agent control approach for supervisory control of large central cooling plants. The starting point for this work was a multi-agent control simulation framework developed by Cai (2015).  To adapt the framework to the problem at hand several tasks were accomplished: agents representing the performance of the different devices of the plant were developed and inserted in the framework and generalized heuristics were incorporated to make the approach less computationally intensive. A case study of an existing cooling plant with significant complexity was utilized to conduct an extensive evaluation of the approach in terms of optimality and computational resources. Simulations were carried out using one year of historical data to predict the performance of the plant under three different control strategies: 1) multi-agent control, 2) centralized optimization based on mathematical programming techniques and 3) a heuristic control strategy. The results showed that significant savings can be achieved through the implementation of multi-agent control. It is expected that, if each hardware component of the plant comes with an integrated agent that represents its behavior, then the proposed multi-agent framework could automatically generate the multi-agent structure and control algorithm after some relatively simple pre-configuration steps. This will reduce the site-specific engineering and will provide a more economic and easy to configure solution for central cooling systems.