A Multi-Agent Control Approach for Optimization of Central Cooling Plants
This research focuses on the application of a multi-agent control approach to optimal supervisory control of central cooling systems. Most of the research related to supervisory control of central cooling systems has focused on centralized control approaches. Although these approaches can be effective at reducing operational costs, the high initial costs associated with site-specific controller design and implementation, and the need to update the plant model and control sequences every time a modification is made, have prevented a greater penetration of these technologies in the market. In light of these limitations, the goal of this research is the development and extensive evaluation of a multi-agent control approach that could provide a more economic and easy to configure solution for optimal control of central cooling plants. The work starts from a multi-agent control simulation framework developed by Cai (2015) for optimization of distributed air-conditioning systems. In this setting, the multi-agent structure and optimization-based control algorithm can be automatically generated for each particular application after some pre-configuration. Although the proposed framework should reduce site-specific engineering and provides good flexibility in control topology design, the distributed optimization algorithms included in the framework are not suitable for handling non-convex functions and discontinuous control variables, such as the multiple operating modes in a large cooling system. To adapt the framework to this problem, several tasks were accomplished: (1) a genetic algorithm solution approach was developed and added to the framework to provide an alternative for finding the global optimal operating point in the presence of non-convex functions and discontinuous design spaces, such as the ones that characterize the problem of optimization of large central cooling systems, (2) agents representing the performance of the physical devices of a cooling plant were developed and inserted in the framework, (3) generalized heuristic rules for sequencing and loading plant equipment were incorporated in the framework to reduce the number of control variables making the approach less computationally intensive, (4) a near-optimal control strategy for control of chilled water storage systems subject to dynamic electricity rates and demand charges was developed. The strategy can be implemented within a micro-processor controller which can work in conjunction with the multi-agent framework for optimization of cooling plants, reducing the complexity of the optimization problem. The Purdue Northwest Chiller Plant, a system with a significant degree of complexity, was utilized as the test facility to conduct an extensive computational simulation of the approach for different operating conditions, including chilled water storage subject to dynamic electricity rates with demand charges. The results were evaluated in terms of optimality and processing time requirements. Comparison with other benchmarks such as heuristic control schemes demonstrated that significant savings can be achieved through the implementation of multi-agent control for central cooling plants.
Horton, Purdue University.
Off-Campus Purdue Users:
To access this dissertation, please log in to our