METHODOLOGY TO DEVELOP ARTIFICIAL NEURAL NETWORK BASED CONTROL STRATEGIES FOR MULTIPLE AIR-COOLING UNITS IN A RAISED FLOOR DATA CENTER

Abstract

A data center consists of a hierarchy of systems with dedicated control algorithms dictating their operational states. We could say, there exists an ensemble of dynamical systems, each executing a control task, while the global objective is to drive the overall system to an optimum i.e. minimum total operational power at desired rack inlet temperatures. The dynamics of the Information Technology Equipment (ITE) workload and the cooling provisioning is non-linear in spatial and temporal parameter space. Data-driven modelling is one method to realistically model such non-linear dynamics and make predictions that are necessary for improved control design of the cooling system. In this study, the data center non-linear dynamics are approximated by well-defined operational scenarios. Multiple ACU’s are to be optimally controlled in provisioning a varying rack-level workload within the data center. CFD tool 6SigmaRoom is used to model and simulate a raised-floor data center with multiple Air-Cooling Unit (ACU)s. The input parameter space and boundary conditions to be applied to the simulations are sampled on a random basis using the Latin Hypercube Sampling technique. Artificial Neural Network (ANN) is a suitable data driven technique that has the ability to capture the non-linear relationships between the dynamic operation of ACU’s setpoints and the ITE’s workload. Training data is obtained from the results and observations of a large number of parametric CFD simulations. An objective function is defined. ANN model is developed to predict the parameters that are necessary in designing a data-driven control framework.