Operation Optimization of Liquid Cooling Systems in Data Centers by the Heat Current Method and Artificial Neural Network

Abstract

Liquid cooling systems in data centers have been attracting more attentions due to its better cooling capability and less energy consumption. In order to propose an effective optimization method for the operation of indirect liquid cooling systems, this paper first constructs an experiment platform and applies the heat current method to build the global heat transfer constraints of the whole system. Particularly, the thermal conductance of each heat exchanger under different working conditions is predicted by the Artificial Neural Networks (ANN) trained by the historical data. On this basis, combining the heat transfer and fluid flow constraints together with the Lagrange multiplier method builds the optimization model with the objective of minimum pumping power consumption (PPC), solving which by the solution strategy designed obtains the optimal frequencies of the variable frequency pumps (VFPs). Operating with the optimal and other feasible operating conditions validates the optimization model. Meanwhile, the experiments with variable heat loads and flow resistances provide some guidelines for the optimal system operation. For instance, to address heat load increase of a branch, it needs to increase the frequencies of the VFPs, not only the corresponding hot loop but also the whole cold loop.