Optimizing Cooling System Operations with Informed ML and a Digital Twin

Today, there are a variety of cooling systems available to serve smaller data centers, industrial plants or office buildings. These are often one-off installations, and in most cases their control parameters are set at the time of installation and are not changed subsequently. In addition, these parameters are set conservatively and are not designed for energy-optimized operation. A digital twin of the plant, including a simulation model, is essential to bring the cooling system closer to energy-optimized operation over its lifetime, but this is not usually the case. One of the main reasons is that digital twins are expensive and time-consuming to create. However, today’s cooling systems are extensively equipped with sensors, so this information can be used and the effort to create a digital twin is greatly reduced.
This chapter proposes an approach to generate parts of the digital twin for the cooling system from measured data by using ML methods. In a subsequent step, this digital twin is used to calculate the effects of alternative control parameters, and the results are presented to the operator in an understandable way. Combined with appropriate monitoring this allows the operator to make informed decisions to adjust the control parameters accordingly.