In our latest news article, Dr Tudor Cioara of the Technical University of Cluj-Napoca in Romania provides an insight into CATALYST Flexibility Prediction and Optimisation.
“In CATALYST we are aiming to transform the Data Centres in active energy players that are operated at the crossroad of electrical, thermal, and data networks and can leverage onto their internal electrical and thermal flexibility to provide energy services to the local grid. In this way the Data Centres will not only contribute to the local energy grid resilience and decrease their carbon footprint, but they will gain new revenue streams on the top of provided services not foreseen before.
To make this vision possible an important component is the Flexibility Prediction and Optimization software that will be deployed in the Data Centre and will manage its electrical and thermal flexibility without endangering its safe operation.
Digital Twin model of a Data Center
The CATALYST flexibility optimization relies on a system of systems enabled thermal and electrical energy model which constitutes the Digital Twin (DT) representation of the Data Centre. It models the operation of various Data Centre components as sub-systems integrated through links of energy streams and a set of flexibility control actions and variables were defined and used to estimate the electrical energy flexibility potential of each component using mathematical functions. The following subsystems are considered for providing energy flexibility: the IT servers by shifting in time the execution of delay-tolerant workload, the cooling system by dynamic usage of non-electrical cooling systems to pre-cool or post-cool the Data Centre and to compensate the electrical one and finally the batteries systems by reducing/increasing the DC energy demand with the amount of energy discharged/charged from batteries.
The server room, cooling system, and heat pump are the main sub-system considered modelled from the perspective of thermal energy flexibility. The server room contains the servers that receive as input the workload to be executed and generates heat that must be dissipated by the cooling system to maintain the temperature setpoints for the IT equipment safe operation. By smart allocation of workload and changing the temperature set points in the server room the thermal flexibility associated is determined and eventually exploited. The cooling system can be used to modify the airflow in the server room and the air temperature to maximize the heat reuse capability of the DC, while the heat pump’s cold and hot water tanks can be used as buffers to vary the heat quantity and quality sent to nearby neighbourhoods.
More details on the CATALYST Data Centre electrical and thermal model and mathematics behind it can be found in .
Digital Twin enabled Data Centre predictive analytics
The constructed Data Centre Digital Twin model is used to conduct data-driven simulations for estimating the Data Centre energy demand baseline and energy flexibility potential and supporting the predictive control of Data Centre operation. For short term forecasting of Data Centre’s energy demand and production deep neural network-based models have been used. They consider energy-related features determined based on historical data acquired by DC energy meters and contextual features that are not specific to energy but are correlated to context, such as seasons, weekdays, calendar days, etc. Due to the fact that is difficult to anticipate the precise thermal behaviour of the server room and potential hot spot formation while temperature setpoints are increased, we have defined and used Computational Fluid Dynamics simulations of the thermal processes of the server room, considering the individual server racks, their actual load, as well as the cooling system operation parameters, such as input airflow and air temperature. To avoid the thermal simulations computational overhead at run-time their results are fed to the neural network models to predict the temperature distribution in the server room and the amount of heat recovered.
Data Centre flexibility management optimization
The problem of optimizing the DC operation to shift energy flexibility to meet different energy grid objectives had been modelled as a multi-criteria optimization problem in which we aim to determine a flexibility shifting action plan that minimizes the distance to the objective while at the same time meets all the constraints for DC systems safe operation. As energy service objectives we have considered (i) the electrical energy trading by shifting energy flexibility to allow it to sell energy when the prices are high and buy energy when the prices are low in the electricity market, (ii) delivery of flexibility services by shifting energy flexibility to match as close as possible a flexibility order provided by a flexibility aggregator via the flexibility marketplace and (iii) heat trading by shifting energy flexibility increase as much as possible the quality of recovered heat and sell it on the heat marketplace directly of via a Heat Broker for example. The DC operator is empowered to define its custom DC optimization strategy by defining weights for each specific objective allowing the simultaneous delivery of more than one service at the same time. The DC flexibility management optimization is a Non-Linear optimization problem (an NP-hard one) because the unknowns are real, and the defined objective functions are nonlinear. To solve it we have defined hybrid heuristics that are combining population-based and gradient-based optimization.
 M. Antal, et al., A System of Systems approach for data centers optimization and integration into smart energy grids, Future Generation Computer Systems, Volume 105, 2020, Pages 948-963, ISSN 0167-739X, https://doi.org/10.1016/j.future.2017.05.021.
 Vesa, A.V.; Cioara, T, et al. Energy Flexibility Prediction for Data Center Engagement in Demand Response Programs. Sustainability 2020, 12, 1417, https://www.mdpi.com/2071-1050/12/4/1417
 Antal, M.; Cioara, T.; Anghel, I.; Gorzenski, R.; Januszewski, R.; Oleksiak, A.; Piatek, W.; Pop, C.; Salomie, I.; Szeliga, W. Reuse of Data Center Waste Heat in Nearby Neighborhoods: A Neural Networks-Based Prediction Model. Energies 2019, 12, 814, https://www.mdpi.com/1996-1073/12/5/814
 T.Cioara, et al., Exploiting data centres energy flexibility in smart cities: Business scenarios, Information Sciences, Volume 476, 2019, Pages 392-412, ISSN 0020-0255, https://doi.org/10.1016/j.ins.2018.07.010.
 T. Cioara, I. Anghel, M. Bertoncini, et al. Optimized flexibility management enacting Data Centres participation in Smart Demand Response programs,Future Generation Computer Systems,Volume 78, Part 1, 2018, Pages 330-342, ISSN 0167-739X, http://www.sciencedirect.com/science/article/pii/S0167739X16301200