Thermal energy storages for cooling applications optimization using Artificial Intelligence algorithm

Panel: 8. Buildings: technologies and systems beyond energy efficiency

Authors:
Jérémie Fricker, ENGIE Lab CYLERGIE, France
Sébatien Sailler, ENGIE Lab Cylergie, France

Abstract

The use of energy for space cooling is steadily and globally growing. Cooling needs tripled since 1990 and at the same time greenhouse gas emission due to cooling systems tripled as well. The need for more efficient cooling systems become crucial, and the role of District cooling networks become more complex. They must provide the cooling capacity to meet each client cooling need, while reducing the operational costs, energy consumptions and CO2 emissions.

Thermal Energy Storage (TES) like chilled water storage or ice storage is regularly mentioned as a way to improve safety and generate operational savings. These potential savings could be achieved if the TES is well designed and managed. The aim of this work is to explain how modeling, optimization algorithms and Artificial Intelligence (AI) could help further than current automated methods to get the maximum benefits of our TES.

Designing a TES – which means choosing its technology, size and control strategy to get high energy and economy efficiencies – is a complex task. Indeed, in addition of space and budget constraints designing calculations must consider simultaneously: electricity prices, equipment’s performances, cooling load and TES production strategy. Thanks to energy balance models, equipment models, and production strategy scenarios, our algorithms test all the combinations over technologies, sizes and control strategies. The results of these high-speed simulations (1-year simulation time ~ 7 minutes of computation time) are displayed in graphs to compare them and finally to assist in decision-making. To go further, an Optimization Algorithm is implemented to determine the best daily charge and discharge curve of the TES with the aim to minimize operational costs, electricity consumption. These computations can be realized with past cooling demand profile to evaluate additional savings or with the future one to do a daily management of the TES.

That is the reason why we implemented a cooling demand profile forecast in our algorithms. In practical terms, a Long Short-Term Memory networks (LSTM), which are a special kind of recurrent neural network capable of learning long-term dependencies, is automatically generated for each dataset provided. Our results show that the cooling demand forecast is closed to real data, thus a daily optimal curve can be forecasted and followed by operators.

These algorithms answer to issues of finding out the best design and production strategy for TES. Even if R&D improvements are possible by getting better accuracy or reducing calculation time, the current real challenge is to integrate these methods into our operational processes. In the future, these features of optimization and demand forecast, could be directly added as an automation tool to do predictive management of our cooling plants as a decision-making tool for operators.