Optimal configuration and control strategy in a multi-carrier-energy system using reinforcement learning methods

The efficient integration of renewable energy in the present infrastructure requires the use of storage technologies and the inclusion of other energy sectors as electricity and natural gas to use synergistic effects. In this paper we present an approach to find an optimal solution using Reinforcement Learning (RL) methods. We use the concept of “Energy Hub” (EH) and “Smart Grid” to improve the operation of multi-carrier energy systems. The RL agent finds the optimal configuration of the EH (the combination of gas-turbines, furnaces, transformers and storage devices) together with the optimal control strategy of these devices to minimize the total device and unit costs while satisfying the customers' electrical and heat loads. This problem is setup as a continuous control problem, different RL algorithms are benchmarked using the RLLab framework. We were able to find and validate a set of optimal configurations and corresponding control strategies. The presented method can be used with arbitrary energy hub configurations and load profiles. The provided solution is able to adapt the control over time and respond to changes in the load profiles.