REinforcement LEarning and Nonlinear MPC algorithm for the Dis tributed Energy Resources (RELENDER)

The Lab-Access User Project RELENDER (REinforcement LEarning and Nonlinear MPC algo rithm for the Distributed Energy Resources) has been hosted in the SGIL lab in JRC in Petten (NL), from 10 till 16 Nov 2024. The members of the group were Prof. Luca Ferrarini and Dr. Alberto Valentini.

The RELENDER project originally aimed at developing a control and optimal real-time management of energy resources in a smart building, in order to help reduce energy (both thermal and electrical) demand in buildings and also to favor the adoption of renewables as well as to encourage the paradigm shift to use buildings as service providers for the grid. However, some equipment was not available and heat pumps were not (easily) controllable by an optimal controller, the project reduced its aim to test the adoption of optimal scheduling of electrical energy stored in batteries, for which the user group developed a fast data-driven identification algorithm, based on Online Kernel Regression (OKR) techniques, to estimate the behavior of the local battery management system. Several tests were conducted to collect the necessary data to train the OKR algorithm, under different operating conditions, initial charge level, and battery type (electric vehicle and e-bike). Some smart meters can be read remotely through a computer, and data acquired for post-processing. Given the short duration of the lab access, the collected data are under filtering and prepared for the tuning of OKR algorithms.