Work

D5.3 Design of the forecasting modes

Year: 2025

Type: article

Abstract: Deliverable D5.3 presents the design and validation of forecasting modes for Distributed Energy Resource (DER) generation and energy demand within the MiniStor Smart Home Energy Management System (SHEMS). For generation forecasting, Multi-Layer Perceptron (MLP) Artificial Neural Networks (ANNs) were trained to predict irradiance, photovoltaic (PV) electrical output and solar thermal power at the five European demo sites, achieving improved accuracy compared to raw meteorological forecasts while also highlighting limitations in complex climatic or operational contexts. For demand forecasting, a range of Machine Learning (ML) models was benchmarked, showing that ensemble methods such as Extra Trees (ETs) and Voting Regressor (VotRegr) provided robust performance, while Deep Learning (DL) and Transformer models showed potential but required further tuning. The forecasting modes developed here have been integrated into the SHEMS alongside the control and optimization strategies and IoT platform defined in deliverables D5.2 and D5.5, supporting MiniStor’s objectives of maximizing renewable self-consumption, reducing costs and ensuring user comfort.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.17152394