D5.4 Methodology for farm based multisource energy management systems development

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:Provides a practical AI methodology for multi-vector energy forecasting and optimization in agricultural settings, with comparative model evaluation.

🏢実務担当者:Farm operators can use the HarvRESt dashboard for real-time energy management, improving self-consumption and reducing costs.

🏛政策担当者:Illustrates how AI can support agricultural decarbonization, offering insights for designing subsidy schemes or reporting frameworks.

This report describes the development of the HarvRESt cloud platform and the HarvRESt Dashboard — the software deliverable of Task 5.4 in Work Package 5. The deliverable in hand covers platform design, AI model training and validation, and the first operational demonstration of the energy management interface across three HarvRESt demo sites: Viñas del Vero (VdV, Spain), Fattoria Solidale del Circeo (FSDC, Italy), and Grønn Gårdsenergi (GGE, Norway). The HarvRESt cloud platform is organised in four layers. The Data Collection Layer ingests site data from three upstream sources: the VdV centralised platform in Spain, the data from various multi-vector sources at the Norwegian pilot that can be potentially integrated with the ELEXIA platform, and direct on-site monitoring exports for the remaining sites. The Data Pre-processing and Quality Layer standardises, aligns, and validates the incoming data. The AI Analytics Layer hosts five categories of pre-trained models: PV generation forecasting, wind generation forecasting, biogas estimation, energy demand forecasting, and battery storage flexibility simulation. The HarvRESt Dashboard presents the analytics outputs to farm operators through a web interface accessible to non-technical users. The AI models are trained and evaluated on site-specific datasets. PV generation forecasting achieves explained variance scores between 0.82 (FSDC, Italy) and 0.92 (VdV, Spain). Demand forecasting ranges between 0.37 (VdV, Spain) and 0.89 (FSDC, Italy). Wind generation forecasting, operational for the Norwegian GGE site only, achieves an explained variance of 0.79. LightGBM is adopted as the primary production algorithm across all sites, with XGBoost evaluated as an alternative for the Norwegian site. The Self-Consumption Maximisation Engine applies a priority-based dispatch rule to 48-hour generation and demand forecasts, producing battery charge/discharge schedules and five KPIs for each site: Self-Consumption Ratio, Self-Sufficiency Ratio, Energy Savings (kWh), Estimated Savings (EUR), and Avoided CO₂ (kg CO₂eq). The HarvRESt Dashboard presents these KPIs alongside a generation-demand timeline, a battery dispatch panel, and plain-language “Next Action” guidance messages. Pre-configured user accounts per active farm provide role-based access to site-specific data. Chapter 7 documents the multi-vector energy systems advisor. The decision-support tool allows farmers to evaluate and simulate combinations of renewable energy technologies (solar PV, vertical-axis wind, hydropower, biogas CHP, wood-chip boilers, and energy storage) through a web-based interface, prior to making investment decisions. Together with the HarvRESt Dashboard, it constitutes a complete farmer-facing toolchain spanning pre-deployment RES planning through to day-to-day operational energy management.

• Zenodo https://zenodo.org/records/21275220first seen 2026-07-10 04:27:14 · last seen 2026-07-10 04:29:19