Part of the
4TU+.
NIRICT
4TU+.
NIRICT
Close

4TU.Federation

+31(0)6 48 27 55 61

secretaris@4tu.nl

Website: 4TU.nl

Agentic AI for Energy Management in Horticulture: Integrating Systems Thinking, Electrical Constraints, and Human Oversight

The requested funding will support two interdisciplinary workshops to establish a 4TU research network on the responsible development and adoption of Agentic AI. Greenhouse horticulture in the Netherlands provides a compelling and urgent use case: growers already deploy solar generation, battery storage, and CHP systems to actively manage grid congestion (Hortidaily, 2025; NL Times, 2025), yet these assets remain coordinated through static, rule-based control rather than adaptive intelligent agents. Energy management in these systems is inherently uncertain, driven by weather variability, model uncertainty, and volatile energy markets (Payne, 2024). Agentic AI offers the potential to address this complexity, but must simultaneously manage distributed energy assets, operate across distributed edge devices, respect strict electrical constraints (e.g., voltage limits, battery bounds, and grid codes), and respond to real-time market signals (Bui et al., 2025; Kiasari and Aly, 2026). Specifically, this proposal focuses on system-level grid coordination, ensuring agentic AI complies with TSO-DSO coordination rules (coordination between transmission and distribution system operators) and remains physically feasible without causing congestion or voltage violations. While local optimisation is well understood, system-level coordination with grid operators remains an open challenge. These requirements must also be met under fast control timescales, necessitating low-latency edge AI deployment on distributed edge devices for real-time control and protection. Crucially, no existing agentic AI framework can safely satisfy these requirements in an integrated manner, representing a critical barrier to deployment. Greenhouse energy systems are tightly coupled socio-technical systems, where decisions in one subsystem cascade across physical, digital, and market layers. We propose a research agenda on human–AI collaboration and systems integration for safe agentic AI under electrical constraints.