Enhancing greenhouse management with interpretable AI: A natural language interface for advanced and optimization-based control

As climate change and resource scarcity threaten global food security, greenhouse systems are becoming critical for sustainable agriculture. Advanced control, such as Model Predictive Control (MPC), effectively regulates temperature, humidity, and CO 2 to enhance crop growth and resource efficiency. However, the widespread adoption of such advanced control systems is limited by their lack of interpretability, as growers struggle to understand complex control decisions, particularly during rapid environmental changes. In this work, a Natural Language Generation (NLG) interface has been developed to bridge this gap and transform MPC control decisions into clear, actionable explanations for greenhouse growers. This interface integrates Large Language Models (LLMs) with greenhouse control systems and mathematical constraints, providing a step toward making AI-driven agriculture more accessible. This integration addresses the need for interpretable AI systems in modern agricultural applications. The system allows growers to interact with the control system, query decisions, and receive contextually relevant explanations through Retrieval Augmented Generation (RAG) mechanisms and instruction prompting techniques. The Adaptive RAG (ARAG) framework was evaluated using semantic similarity, information retrieval, and contextual relevance metrics, demonstrating a 12.1% improvement in BERTScore, over baseline methods. These results highlight the system's ability to deliver accurate, well-structured explanations without compromising control performance. By improving the interpretability and accessibility of AI-powered greenhouse automation, this research advances the development of sustainable greenhouse practices that can adapt to the challenges of climate change and resource scarcity. The proposed system represents a significant step toward transforming traditional greenhouse control into more interpretable solutions for modern agriculture.