Automated analysis and textual summarization of time-varying references in advanced greenhouse climate control

The growing need for energy-efficient and sustainable crop production has made advanced control systems, such as Model Predictive Control (MPC), essential in greenhouse farming. MPC is an optimization-based control strategy that uses mathematical models and weather forecast data to regulate greenhouse climates effectively. This technique generates time-varying climate reference trajectories, which are sent to the local process computer to control the corresponding climate parameter or equipment. While MPC and artificial intelligence-based techniques are becoming more common in advanced agricultural setups, their widespread adoption remains limited. Potential reasons are the lack of transparency and the understandability of the control algorithms. This study introduces a language-based support system to improve the usability of advanced control strategies like MPC. The system segments time-series data using the change point detection method to identify significant changes. The identified trend information is converted into detailed textual descriptions using the natural language generation technique. These descriptions are refined into user-friendly summaries with the assistance of a pretrained large language model. The results demonstrate that this support system can improve the accessibility and usability of advanced control strategies like MPC, making them more practical for greenhouse growers.