Coding-free virtual flowmeter for building chilled water using pump VFD data and LLM

With the development of artificial intelligence technologies, large language models (LLMs) have been adopted by an increasing number of researchers as assistants. With additional extensions, such as Wolfram Alpha, web search engine and code interpreters, LLMs are promising to enable data-driven modelling without manual coding. In building HVAC (heating, ventilation and air-conditioning) systems, chilled water flow rate is a critical variable for real-time optimal control of HVAC systems. Typically, the flow rate is measured by an on-site flowmeter. Due to the vulnerability of flowmeters, various methods have been developed to predict the real-time flow rate with other monitored variables (i.e., virtual flowmeter), as a supplement to the flowmeter reading. To develop a virtual flowmeter with minimal manual coding effort, this study proposes an approach based on LLMs and pump variable frequency drive (VFD) data to model and predict chilled water flow rate. Three conventional modelling methods were manually implemented for comparison. When predicting chilled water flow rate using the proposed approach (with Doubao), the root mean square error (RMSE) is below 9.05 m³/h, which is slightly higher than the result of artificial random forest method (8.70 m³/h), but lower than the results of artificial physical model (11.28 m³/h) and artificial KNN (15.69 m³/h). Hence, it is verified that the accuracy performance of the proposed approach is comparable to conventional methods, within the acceptable criteria. Moreover, the proposed approach outperformed the conventional methods on reducing manual labor on coding.