Large Language Model-Assisted Parameter Prediction in Integrated Energy Systems

Accurate parameter predictions contribute to the operation optimization of integrated energy systems. Existing prediction methods have some limitations in fully capturing the complex temporal dependencies inherent in parameter patterns. To overcome such limitations, this paper investigates the problem of parameter prediction in integrated energy systems based on large language models (LLMs), which show the powerful logical reasoning capabilities when dealing with complex time series problems. To be specific, this paper utilizes Time-LLM framework to predict electric load, heat load, cold load, and photovoltaic generation. Note that the utilized framework facilitates the reprogramming of the LLMs for time series prediction while preserving the integrity of the backbone language models. Numerical results demonstrate that the proposed prediction method can achieve higher accuracy by 5.7% to 83.6% compared with existing machine learning based benchmarks.