Flow Prediction of Water Supply Network Based on Time Series Large Language Model

The acceleration of urbanization in China has resulted in a direct impact on the efficiency of the rational use of water resources, as the urban water supply system plays an important role in urban construction. However, the issue of water supply pipe network leakage is becoming increasingly problematic, as it is directly related to the quality of life of urban residents and the smooth progress of industrial production. Therefore, it is crucial to accurately predict flow changes in urban water supply pipe networks and promptly identify instances of leakage. However, traditional deep learning models are not well-suited for effectively capturing long-term dependencies. In this study, we propose a water supply network flow prediction method based on the time series large language model (LLM) Chronos. Chronos tokenizes time series values using scaling and quantization into a fixed vocabulary and trains existing transformer-based language model architectures on these tokenized time series via the cross-entropy loss. In the course of our experiments, we utilized traffic data from the Haida neighborhood for the year 2023 for the purpose of model evaluation and comparison with traditional LSTM and GRU and the extant time series LLMs, MOMENT and Timer. The results of our experiments demonstrate that Chronos outperforms LSTM, GRU, MOMENT and Timer in all evaluation metrics, thereby substantiating its high degree of accuracy and effectiveness in predicting complex time series data.