Research and application of deep learning techniques for forecasting groundwater levels in the Hanoi area

Abstract

Groundwater is essential for keeping the ecosystem in balance and meeting the water needs of homes, farms, and businesses, especially in big cities like Hanoi, which are strongly affected by climate change and overuse. In the context of climate change and overexploitation, accurate groundwater level forecasts not only help successful water resource management, but also serve as a foundation for long-term development. However, the intricacy, nonlinearity, and long-term dependency of groundwater levels present significant problems for standard approaches. Ours provides a novel approach for predicting groundwater levels at monitoring station Q64 in the Hanoi area using four deep learning models: RNN, LSTM, Transformer and Autoformer. We forecast short-term (48 hours), medium-term (120 hours) and long-term (360 hours). Experimental results reveal that Autoformer is clearly superior in forecasting situations and performs well in short-term, in the medium term forecasting. This demonstrates that the model based on an attention-based architecture ơan capture long-term properties of groundwater level time series. These findings support the use of deep learning in groundwater level forecasting, paving the way for the creation of intelligent forecasting systems, aiding decision-making in water resource management, and developing ways to adapt to climate change in large cities.