Advancing climate-resilient flood mitigation: Utilizing transformer-LSTM for water level forecasting at pumping...

Area of Science:

• Environmental Science and Engineering
• Artificial Intelligence in Water Management
• Climate Change Adaptation

Background:

• Climate change necessitates proactive flood management strategies.
• Intelligent operation of pumping stations relies on accurate water level forecasting.
• Artificial intelligence (AI) offers potential for advanced flood mitigation.

Purpose of the Study:

• To propose a novel Transformer-LSTM model for accurate multi-step-ahead water level forecasting.
• To enhance the intelligent operation of pumping stations for flood management.
• To analyze the interconnected dynamics influencing flood events.

Main Methods:

• Utilized a Transformer-LSTM model integrating Transformer and LSTM modules.
• Collected and processed 19,647 ten-minute-based datasets (2014-2020) for training, validation, and testing.
• Evaluated model performance against benchmark models for water level prediction.

Main Results:

• The Transformer-LSTM model significantly outperformed benchmark models in water level forecasting accuracy.
• Achieved reliable forecasts at 10-minute to 60-minute horizons (T+1 to T+6).
• Demonstrated effective enhancement of input factor connections and time series connectivity.

Conclusions:

• The proposed model improves flood management through accurate water level predictions.
• Understanding interconnected dynamics is crucial for climate impact resilience and infrastructure.
• The Transformer-LSTM model supports water practices, resilience, and disaster risk reduction.