Water demand forecasting in multiple district metered areas based on a multi-scale correction module neural network...

Area of Science:

• Hydrology and Water Resources Management
• Artificial Intelligence in Environmental Science
• Urban Water Systems Engineering

Background:

• Accurate short-term water demand forecasting (STWDF) is crucial for efficient urban water supply network management.
• Forecasting demand for individual District Metering Areas (DMAs) presents greater uncertainty than aggregate demand.
• Existing models often struggle with the spatial and temporal correlations inherent in multi-DMA water usage.

Purpose of the Study:

• To develop an innovative neural network architecture for simultaneous STWDF across multiple DMAs.
• To enhance prediction accuracy by incorporating multivariate corrections and inter-DMA correlations.
• To provide a unified framework for simplifying multi-DMA STWDF and improving urban water management strategies.

Main Methods:

• Development of a multi-scale correction module neural network combining Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN) with Attention mechanisms.
• Application of the model for hourly STWDF over a week for 10 DMAs in northern Italy.
• Utilizing multivariate corrections and analyzing the impact of inter-DMA correlations and meteorological features.

Main Results:

• The proposed model demonstrated an average performance improvement of 5%-20% across evaluation metrics compared to traditional Gated Recurrent Unit (GRU) or LSTM models.
• Superior accuracy was consistently achieved across single DMA, total water demand, and extreme condition forecasting scenarios.
• Interpretability analysis confirmed the model's feasibility and highlighted the predictive contribution of meteorological data for specific DMAs.

Conclusions:

• The novel multi-scale correction module neural network offers a significant advancement in STWDF for urban water networks.
• Leveraging inter-DMA correlations and multivariate corrections is key to improving forecasting accuracy and managing water resources effectively.
• The unified input-output framework simplifies multi-DMA STWDF, offering valuable insights for future research and practical applications.