Development of an Artificial Neural Network Algorithm Embedded in an On-Site Sensor for Water Level Forecasting.

Area of Science:

• Environmental Science
• Computer Engineering
• Hydrology

Background:

• Extreme weather events frequently cause urban inundation due to stream overflow.
• Traditional centralized flood monitoring systems have limitations in real-time data processing and prediction accuracy.
• There is a need for innovative, efficient flood monitoring solutions for timely urban flood management.

Purpose of the Study:

• To propose a decentralized flood monitoring system for predicting stream water levels up to three hours in advance.
• To develop and evaluate artificial neural network (ANN) models for water level forecasting using edge computing.
• To identify key factors influencing water level variations for improved prediction accuracy.

Main Methods:

• A customized sensor system was developed to measure water levels and implement edge computing for real-time predictions.
• Correlation analysis was performed to determine significant input factors for the ANN models.
• Three artificial neural network (ANN) models were developed and tested using data from four streams, with one case for training/testing and others for validation.

Main Results:

• The ANN model incorporating the second-order water level difference as an input factor demonstrated superior performance.
• This model achieved the lowest Root Mean Square Error (RMSE) compared to other developed ANN models.
• The customized sensor with embedded ANN algorithms proved effective for edge computing in flood monitoring.

Conclusions:

• The developed decentralized flood monitoring system enhances edge computing capabilities for real-time water level prediction.
• The findings support the adoption of microprocessor-based sensors with embedded ANN for improved urban flood forecasting.
• This technology can significantly aid emergency response and decision-making processes during extreme weather events.