Artificial neural network analysis for reliability prediction of regional runoff utilization

Area of Science:

• Hydrology and Water Resources Engineering
• Artificial Intelligence in Environmental Management
• Reliability Engineering

Background:

• Planning regional rainwater utilization tank capacity involves complex reliability analysis.
• Historical daily rainfall data is crucial for assessing water supply systems.
• Understanding rainfall frequency, runoff, and water continuity is essential for capacity planning.

Purpose of the Study:

• To analyze regional daily rainfall frequency, runoff amount, water continuity, and reliability.
• To determine suggested designed storage capacity based on demand and supply reliability.
• To simulate and predict the reliability of small area rainfall-runoff supply systems using artificial neural networks.

Main Methods:

• Analysis of historical daily rainfall data for frequency, runoff, continuity, and reliability.
• Development of small area rainfall-runoff supply systems using two artificial neural network models: Radial Basis Function Neural Network (RBFNN) and Back-Propagation Neural Network (BPNN).
• Assessment of model stability and learning speed, comparing RBFNN and BPNN performance.

Main Results:

• The RBFNN model demonstrated superior stability compared to the BPNN model.
• Despite better reliability, RBFNN provided a conservative estimate for actual monitoring data.
• The BPNN model (4-3-1-1 configuration) exhibited a lower error rate than RBFNN.

Conclusions:

• RBFNN offers better stability for rainwater supply system reliability analysis than BPNN.
• BPNN may provide more accurate predictions for actual monitoring data in this context.
• The findings support the future application of intelligent control equipment for instantaneous prediction and management of rainwater utilization systems.