Concrete crack opening forecasting by back propagation neural network and differential equation

Area of Science:

• Hydraulic Engineering
• Computational Mechanics
• Environmental Monitoring

Background:

• Concrete crack opening (CCO) is critical for hydraulic structure maintenance.
• Accurate CCO forecasting is essential for proactive engineering interventions.

Purpose of the Study:

• To develop and assess a novel forecasting method for daily CCO using BPNN and DE.
• To evaluate the contribution of DE in modeling CCO, contrasting with a sole BPNN approach.

Main Methods:

• Calibrating BPNN models with historical data to obtain residuals.
• Employing particle swarm optimization to calibrate DE coefficients.
• Integrating BPNN and DE (BPNN-DE-2TD) for enhanced CCO prediction.

Main Results:

• Sole BPNN models provided reasonable CCO predictions.
• The BPNN-DE-2TD model significantly improved prediction accuracy (e.g., 37% increase in KGE for JB-3).
• DE effectively modeled seasonal and linear temperature trends, while BPNN captured nonlinear water level and precipitation effects.

Conclusions:

• The combined BPNN-DE approach enhances CCO forecasting accuracy.
• This method offers valuable insights for global CCO prediction in hydraulic engineering.
• The study highlights the synergistic benefits of integrating data-driven and physics-based models.