Hybrid ANN ‐optimization methods for seismic collapse probability prediction of reinforced concrete structures

Abstract Reinforced concrete (RC) structures in earthquake‐prone regions require accurate and efficient tools for predicting collapse probability (CP) and improving seismic performance. Although performance‐based earthquake engineering methods are reliable, their high computational cost often limits their practicality for large‐scale or iterative design applications. This study introduces a hybrid surrogate modeling framework that integrates artificial neural networks (ANNs) with Pattern Search (PS) optimization (ANN‐PS) to address these challenges. The method is compared with two alternative metaheuristic approaches—ANN‐Genetic Algorithm (ANN‐GA) and ANN‐Ant Colony Optimization (ANN‐ACO)—to evaluate relative performance. A dataset of over 100 RC archetype structures was generated, and more than 5000 Incremental Dynamic Analyses (IDA) were performed using 22 far‐field ground motions. Key seismic and structural parameters were optimized using the hybrid models. Model performance was assessed using 20 statistical metrics covering accuracy, bias, and correlation, and ranked using scoring. Results show that ANN‐PS achieved the highest predictive accuracy, reaching a minimum collapse‐probability estimate of CP ≈ 0.0100, outperforming ANN‐GA (typical CP = 0.0101–0.0149) and ANN‐ACO, which exhibited significant variability (CP = 0.0119–1.8893). Quantitatively, ANN‐PS matched the ANFIS benchmark within 0.0000–0.0001 absolute error, whereas ANN‐ACO deviated by up to +1.8893. Overall, Pattern Search demonstrated superior stability and consistency, confirming it as the most reliable optimization strategy for ANN‐based seismic collapse prediction of RC structures.