Neural Surrogate Modeling with Odd-Symmetric Activation for Cyclic Response Prediction and Multiobjective Optimization

A fractional-inspired neural surrogate is developed for cyclic performance prediction and multiobjective exploration of reinforced concrete (RC) bridge piers. Its signed-power pre-activation, with exponent α = 0.8, satisfies h(−z) = −h(z), preserving sign-reversal symmetry locally. Positive response indicators retain their physical definitions; odd symmetry is imposed only on the internal transformation. An experimentally calibrated OpenSees fiber model provides 1000 converged cyclic simulations generated by Latin hypercube sampling over five design parameters. Calibration errors are 3.60% and 1.16% for positive and negative peak forces and 7.46% for cumulative energy. Under five-fold cross-validation, the neural surrogate exceeds random forest, XGBoost, and gradient boosting regression in reported predictive accuracy. The respective R2 values for initial stiffness, yield displacement, peak lateral resistance, ultimate displacement, unloading stiffness, and cumulative hysteretic energy are 0.9944, 0.9955, 0.9953, 0.9726, 0.9486, and 0.9973. TreeSHAP analysis of tree surrogates assigns 41.6% of summed mean absolute peak-resistance attribution to shear span ratio and identifies opposing axial load and confinement associations with ductility. NSGA-II yields representative cost–ductility configurations of 0.77, 1.03, and 1.31 kRMB with predicted ductilities of 4.94, 8.30, and 11.35, respectively. These results connect activation-level symmetry with surrogate-assisted exploration of cyclic pier behavior within the sampled domain. Methodologically, the scientific novelty of FANN stems from embedding a sublinear power transformation into the pre-activation mapping, whose derivative dampens extreme gradient spikes during multi-target fitting of quantities spanning diverse physical scales. The mathematical odd symmetry of the activation strictly enforces sign-reversal parity invariance under cyclic reversals, which is decoupled from, yet synergistic with, the sublinear scaling that governs global generalization.

Authors

Institutions

Publication Details

Journal
Symmetry
Published
2026-11-16
DOI
https://doi.org/10.3390/sym18111916
Primary Topic
Seismic Performance and Analysis
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Neural Surrogate Modeling with Odd-Symmetric Activation for Cyclic Response Prediction and Multiobjective Optimization

Hao Li, Yaqing Gu, Mingyuan Hu, Lei Zhang et al.
2 links
Symmetry
Seismic Performance and Analysis
article

Neural Surrogate Modeling with Odd-Symmetric Activation for Cyclic Response Prediction and Multiobjective Optimization

Hao Li, Yaqing Gu, Mingyuan Hu, Lei Zhang, Jingbo Hu, Zhiliang An, Rixia Wang, Sheng Li
article en

Abstract

A fractional-inspired neural surrogate is developed for cyclic performance prediction and multiobjective exploration of reinforced concrete (RC) bridge piers. Its signed-power pre-activation, with exponent α = 0.8, satisfies h(−z) = −h(z), preserving sign-reversal symmetry locally. Positive response indicators retain their physical definitions; odd symmetry is imposed only on the internal transformation. An experimentally calibrated OpenSees fiber model provides 1000 converged cyclic simulations generated by Latin hypercube sampling over five design parameters. Calibration errors are 3.60% and 1.16% for positive and negative peak forces and 7.46% for cumulative energy. Under five-fold cross-validation, the neural surrogate exceeds random forest, XGBoost, and gradient boosting regression in reported predictive accuracy. The respective R2 values for initial stiffness, yield displacement, peak lateral resistance, ultimate displacement, unloading stiffness, and cumulative hysteretic energy are 0.9944, 0.9955, 0.9953, 0.9726, 0.9486, and 0.9973. TreeSHAP analysis of tree surrogates assigns 41.6% of summed mean absolute peak-resistance attribution to shear span ratio and identifies opposing axial load and confinement associations with ductility. NSGA-II yields representative cost–ductility configurations of 0.77, 1.03, and 1.31 kRMB with predicted ductilities of 4.94, 8.30, and 11.35, respectively. These results connect activation-level symmetry with surrogate-assisted exploration of cyclic pier behavior within the sampled domain. Methodologically, the scientific novelty of FANN stems from embedding a sublinear power transformation into the pre-activation mapping, whose derivative dampens extreme gradient spikes during multi-target fitting of quantities spanning diverse physical scales. The mathematical odd symmetry of the activation strictly enforces sign-reversal parity invariance under cyclic reversals, which is decoupled from, yet synergistic with, the sublinear scaling that governs global generalization.

SymmetryVol. 18(11)
Shandong University (CN), Beijing Jiaotong University (CN), Beijing Urban Construction Design & Development Group (China) (CN), Shandong Huayu University of Technology (CN), University of Science and Technology Beijing (CN)
Natural Science Foundation of Shandong Province
Industry, innovation and infrastructure
Seismic Performance and Analysis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon