Interval neural network for structural distributed load identification under uncertainty

Accurate characterization of loading conditions is essential for ensuring the safety and reliability of modern engineering structures. In real scenarios, uncertainties in both environmental inputs and structural properties lead to variability in measured data, making uncertainty-aware load identification crucial. To address this challenge, this paper develops an Interval Backpropagation Neural Network (IBPNN) for distributed load identification. By introducing interval analysis into a conventional Backpropagation Neural Network (BPNN), the model can represent uncertainty in outputs. Considering spatial variations of loads and the continuity of physical fields, the target load is reconstructed using basis functions through modal decomposition. Under uncertain conditions, the proposed IBPNN identifies credible interval ranges of distributed loads. Furthermore, an improved interval prediction quality metric, the Improved Coverage Width Based Criterion (ICWC), is designed to balance coverage and interval width while ensuring consistency between the interval median and true parameter values. Finally, three structural case studies and an experimental example verify the feasibility and effectiveness of the proposed approach.