This study presents an innovative method for predicting coal water supply pipeline leakage aperture sizes. Firstly, an Improved Grey Wolf Optimization (IGWO) algorithm is introduced, which combines Bernoulli chaotic initialization, elliptical convergence parameters, and dynamic weight updates. The algorithm’s optimization capabilities are enhanced by IGWO. Secondly, the effectiveness of IGWO is validated through benchmark tests. Finally, IGWO is applied to optimize the initial structural parameters of a Backpropagation (BP) neural network. The effectiveness of IGWO-BP is verified using laboratory pipeline infrasound leakage signals. The optimized IGWO-BP model demonstrates superior performance in predicting leakage apertures, with a substantial 69.12% reduction in mean absolute percentage error (MAPE) and a 66.96% decrease in mean square error (MSE) compared to traditional BP models, offering valuable insights for future leak remediation efforts.