Assessing Environmental Performance of Water Infrastructure Based on an Attention-Enhanced Adaptive Neuro-Fuzzy Inference System and a Multi-Objective Optimization Model

This study aims to develop an integrated framework that combines an attention-enhanced adaptive neuro-fuzzy inference system (ANFIS) with multi-objective optimization to address the challenges of subjective indicator weight allocation and insufficient nonlinear relationship modeling in environmental performance evaluation of water infrastructure. Drawing on the tri-dimensional theory of performance evaluation—success, results, and actions—the framework organizes environmental performance indicators into five primary and nine secondary dimensions. Through empirical analysis across China’s five major river basins (Yangtze, Yellow, Pearl, Huai, and Hai Rivers), our model demonstrates comprehensive superiority with faster convergence (46 iterations) and superior accuracy (R2 = 0.915), significantly outperforming basic attention (62 iterations, R2 = 0.862) and traditional ANFIS (85 iterations, R2 = 0.828) models across all metrics. There is a significant gradient difference in environmental performance scores across the five major river basins: the Yangtze River Basin performs the best (0.99), followed by the Yellow River Basin (0.98), with the Hai River (0.92) and Huai River (0.86) in the middle, and the Pearl River Basin scoring the lowest (0.77). This disparity reflects the differences in basin characteristics and governance.