Ecological Evaluation of Waterways Based on Modified Neural Networks

Abstract The ecological condition of waterways has attracted increasing public attention in recent years, and the evaluation status is practical scientific significance. To carry out an objective credential waterways, this study compares performance two artificial neural networks (ANN) models, including traditional Back Propagation (BP)-ANN model Particle Swarm Optimization (PSO)-BP-ANN model. BP-ANN characterized by a local search strategy usually converges to minima. This combined PSO algorithm with improve latter. An indicator system was established first based on main features waterway ecology, which includes 13 indicators four aspects transportation function, landscape economic function. A ecology dataset then major China. results show that prediction accuracy around 0.6 unstable due its searching strategy. In contrast, coefficient determination PSO-BP-ANN reaches 0.98, indicating high On basis, further analyzed relative importance functions using Results showed grades can be significantly improved advancing function improvement these scores greatly reduce probability lowest rating. highest rating requires all greater than certain threshold values.