PIANC Smart Rivers 2022

Reviewing Presentation documents

Topic:

Title:

Author(s):

Author 1

Name：

Affiliations：

Region：

Email：

Is corresponding author or not (one only)：

Author 2

Name：

Affiliations：

Region：

Email：

Is corresponding author or not (one only)：

Author 3

Name：

Affiliations：

Region：

Email：

Is corresponding author or not (one only)：

Abstract : *(250~1000 words)

The ecological condition of waterways has attracted increasing public attention in recent years, and the evaluation of the ecological status of waterways is of practical and scientific significance. In order to carry out an objective and credential evaluation of the ecological condition of waterways, this study compares the performance of two artificial neural networks (ANN) models, including the traditional Back Propagation (BP)-ANN model and the Particle Swarm Optimization (PSO)-BP-ANN model. The traditional BP-ANN model randomly selects data during training processes, indicating a local search model. The searching strategy may cause the model to converge to local minima, and different minima may lead to inconsistent results each time. To improve the traditional BP-ANN model, this study applies the PSO to solve the weights and thresholds first, and then substitutes them into the BP-ANN model for continuous iteration, so that local optimal solutions can be avoided. This study firstly establishes an evaluation index system for the ecological evaluation of waterways based on the main features of waterway ecology, which includes 13 indexes in four aspects: shipping functions, ecological functions, landscape functions, and economic functions. According to these indexes, a waterway ecology database was established based on major waterway networks in China, including Yangtze River, Pearl River, Beijing-Hangzhou Canal, and Huai River. The results show that the BP-ANN has a coefficient of determination of 0.6209 when using the elastic BP optimization algorithm, and the best calibration error can be achieved after about 15 training iterations, but the prediction results of the BP-ANN are unstable due to its local searching strategy. In contrast, the coefficient of determination of the PSO-BP-ANN model reaches 0.98, indicating the high prediction accuracy and stability of the model. On this basis, this study further analyzed the relative importance of the above functions using the PSO-BP-ANN model. Results showed that the optimal range of the shipping function is between 0.6 and 0.85; the optimal range of the ecological function is around 0.65. The waterway ecological grade tends to be stable after the ecological function value is greater than 0.8 and continuing to improve this function cannot increase the waterway ecological grade. The optimal range of landscape function value is around 0.8 and the optimal range of economic function score should be higher than 0.6. Meanwhile, for the waterway with a poor rating, the evaluation results can be significantly improved by advancing the landscape function and the shipping function, and the increase of these two scores can greatly reduce the probability of the lowest rating. The highest rating requires all four functions to be greater than certain threshold values.

Key words: (max: 5)

Reference no.

Slides：

click to download