| Abstrakt: |
Based on a series of indoor simulating infiltration tests, the influences of postponed water feeding time, the amount of cement, the permeation headwater, the texture and the bulk density of the sediment on the water infiltration and infiltration-reducing rate were analyzed. The BP neural network was optimized by genetic algorithm and particle swarm optimization and the prediction models of the infiltration-reducing rate were established based on three BP neural networks with 60 groups of sampled data of different test conditions. Although the results reveal that the forecast precision of three BP neural network forecasting models are different, those mean absolute percentage errors are all within 10%, and all can be used to forecast the infiltration-reducing rate. The mean absolute percentage error of traditional BP prediction model is 9.4%, while that of GA-BP and PSO-BP were 7.2% and 7.5%. The forecast accuracies of GA-BP and PSO-BP are partly improved and GA-BP and PSO-BP can better forecast the infiltration reduction results under the condition of fine particulate matter permeating river sediment. [ABSTRACT FROM AUTHOR] |
