Prediction of slope infiltration based on artificial neural networks by free search

Attempts of using FSBP were made to predict infiltration of natural rainfall on the slop surface of red soil under different land use patterns. Seven indexes such as precipitation, maximum rainfall intensity, rainfall duration, initial soil water content, soil bulk density, soil porosity and underlaying surface were selected as input variable, and the amount of infiltration as output variable. Results show that the mean relative error of the prediction is 11.08%, and t test and regression analysis indicates that the predicted value differs just slightly from the observed value and their correlation coefficient was 0.9715. The model is quite high in accuracy and stability, and serves as useful tool in further research on prediction of infiltration of nature rainfall on slopes.