The BP Neural Network Model of Soil Water-Salt Dynamic State Analysis

Abstract

With the survey data of Luohui Canal Irrigation District, Shaanxi, China as the example, we employed the three-layer feed forward BP network modeling method to study the soil water-salt dynamic state under the comprehensive conditions of the irrigation district, and adopted the Additional Momentum Method and Self-adaptive Learning-rate Adjustment Strategy to modify the back propagation algorithm; on this basis, we employed the default-factor testing method to analyze the sensitivity degrees of soil salt content and soil alkalinity to every factor in the input layer. The results show this model has a high accuracy and can characterize effectively the internal relationships between the change of farmland soil water-salt dynamic state at a shallow water table during crop growth period and its influential factors. Soil moisture content, groundwater salt content and groundwater evaporating capacity are main sensitive factors of soil water-salt dynamic state; the factors interact and affect each other, giving rise to a coupling relationship under complex conditions. Combining the above methods can provide a feasible and effective approach to study the law of soil water-salt dynamic state under a shallow water table during crop growth period, which is a supplement to and improvement in conventional research methods for soil water-salt dynamic state.