Research On Frozen Soil Water Migration Based On Neural Network Technique

Water migration performs a core role in the process of Froze soil freezing or thawing. The composition characteristics of froze soil mediums make its different features from the general nature of the soil. The water migration of Froze soil not only relates to characteristics of the soil, but also relates closely to many other factors, such as water content, temperature, loading time, etc., among them there are complex nonlinear relationship.Because Artificial Neural Network can better imitate nonlinear systems, which is also a good network structure with strong learning ability, therefore it has more extensive usage in scientific prediction. In this essay, on the basis of its outstanding function of Back Propagation Neural Network algorithm, do improvement of its weakness, and add momentum factor, adaptation factor and elasticity gradient into it during the weights and threshold changes, and select intermediate nodes of middle layer at the test methodology.In this essay, I use improved Back Propagation Neural Network algorithm, and select related test data, establish the forecasting model of unfrozen water content in froze soil, water migration of freezing soil in the open system, water migration of freezing soil in the close system and water migration of frozen soil in the close system. The essay is based on the established neural network predictive model, it presents systematic and specific analysis on the main factors which will influence water migration in froze soil, and discusses something which is different from dry bulk density, initial moisture content, salinity, external load, temperature gradient, etc., such factors will impact on predictive value, and set the corresponding relations of fitting function.