| Frozen soil is a type of soil that is extremely temperature sensitive.It is crucial for the design and construction of the frozen ground temperature in the frozen water area.The traditional calculation formula does not consider the factors such as solar radiant heat,which has some limitations.This paper takes the low liquid limit clay of the embankment of the main stream of Heilongjiang Province as the research object,nuclear magnetic resonance test is carried out.Freeze-thaw tests on the physical slope model are carried out,considering the impact of irradiation.The variation laws of factors such as temperature,unfrozen water content,and radiant heat and their influence on the freezing depth of seasonal frozen soil are analyzed,and the artificial neural network model is optimized using the whale optimization algorithm to predict the freezing depth and melting depth of seasonal frozen soil.This research investigates the change in unfrozen water content in frozen soil,as well as the seasonal freezing and thawing process and the rule of low liquid limit clay,using test data from a physical slope model.The goal is to add to and build on previous studies on the freezing depth of seasonal frozen soil in cold climates.The following are the key research findings and conclusions:（1）Conduct NMR test.The T2 relaxation time of six clay samples with different moisture content was measured by NMR tester,and the unfrozen water content of soil samples at 16 different temperature points was calculated.According to the test data,it is found that when the temperature increases,the unfrozen water content also increases.With the negative temperature,the unfrozen water content also increases.At-80°C,the unfrozen water content of six samples with different initial moisture content is basically the same.The freezing process of soil samples is also reflected in the change in unfrozen water content.The relaxation time curve reflects the internal changes of soil during freezing,the peak value of the relaxation time curve drops as the temperature decreases,and dramatically declines at-30°C.In the process of soil freezing,compression occurs continuously in the soil to form small pores,and the position of wave crest shifts to the left.The relaxation time curves of-60°C～-80°C are basically the same.（2）Unfrozen water content prediction models based on BP neural networks and BP neural networks based on whale optimization algorithms are constructed and compared to the usual three fitting formula techniques.The fitting effects of the three fitting formulas in the formula fitting method are essentially the same,and the overall change is consistent with the trend of the measured data,but the fitting effect is poor in the range of 0°C～5°C,which does not meet the prediction requirements.The fitting accuracy of the BP neural network prediction model is greater than that of the formula,and the fitting correlation coefficient may reach 0.9745,however due to its own setup,it is easy to slip into local minimum points and convergence speed is sluggish.The unfrozen water content prediction model WOA-BP has a high fitting accuracy,a faster convergence speed,and a lower convergence value.The prediction model’s fitting mean square error is as low as 3.9734e-10,and the absolute value of the prediction error is less than 10-6,allowing it to estimate the unfrozen water content of soil samples with varying beginning moisture content and negative temperatures,and provide input data for the establishment of freezing depth and melting depth prediction model in Chapter 4.（3）Carry out the freeze-thaw test of physical slope model considering the influence of irradiation.According to the similarity criteria,combined with laboratory specifications and field temperature data,the size of slope model is designed,and the solar radiation simulator and temperature and displacement sensors are arranged.Using the laboratory correction coefficient to improve the indoor temperature control mode and the control mode of solar radiation simulator,the control freeze-thaw test with or without radiation effect was carried out.Under the impact of irradiation,the soil temperature of the slope presents a sinusoidal curve,and the test group fluctuates periodically.The temperature of the soil is less impacted by ambient temperature as soil depth increases,while the temperature of deep soil swings around 0°C.The freezing process of slope is consistent with the change process of unfrozen water content in soil sample.The soil will experience frost heave and thawing settling as a result of the freezing process.There would be some residual distortion after the test.The test group influenced by solar irradiation had less residual deformation than the control group,and the maximum deformation is likewise higher.The duration of the slow freezing phase was longer and the forward melting time was earlier in the test group.（4）The influence of irradiation is taken into account while creating the freezing curve prediction model and melting curve prediction model based on BP neural network and BP neural network based on whale optimization algorithm.The prediction model can predict the freezing curve and melting curve on the premise of known temperature,moisture content and radiant heat,so as to achieve the prediction effect of clay freezing depth.Compared with the freezing depth obtained from physical model test,the traditional Stephen formula not only calculates the maximum freezing depth,but also the freezing time and melting time are short.The improved Stephen formula considering the influence of irradiation can predict the maximum freezing depth,but there is still a gap between the freezing curve and melting curve and the measured value.On the melting curve,the prediction impact of the BP neural network prediction model is better than on the freezing curve.The overall prediction accuracy is higher than that calculated by formula,but its convergence value is large.The BP neural network freezing curve prediction model has a mean square error of 1.59,whereas the BP neural network melting curve prediction model has a mean square error of 0.26.WOA-BP neural network has higher prediction accuracy.WOA-BP neural network freezing curve prediction model has a mean square error of 0.03,and WOA-BP neural network melting curve prediction model has a mean square error of 0.02,which are much superior than the BP neural network prediction model and formula calculation.The prediction result of WOA-BP neural network prediction model is closer to the measured value of physical model test,and the prediction effect is better. |
PDF Full Text Request