Construction of strength prediction model for freeze-thaw saline soil based on ultrasonic velocity
The existing research on the macroscopic strength of frozen soil under the influence of macroscopic control factors mainly relies on experimental methods,and has achieved good results based on actual conditions.Generally speaking,both indoor and outdoor tests have shortcomings such as long cycle time and high cost.With the emergence of new technological means,exploring simpler methods and building predictive models has been a long-term endeavor of scientific researchers.At the same time,the influence of macroscopic control fac-tors on the macroscopic strength of frozen soil is exerted through the medium of the internal characteristics of the soil.Since ultrasonic waves are a good carrier of relevant information such as the physical and mechanical prop-erties of rock and soil media,ultrasonic testing can reflect the internal characteristics of the soil due to its non-de-structive,fast and simple characteristics.Therefore,this paper designs a strength prediction model containing different types of parameters based on different ideas.Idea 1-macro-controlling factors to macro-strength charac-teristics,idea 2-macro-controlling factors to internal soil characteristics reflected by ultrasonic wave velocity and then to macro-strength properties.Through experiments,the ultrasonic wave velocity and uniaxial compressive strength of soils with different salt contents undergoing different freeze-thaw cycles were obtained as basic data.The experimental control variables are used as idea 1 parameters,the ultrasonic characteristic parameter group constructed with compressional and shear wave velocities is used as idea 2 parameters,and the combined two ideas parameters are used as model input.A BP neural network prediction model for uniaxial compressive strength was established,and the prediction model was evaluated using the default factor test method.Tests show that as the number of freezing and thawing times and the salt content increase,the uniaxial compressive strength decreases overall.The wave velocity fluctuates significantly in the early stages of freezing and thawing,slows down in the middle stage,and returns to near the initial value in the later stage.Under the action of con-trolling factors,the uniaxial compressive strength decreases step by step as the wave velocity increases.The idea 2 parameters after gray correlation and rough set optimization are used to establish a BP neural network model for the optimal subsequence responding to the internal characteristics of the soil.The average absolute error of the model is less than 0.05 kPa,the coefficient of determination is greater than 0.96,and the average sensitivity index of each parameter is 1.4251.Sensitivity analysis successfully verified the assumed status of controlling factors and optimal subsequences in the model building process.A single controllable parameter has a greater im-pact on uniaxial compressive strength than a single ultrasonic characteristic parameter.The 29 parameters can be divided into four levels according to their contribution weight to the model.In the subsequent dimensionality re-duction and feature selection of the number of parameters,the fourth level parameters should be discarded first,and the third level parameters should be optimized through parameter construction innovation and data sample ex-pansion.This can reduce the number of overall parameters and increase the contribution weight,thereby better optimizing the model.The BP neural network model of uniaxial compressive strength established based on the different ideas of ultrasonic testing has strong predictive ability and good interpretability of model parameters.The ultrasonic characteristic parameter group under the influence of control factors plays an important role in the construction of the strength model.It also verifies the reliability and effectiveness of the BP neural network mod-el in predicting the uniaxial compressive strength of saline soil.The model has high accuracy and strong practica-bility,and can provide a reference for strength prediction and parameter selection of frozen soil models.
BP neural network modelultrasonic characteristic parameter groupfreezing and thawing saline soilstrength forecast
NETL
NSTL
万方数据
