Study On The Content Of Clay Modifier With High Liquid Limit Based On Improved BP Neural Network

High liquid limit clay is widely distributed in humid and hot areas in South China.Its natural moisture content is high,resulting in poor rolling effect and compaction performance.Filling embankment with it will often induce subgrade settlement,local cracking and slope instability.How to scientifically and reasonably use high liquid limit clay as subgrade filler and avoid massive waste is of great significance to guide expressway subgrade filling.Therefore,based on the investigation of treatment technology of high liquid limit clay at home and abroad,this paper takes the high liquid limit clay of Laidu K274+400～K280+100 excavation section of Hezhou Bama Expressway in Guangxi as the research object,discusses the advantages and disadvantages and application scenarios of three technical means of airing,lime and cement,and puts forward the prediction model of chemical modifier content based on LM and Br algorithm,The main research contents and results are as follows:(1)Through particle analysis,limit moisture content,dry and wet compaction test and CBR test,the basic characteristics of soil are explored,and the rolling moisture content is 14%～22.5%;Taking the consistency,CBR value and CBR expansion as the soil grade judgment index,it is judged that the soil of the test section is grade 3 soil,and measures shall be taken to reduce the water before it can be used as subgrade filler;In order to obtain better rolling performance and higher degree of compaction,the optimal moisture content of wet compaction(21%)is proposed as the precipitation target moisture content.(2)Based on indoor and outdoor drying experiments,the applicability of drying methods is analyzed;The compaction test and limit moisture content test of soil materials with different lime and cement contents are carried out,and the compaction performance and compactability of the treated soil are analyzed.The results show that the treated soil has good compaction performance,and the compactness of the subgrade filled with it can meet the requirements;Based on the orthogonal test,the influence of different factors on the precipitation effect is explored.It is found that the precipitation effect is significantly related to the content of lime and cement,the initial moisture content of soil,atmospheric temperature and humidity.(3)Because BP neural network is easy to fall into local minimum,low prediction accuracy and slow convergence speed,Br algorithm and LM algorithm are used to improve and optimize it,so as to obtain an improved BP neural network model with faster convergence speed and higher prediction accuracy;The improved model is used to predict the amount of lime and cement required under different working conditions,and its accuracy is verified by field test and stable moisture content prediction model.