| In recent years,with the rapid development of highway and railway construction in China,a large amount of materials are needed to fill the subgrade.Coal gangue has been widely used as a filling material for geotechnical engineering in foreign countries,and therefore it has broad application prospects as a subgrade material.The key issue in the construction of highways on coal gangue foundations is to control the subsidence of subgrade.Therefore,it is of great practical significance to accurately predict the subsidence of coal gangue subgrade and avoid the construction progress and quality due to the subsidence problem.Relying on the Science and Technology Project of the Ministry of Transport of the "Research on engineering application of coal gangue embankment fillings in expressway"(Project number 2010-353-343-290)and the transportation technology projects of Hunan province(Project number 200908)and the An Shao Expressway Coal Gangue Subgrade Project.Based on the back-propagation neural network non-linear mapping ability and learning ability,a variable learning rate BP is proposed to predict the subsidence of coal gangue subgrade in An Shao Expressway.The history data is used to establish the model of subgrade subsidence.The model has high prediction accuracy which can be used for subsidence prediction of coal gangue subgrade.In order to complete the above research,there are mainly works on this paper:(1)The research status of coal gangue at home and abroad,the research status of coal gangue subgrade subsidence,and the research status of application for BP neural network in subgrade subsidence are systematically reviewed.(2)Through investigating the gangue mining area in nine provinces,and the subgrade of highway is filled by coal gangue,and consulting relevant documents,combining with the requirement of highway subgrade filling,founding that it is feasible for the highway subgrade that filled by the coal gangue of part in the south.(3)A brief overview of the work processes,learning methods,and structural networks of artificial neural networks is provided.The principle and implementation process of BP artificial neural network are elaborated in detail.Theoretically,the advantages and disadvantages of BP neural network are analyzed and summarized.At the theoretical level,two optimization methods of momentum BP network and variable learning rate BP network are briefly compared.(4)The construction technology of coal gangue subgrade is introduced in detail.Hyperbolic curve fitting method,exponential curve fitting method,and Hoshino method are used to predict the K127+700 section of the subgrade subsidence data of the Anshao Expressway,and the subsidence curve is finally obtained through calculation and the help of the trend line.(5)The subgrade subsidence data from the Anshao Expressway was measured and modeled by the BP neural network.Combining the structural method and the empirical test,and the network topology suitable for the subsidence prediction of the coal gangue subgrade was found.With the aid of artificial neural network toolbox in MATLAB,using the momentum BP algorithm(corresponding to the tramingdm function in the toolbox)and variable momentum BP algorithm(corresponding to the toolbox traingdx function)are to predict subgrade subsidence,and a comparative analysis is made to determine the latter more suitable for the prediction of the subsidence of the coal gangue subgrade.By comparing the prediction results of BP neural network method and curve fitting method,it is concluded that the BP neural network algorithm is more suitable for the subsidence prediction of coal gangue subgrade than the curve fitting method.The above research work provides a more accurate method for the subsidence prediction of coal gangue subgrade,providing a basis for subsidence deformation of coal gangue subgrade,and reveals the subsidence change law of coal gangue subgrade.Based on this,it is proved that the momentum BP algorithm with variable learning rate is more accurate than the momentum BP algorithm. |
PDF Full Text Request