Research On Prediction,optimization And Variation Patterns Of Earth Pressure In Earth Pressure Balance Shields

The earth pressure balance shield is one of the common pieces of equipment for drilling subway tunnels.As the core of this machine,earth pressure(in chamber)greatly affects surface settlement,but controlling it is a complex problem due to joint influences of multiple variables.As for predicting earth pressure,the reasonable intelligent model can effectively improve the accuracy and timeliness of the theoretical one,and reduce the difficulty of determining various parameters.However,existing research on intelligent models still has the following problems:(1)prediction intervals that are inconsistent with engineering;(2)neglecting the influence of soil conditioning parameters;(3)lacking analysis of variation patterns of important variables in models.In order to further perfect intelligent models and provide guidance for controlling earth pressure,many intelligent models for predicting earth pressure were built in this research based on the big data collected from the Chengdu Metro Line 19 project.Meanwhile,the impact of crucial variables on earth pressure was analyzed,and the unreasonable earth pressure was optimized in a certain drilling cycle.The main content and conclusions are as follows:(1)A massive amount of data was processed at a time interval of 1 second,and intelligent models for predicting earth pressure were established based on various machine learning algorithms.Then,taking the random forest model as an example,its generalization ability was tested and the cause of prediction error was analyzed.The results show that most models can meet the engineering requirements(absolute error<0.1 bar)at the prediction interval consistent with the project.The random forest model has a strong generalization ability,and part of its wrong prediction may result from the frequent changes in the habits of driving shields.(2)Water and foam were added to models as two kinds of soil conditioning parameters,and their effects on the prediction performance and feature importance were analyzed.The results show a significant correlation between soil conditioning parameters and earth pressure,which can obviously improve the performance of models.In addition to propulsion pressure and rotational speed of cutterhead,the four most important variables in prediction models also include foam air flow and central water flow.(3)The method of controlling variates was used to analyze the variation rules of earth pressure when crucial features were changing.The genetic algorithm was utilized to optimize some earth pressure that was not between the valid interval calculated by theory.The results show that when more earth pressure is required,a driver can consider raising propulsion pressure besides increasing advance rate and reducing rotational speed of screw conveyor,and should pay attention to the loss of earth pressure caused by adding foam air flow.