Intelligent construction control method, apparatus and system for mixing pile with split-grouting

An intelligent construction control system for splitting jet grouting mixing piles, comprising

• a real-time perception information acquisition module, configured to:
• receive real-time perception information from construction information self-perception system, where the real-time perception information includes data of drilling depth, output power of a drilling rig, torque of a drill bit, axial force of a drill rod and pore water pressure of the drill bit;
• a real-time stratum state information prediction module, configured to:
• input the real-time perception information into the trained machine-learning model for stratum information, establish a nonlinear implicit correspondence between the real-time perception information and construction stratum conditions, and output real-time stratum state information;
• and
• a current optimal construction parameter acquisition module, configured to:
• input the real-time stratum state information into a trained self-matching machine-learning model for optimal construction parameters, establish a nonlinear implicit correspondence between the real-time stratum state information and optimal construction parameters, and output current optimal construction parameters;
• 1) data preprocessing:
• removing data outliers based on a statistical analysis method, and using a regression interpolation method to fill local sparse data;
• 2) Feature selection and standardization:
• evaluating a correlation degree between features and target variables using a chi-square test method, and selecting five features with a highest correlation degree, and converting selected five features to a normal distribution with a mean value of 0 and a variance of 1 using a Z-score standardization method, to make them comparable;
• 3) Data division:
• dividing a database into n datasets with approximate sizes using a K-fold cross validation method;
• 4) User-defined kernel function:
• using a user-defined polynomial kernel function to map the data to a high-dimensional feature space, therefore transferring nonlinear problems to linear problems;
• 5) model training and evaluating:
• performing K-round training to evaluate on the built construction parameter self-matching model, where n−1 datasets in each round are used for training, and remaining datasets are used as a test set, and finally calculating the average value among K-round evaluation results as a performance index of the model, and performing next round of training if not meet the requirement;
• 6) optimal construction parameters prediction:
• inputting real-time stratum state information into a trained self-matching machine-learning model for optimal construction parameters, and obtaining predicted optimal construction parameters through forward propagation;
• the self-matching machine-learning model for optimal construction parameters is a specially designed SVM class model, a kernel function of the model is the user-defined polynomial kernel function.