DCM pile compressive strength prediction method based on PCA-GA-RF
Through the PCA-GA-RF method, the hyperparameters are optimized by principal component analysis and random forest model, and the problems of limited dimensions and low accuracy in the prediction of compressive strength of DCM piles are solved, achieving high-precision and stability prediction effects, suitable for DCM pile construction in complex geological environment.
Patent Information
- Application Number
- CN202411138721.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-19
- Publication Date
- 2024-12-06
AI Technical Summary
The traditional DCM pile compressive strength prediction method is limited to specific working conditions, and the existing machine learning method has limited dimensions, resulting in low prediction accuracy and lack of ability to process high-dimensional data, making it difficult to provide efficient and accurate quality control in complex geological environments. .
Using a PCA-GA-RF-based method, features are extracted through principal component analysis, and the random forest model is combined with genetic algorithm to optimize hyperparameters, and multiple decision tree models are constructed to process high-dimensional data and improve prediction accuracy.
The stability and robustness of the compression strength prediction of DCM piles is improved, high-precision prediction results are achieved, and construction efficiency and accuracy are improved.
Smart Images

Figure CN119089320A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of foundation reinforcement DCM pile construction, and in particular to a DCM pile compressive strength prediction method based on PCA-GA-RF. Background Art
[0002] Traditional underwater soft foundation reinforcement methods mainly include underwater explosive silt squeezing, underwater excavation and replacement, underwater vibration, etc. These methods will cause large engineering volume, low efficiency, environmental pollution and other problems in the process of deep water and large-scale development of hydraulic structures. In order to solve the above problems, the deep cement mixing pile method (DCM) to reinforce underwater soft soil foundation came into being. This method forcibly mixes cement and in-situ soil to form cement soil with a certain strength, so as to achieve the construction method of reinforcing underwater soft foundation, which has the characteristics of flexible layout and environmental friendliness. However, my country has little experience in the application of underwater DCM technology, and can only rely on limited construction experience for the quality control of DCM piles. These experiences may lead to misjudgment of the quality of piles, and empirical judgment of a large amount of data will face the problem of low efficiency. In order to further improve the efficiency of pile formation, it is necessary to control the quality of DCM piles in a more reliable way. The quality influencing factor of DCM piles is mainly its compressive strength. Therefore, the prediction of the compressive strength of DCM piles is extremely important in the construction process.
[0003] However, the existing research on the prediction of compressive strength of DCM piles is relatively limited. Traditional prediction methods mainly use empirical methods to predict based on on-site construction experience combined with indoor model tests. Or use indoor or on-site test results to recommend relevant formulas for pile quality prediction through fitting methods. However, the above methods all have limitations. For example, the empirical method is limited in prediction when facing complex geological environments, which affects the prediction accuracy. Most of the methods using fitting formulas do not consider the impact of water spraying and grouting volume on the compressive strength of the pile body during construction. In addition, as too many factors are considered, the difficulty of fitting the formula will increase exponentially. In order to overcome the limitations of traditional methods, the rapid development of machine learning technology in recent years has provided a new solution for the prediction of DCM piles. Some existing studies use support vector machines and BP neural networks to predict the compressive strength of DCM piles. However, these predictions consider limited dimensions, resulting in low prediction accuracy.
[0004] Therefore, it is necessary to propose a DCM pile compressive strength prediction method based on PCA-GA-RF that can process high-dimensional data while ensuring computational efficiency and accuracy. Summary of the invention
[0005] The purpose of the present invention is to solve the problems that the traditional DCM pile compressive strength prediction method is limited to specific working conditions and the existing machine learning prediction method has limited consideration dimensions and low prediction accuracy. A DCM pile compressive strength prediction method based on PCA-GA-RF is provided, which can be widely used in the field of foundation reinforcement DCM pile construction.
[0006] To achieve the above object, the present invention provides the following technical solutions:
[0007] S101, data collection and preprocessing;
[0008] The data collection and preprocessing include the DCM pile collection system acquiring relevant parameters during the construction process, including construction elevation, cement content, total water content, BRN, pre-mixing times, total mixing times, age, water depth, moisture content, natural density, void ratio, liquid limit, plastic limit, compression coefficient, compression modulus, quick shear c value, quick shear φ value, soil pressure and shear strength, and then obtaining the actual DCM pile compressive strength measurement value through on-site coring test, and arranging the relevant parameters and the corresponding sampling height and DCM pile compressive strength as a data set. The preprocessing process uses a local anomaly factor algorithm to identify outliers, and uses an interpolation method to replace outliers and missing values, and then uses the mean μ of each parameter in the data set to obtain the measured value. j , variance σ j , for each parameter data x i,j Perform standardization transformation and divide the preprocessed data set into training set and test set in a ratio of 8:2;
[0009]
[0010] In the formula, z i,j -The jth parameter data of the i-th standardized sample;
[0011] S102, extracting features using principal component analysis;
[0012] The principal component analysis method extracts features, including taking various parameters in the data set except the compressive strength of DCM piles as features, and calculating the covariance matrix of the data to measure the linear correlation between different features.
[0013]
[0014] In the formula, C is the covariance matrix, z is i - the i-th standardized sample, - sample mean, n- number of samples; then calculate the eigenvalues and eigenvectors of the covariance matrix, set the eigenvalue threshold K, and take the features with eigenvalues greater than K as the principal components and input them into the prediction model;
[0015] S103, Random Forest Model Hyperparameter Optimization;
[0016] The random forest model hyperparameter optimization includes constructing a random forest model and optimizing the model's hyperparameters using a genetic algorithm, and the optimization process includes:
[0017] a. Population initialization: Initialize the hyperparameters of the random forest model and obtain N sets of hyperparameter combinations as the initialization population. Each set of hyperparameter combinations is an individual in the population. The initialization population is composed of random chromosomes. All chromosomes are binary coded. When the code is "0", the hyperparameter corresponding to the code is not selected.
[0018] b. Construct a fitness function, calculate the fitness function value according to the error function MSE, and select individuals with large fitness function values to be inherited to the next generation;
[0019]
[0020] In the formula, y i - Actual value of compressive strength of DCM pile, -Prediction of compressive strength of DCM piles;
[0021] c. Genetic algorithm selection, crossover, and mutation. The model uses the hyperparameters in the initialized population to predict the compressive strength of DCM piles, evaluates the individuals in the population according to the fitness function, selects the first m excellent individuals as the parent population according to the fitness of different individuals through the selection operator, and then uses the set crossover operator and mutation operator to perform crossover and mutation operations on the individuals in the parent population to form candidate individuals with new features. The crossover and mutation operations are repeated until the stop condition is met or the maximum number of iterations is reached, and the individual with the largest proportion of the total number of individuals in the new population in the last iteration is output;
[0022] S104, model testing;
[0023] The model test includes using the test set data to verify the random forest regression prediction model optimized by the genetic algorithm, and using the determination coefficient R 2 , root mean square error RMSE, and mean absolute error MAE are used as performance evaluation indicators.
[0024] As a preferred technical solution of the present invention, in step S101, the training set and the test set are divided by a 5-fold cross-validation method, the data set is randomly divided into 5 equal subsets, and then steps S102 to S104 are iterated 5 times, one of the subsets is selected as the validation set in each iteration without repetition, and the remaining 4 subsets are selected as training sets.
[0025] As a preferred technical solution of the present invention, in step S102, the characteristic value threshold K is 0.8.
[0026] As a preferred technical solution of the present invention, in step S103, the random forest model hyperparameters must include the minimum number of leaf node samples and the number of decision trees.
[0027] As a preferred technical solution of the present invention, in step S103, the crossover and mutation operations both randomly select individuals for operation, and the crossover operation needs to add a discrimination condition so that individuals with large differences are cross-transposed.
[0028] The beneficial effects of the present invention are as follows: Random Forest (RF) improves the stability and robustness of the compressive strength prediction model of DCM piles by constructing multiple decision trees and combining their prediction results, uses principal component analysis (PCA) to extract effective information from high-dimensional data, uses genetic algorithm (GA) to seek the optimal hyperparameters of random forest, further improves the prediction performance of random forest, and compensates for the shortcomings of existing methods by improving the random forest method, bringing an innovative solution to the problem of compressive strength prediction of DCM piles. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 It is a flow chart of the DCM pile compressive strength prediction method based on PCA-GA-RF of the present invention. DETAILED DESCRIPTION
[0030] The specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. It should be understood that the specific embodiments given here are only used to illustrate and explain the present invention and cannot be used to limit the present invention. It should be noted that many specific details are set forth in the following description to facilitate a full understanding of the present invention, but the present invention may also have other embodiments and variations thereof, and therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.
[0031] The specific implementation steps of Example 1 are as follows:
[0032] S101, data collection and preprocessing;
[0033] The DCM pile acquisition system obtains relevant parameters during the construction process, including construction elevation, cement content, total water content, BRN, pre-mixing times, total mixing times, age, water depth, moisture content, natural density, void ratio, liquid limit, plastic limit, compression coefficient, compression modulus, quick shear c value, quick shear φ value, soil pressure and shear strength. Then, the actual DCM pile compressive strength measurement value is obtained through on-site coring test. The relevant parameters and the corresponding sampling height and DCM pile compressive strength are sorted as a data set. The data set contains 21 parameters and 257 samples. The preprocessing process uses the local anomaly factor algorithm to identify outliers and uses the interpolation method to replace outliers and missing values. Then, the mean μ of each parameter in the data set is used. j , variance σ j , for each parameter data x i,j Perform standardization transformation, randomly divide the preprocessed data set into 5 equal subsets, and then perform steps S102 to S104 for 5 iterations, and select one of the subsets as the validation set in each iteration without repetition, and the remaining 4 subsets as the training set. In this embodiment, the data set cannot be accurately divided into 5 equal parts, and 51 items are taken from each of the 3 subsets, and 52 items are taken from each of the other two subsets;
[0034]
[0035] In the formula, z i,j -The jth parameter data of the i-th standardized sample;
[0036] S102, extracting features using principal component analysis;
[0037] The parameters in the data set except the compressive strength of DCM piles are taken as features, and the covariance matrix of the data is calculated to measure the linear correlation between different features.
[0038]
[0039] In the formula, C is the covariance matrix, z is i - the i-th standardized sample, - sample mean, n- number of samples; then calculate the eigenvalues and eigenvectors of the covariance matrix, and take the features with eigenvalues greater than 0.8 as the principal components and input them into the prediction model;
[0040] S103, Random Forest Model Hyperparameter Optimization;
[0041] A random forest model is constructed and the hyperparameters of the model are optimized using a genetic algorithm. In this embodiment, two key parameters, the minimum number of leaf node samples and the number of decision trees in the random forest model, are optimized. The optimization process includes:
[0042] a. Population initialization: Initialize the hyperparameters of the random forest model and obtain 50 sets of hyperparameter combinations as the initialization population. Each set of hyperparameter combinations is an individual in the population. The initialization population is composed of random chromosomes. All chromosomes are binary coded. When the code is "0", the hyperparameter corresponding to the code is not selected.
[0043] b. Construct a fitness function, calculate the fitness function value according to the error function MSE, and select individuals with large fitness function values to be inherited to the next generation;
[0044]
[0045] In the formula, y i - Actual value of compressive strength of DCM pile, -Prediction of compressive strength of DCM piles;
[0046] c. Genetic algorithm selection, crossover, and mutation. The model uses the hyperparameters in the initialized population to predict the compressive strength of DCM piles, evaluates the individuals in the population according to the fitness function, and selects the top 20 excellent individuals as the parent population according to the fitness of different individuals through the selection operator. Then, through the set crossover operator and mutation operator, the individuals in the parent population perform crossover and mutation operations, with a crossover probability of 0.8 and a mutation ratio of 0.2 to form candidate individuals with new features. Both crossover and mutation operations randomly select individuals, and the crossover operation needs to increase the discrimination condition so that individuals with large differences can be cross-changed. Repeat the crossover and mutation operations until the stop condition is met or the maximum number of iterations is reached. The maximum number of iterations is set to 100, and the individual with the largest proportion of the total number of individuals in the new population of the last iteration is output;
[0047] S104, model testing;
[0048] The test set data was used to verify the random forest regression prediction model optimized by the genetic algorithm. The performance evaluation index determination coefficient R 2 =0.95, root mean square error RMSE =0.42, mean absolute error MAE =0.28.
[0049] In summary, the DCM pile compressive strength prediction method based on PCA-GA-RF in the present invention has the characteristics of high prediction accuracy, strong stability and robustness in the field of foundation reinforcement DCM pile construction.
[0050] It should be understood that the above-mentioned embodiments are one or more embodiments of the present invention, and there are many other embodiments and variations thereof based on the present invention; the variations and modifications made by ordinary technicians in this industry through the present invention without making pioneering innovations all fall within the scope of protection of the present invention.
Claims
1. A method for predicting the compressive strength of DCM piles based on PCA-GA-RF, characterized in that The specific steps include: S101, data collection and preprocessing; The data collection and preprocessing include the DCM pile collection system acquiring relevant parameters in the construction process, including construction elevation, cement content, total water content, BRN, pre-mixing times, total mixing times, age, water depth, moisture content, natural density, void ratio, liquid limit, plastic limit, compression coefficient, compression modulus, quick shear c value, quick shear φ value, soil pressure and shear strength, and then obtaining the actual DCM pile compressive strength measurement value through on-site coring test, and arranging the relevant parameters and the corresponding sampling height and DCM pile compressive strength as a data set. The preprocessing process uses the local anomaly factor algorithm to identify outliers, and uses the interpolation method to replace outliers and missing values, and then uses the mean μ of each parameter in the data set to obtain the measured value. j , variance σ j , for each parameter data x i,j Perform standardization transformation and divide the preprocessed data set into training set and test set in a ratio of 8:2; In the formula, z i,j -The jth parameter data of the i-th standardized sample; S102, extracting features using principal component analysis; The principal component analysis method extracts features, including taking various parameters in the data set except the compressive strength of DCM piles as features, and calculating the covariance matrix of the data to measure the linear correlation between different features. In the formula, C is the covariance matrix, z is i - the i-th standardized sample, - sample mean, n- number of samples; then calculate the eigenvalues and eigenvectors of the covariance matrix, set the eigenvalue threshold K, and take the features with eigenvalues greater than K as the principal components and input them into the prediction model; S103, Random Forest Model Hyperparameter Optimization; The random forest model hyperparameter optimization includes constructing a random forest model and optimizing the model's hyperparameters using a genetic algorithm, and the optimization process includes: a. Population initialization: initialize the hyperparameters of the random forest model, obtain N sets of hyperparameter combinations as the initialization population, each set of hyperparameter combinations as an individual in the population, and the initialization population is composed of random chromosomes. All chromosomes are binary coded. When the code is "0", the hyperparameter corresponding to the code is not selected; b. Construct a fitness function, calculate the fitness function value according to the error function MSE, and select individuals with large fitness function values to be inherited to the next generation; In the formula, y i - Actual value of compressive strength of DCM pile, -Prediction of compressive strength of DCM piles; c. Genetic algorithm selection, crossover, and mutation. The model uses the hyperparameters in the initialized population to predict the compressive strength of DCM piles, evaluates the individuals in the population according to the fitness function, selects the first m excellent individuals as the parent population according to the fitness of different individuals through the selection operator, and then uses the set crossover operator and mutation operator to perform crossover and mutation operations on the individuals in the parent population to form candidate individuals with new features. The crossover and mutation operations are repeated until the stop condition is met or the maximum number of iterations is reached, and the individual with the largest proportion of the total number of individuals in the new population in the last iteration is output; S104, model testing; The model test includes using the test set data to verify the random forest regression prediction model optimized by the genetic algorithm, and using the determination coefficient R 2 , root mean square error RMSE, and mean absolute error MAE are used as performance evaluation indicators.
2. The DCM pile compressive strength prediction method based on PCA-GA-RF according to claim 1, characterized in that: In step S101, the training set and the test set are divided by a 5-fold cross-validation method, the data set is randomly divided into 5 equal subsets, and then steps S102 to S104 are iterated 5 times, one of the subsets is selected as the validation set in each iteration without repetition, and the remaining 4 subsets are selected as the training sets.
3. The DCM pile compressive strength prediction method based on PCA-GA-RF according to claim 1, characterized in that: In step S102, the characteristic value threshold K is set to 0.
8.
4. The DCM pile compressive strength prediction method based on PCA-GA-RF according to claim 1, characterized in that: In step S103, the random forest model hyperparameters must include the minimum number of leaf node samples and the number of decision trees.
5. The DCM pile compressive strength prediction method based on PCA-GA-RF according to claim 1, characterized in that: In step S103, the crossover and mutation operations both randomly select individuals for operation, and the crossover operation needs to add a discrimination condition so that individuals with large differences can be cross-changed.
Citation Information
Cited By
GNSS coordinate time sequence prediction method for optimizing SVM based on genetic algorithm
CN119939258A
