A method for predicting the unconfined compressive strength of in-situ fluidized solidified sediment with different clay mineral components based on machine learning

Abstract

The application discloses a kind of different clay mineral component cast-in-place flow state solidified silt unconfined compressive strength prediction method based on machine learning, including collecting and arranging existing flow state solidified silt unconfined compressive strength data, preparation multiple groups with different clay mineral component and mix proportion solidified silt sample, and its unconfined compressive strength is determined by standard test, all sample data is divided into training set and test set according to 8:2, based on key influence factor analysis, establish four kinds of machine learning prediction model, training and optimization are carried out to model by training set data, obtain integrated model according to model accuracy distribution weight, and the steps of using integrated model to carry out strength prediction to cast-in-place solidified silt.The method can effectively replace the strength determination mode of traditional dependence on a large number of repeated trial and test, provide intelligent support for mix proportion optimization, significantly save cost and time.

Description

Technical Field

[0001] This invention relates to a method for predicting the unconfined compressive strength of solidified sediment, and more particularly to a machine learning-based method for predicting the unconfined compressive strength of cast-in-place solidified sediment. Background Technology

[0002] Dredged sediment and excavated soil typically have high moisture content, low strength, and complex composition. Direct dumping or transportation of these materials not only occupies significant land resources but also poses a risk of secondary environmental pollution. Fluidized solidification technology, with its advantages of good fluidity, pumpability, self-compacting properties, and adjustable strength, has become a key technology for the large-scale, resource-efficient utilization of such sediment, and is widely used in roadbed backfilling, engineering pile foundations, and pond backfilling.

[0003] The unconfined compressive strength of cast-in-place solidified soil is a core mechanical indicator for its engineering applications, directly related to the safety and stability of engineering structures. However, this strength is subject to complex influences from various factors, primarily including: The clay mineral composition of sediment: The type and content of clay minerals (such as montmorillonite, illite, kaolinite, chlorite, etc.) in sediment are key intrinsic factors determining its engineering properties. Different clay minerals vary greatly in specific surface area, cation exchange capacity, and water absorption and swelling properties. For example, sediments with high montmorillonite content require a large amount of water, are difficult to solidify, and significantly inhibit strength development; while sediments dominated by kaolinite are relatively easy to solidify. Traditional methods struggle to accurately quantify this complex nonlinear effect.

[0004] The proportions of the curing system include the type and dosage of curing agents (such as cement, lime, industrial waste slag-based activators, etc.), the use of admixtures (water-reducing agents, early-strength agents), water-to-solid ratio, water-to-binder ratio, etc.

[0005] Curing and molding conditions: such as curing age, curing environment (temperature, humidity), specimen size effect, etc.

[0006] However, in existing technologies, the prediction of the unconfined compressive strength of cast-in-place fluidized bed sediment mainly relies on trial and error methods and empirical formulas. Trial and error methods require numerous repeated experiments, which are time-consuming and labor-intensive; empirical formulas are difficult to accurately reflect the complex nonlinear effects of different clay mineral components (such as montmorillonite, illite, etc.) on strength, resulting in low prediction accuracy and poor universality. This invention introduces a machine learning model, which can learn and characterize the complex nonlinear mapping relationship between these factors and strength, thereby achieving accurate prediction of the solidification strength of bed sediment containing different clay minerals. Summary of the Invention

[0007] To address the shortcomings of existing technologies that rely on time-consuming and laborious trial-and-error methods and low-precision empirical formulas that are difficult to quantify the influence of clay minerals, this invention aims to establish an intelligent prediction method that can scientifically reflect the complex influence of different clay mineral components, with high precision and strong generalization ability. This method can replace traditional repeated experiments, enabling rapid and accurate prediction of unconfined compressive strength and intelligent optimization design of mix proportions, thereby significantly improving the efficiency and scientific nature of the resource utilization of cast-in-place fluidized solidified sediment.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: A method for predicting the unconfined compressive strength of cast-in-place fluidized solidified sediment with different clay mineral compositions based on machine learning, comprising the following steps: Step 1: Collect and organize the existing test data of fluidized solidified sediment, obtain the composition information and strength of each test material, and prepare multiple solidified sediment samples with different clay mineral compositions and mix proportions. Determine their unconfined compressive strength through standard tests. Step 2: Divide all sample data into training set and test set according to a preset ratio; Step 3: Based on the key factors affecting unconfined compressive strength, construct various machine learning prediction models, input the data input variables into the above models, and the output of the models is the predicted strength; Step 4: Train and optimize the machine learning model using the training set, assign weights according to the model accuracy, and finally integrate the prediction model. Step 5: Use the integrated model to predict the unconfined compressive strength of the cast-in-place solidified slurry.

[0009] Preferably, the test material composition information in step 1 includes: the content of each component in the sediment, the type and content of clay minerals, the amount of curing agent, the curing time, the water-cement ratio, and the water-solid ratio.

[0010] Preferably, the aforementioned clay mineral types include, but are not limited to, montmorillonite, illite, kaolinite, and chlorite.

[0011] Preferably, the machine learning prediction model includes, but is not limited to, at least one of the following algorithms: random forest algorithm, support vector machine algorithm, gradient boosting decision tree algorithm, multilayer perceptron neural network algorithm, and ensemble learning algorithm.

[0012] Preferably, the selection of the input variables is based on the degree of synergistic change between the eigenvalues ​​and the unconfined compressive strength, quantified using the Pearson correlation coefficient, and selecting features with an absolute value of a correlation coefficient greater than 0.5 with the unconfined compressive strength as input variables. The Pearson correlation coefficient is calculated using the following formula: ; Where Xi and Yi are the i-th actual values ​​of the two variables, respectively; and are the sample means of the two variables, respectively; n is the sample size.

[0013] Preferably, before inputting the input variables into the model, a normalization preprocessing step is included. Data preprocessing includes: allocating the data according to a test set:training set ratio of 2:8, and then normalizing the data.

[0014] The preferred normalization formula is: ; in, X norm For X, the normalized data is... X max The maximum value of the experimental data. X min The minimum value of the experimental data. X This refers to data from a specific experiment.

[0015] Preferably, the model optimization process includes tuning the key hyperparameters of the model and may employ a weighted ensemble strategy to integrate the final model.

[0016] Preferably, the weights are dynamically allocated based on the accuracy of the model, using the coefficient of determination R² as an indicator, as follows: ; in, w i Let be the weights of the i-th model; R i Let be the accuracy of the i-th model.

[0017] Preferably, the predictive performance of the above-mentioned ensemble model is measured by the correlation coefficient between the training set and the test set results ( R 2 ), mean square error ( RMSE Mean absolute error ( RAE ) and mean absolute percentage error ( MAPE To conduct an evaluation: ; ; ; ; in, For the total number of data, For test values, For predicted values, This represents the average test value. The closer the value is to 1, the better the fit. RMSE , MAE , MAPE The smaller the value, the higher the model's precision and accuracy.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: This paper presents a machine learning-based method for predicting the unconfined compressive strength of cast-in-place fluidized bed sediment with different clay mineral compositions. The method effectively captures the complex nonlinear influence of multiple factors, such as clay mineral type and content, and solidifier ratio, on strength development through machine learning models. It overcomes the shortcomings of traditional empirical formulas, such as low prediction accuracy and poor universality, achieving high-precision intelligent prediction. Simultaneously, this method significantly reduces the large number of repeated experiments required by traditional trial-and-error methods, greatly reducing material and time costs and improving the efficiency of mix design. Furthermore, through normalization preprocessing, feature screening, and multi-model integration, the method enhances the generalization ability and stability of the models, promoting the intelligent transformation of this field from "experience-dependent" to "data-driven," and providing efficient and reliable technical support for the resource utilization of bed sediment. Attached Figure Description

[0019] Figure 1 This is a detailed flowchart of the present invention. Figure 2 This is a scatter plot comparing measured values ​​and predicted values ​​in an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram illustrating the percentage of predicted value errors in an embodiment of the present invention.

[0021] Figure 4 This is a comparison chart of measured and predicted values ​​in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0023] This embodiment provides a method for predicting the unconfined compressive strength of cast-in-place fluidized solidified sediment with different clay mineral compositions based on machine learning. (See also...) Figure 1 The specific implementation steps are as follows: Step 1, Data Collection and Sample Preparation: 1) Experimental data on cast-in-place fluidized bed slurry were collected and organized from literature, engineering reports, and historical laboratory data, resulting in a total of 200 sets of valid data. Each set of data included the following input variables: clay mineral type (montmorillonite, illite, kaolinite, chlorite) and its content (%), curing agent dosage (cement and slag co-mixing, %), water-cement ratio, water-solid ratio, and curing age (d). The output variable was the 28-day unconfined compressive strength (MPa).

[0024] 2) To expand the dataset and cover a wider range of mineral assemblages, 80 sets of cast-in-place fluidized solidified soil specimens with different clay mineral compositions and mix proportions were prepared in the laboratory. Commercially available standard mineral powders were used for each clay mineral, mixed into simulated sediment at predetermined ratios, controlling the total clay mineral content to be 20%–40%, the solidifying agent dosage to be 5%–15%, the water-cement ratio to be 0.4–0.6, and the water-solid ratio to be 0.8–1.2. All specimens were cured under standard curing conditions (temperature 20±2 ℃, humidity ≥95%) to the specified age, and unconfined compressive strength was determined by unconfined compressive strength testing according to the "Standard for Geotechnical Testing Methods" (GB / T 50123).

[0025] Step 2: Dataset Partitioning and Preprocessing: All 280 sets of sample data were randomly divided into a training set (224 sets) and a test set (56 sets) in an 8:2 ratio. To eliminate the influence of different feature units, the input variables were normalized using the following normalization formula: ; Where Xnorm is the normalized data corresponding to X, Xmax is the maximum value of the experimental data, Xmin is the minimum value of the experimental data, and X is a certain experimental data.

[0026] Step 3, Model Building and Feature Input: Based on Pearson correlation coefficient analysis, features with an absolute value greater than 0.5 in the correlation coefficient with unconfined compressive strength were selected as model inputs. The final input variables included: montmorillonite content, kaolinite content, illite content, chlorite content, curing agent dosage, water-cement ratio, and curing age. Four machine learning regression prediction models were constructed: Random Forest (RF), Support Vector Machine (SVM), Linear Regression, and Multilayer Perceptron Neural Network (MLP).

[0027] Step 4, Model Training, Optimization, and Integration: The four models were trained using the training set, and key hyperparameters were tuned using a grid search method, such as the number and depth of trees in RF, the kernel function and penalty coefficient in SVM, and the hidden layer structure and learning rate in MLP. After training, five-fold cross-validation was used to evaluate the accuracy of each model on the training set (using R² as the evaluation metric), and a weighted ensemble was performed based on model accuracy. The weights were dynamically allocated according to the accuracy of each model, using the following formula: ; Where wi is the weight of the i-th model; Ri is the accuracy of the i-th model.

[0028] The performance of the ensemble model was evaluated using a test set, and four metrics were calculated: R2, RMSE, MAE, and MAPE. The calculation formulas are as follows: ; ; ; .

[0029] In this example, the accuracies (percentage of errors less than 10%) of Random Forest (RF), Support Vector Machine (SVM), Linear Regression, and Multilayer Perceptron Neural Network (MLP) are 86.3%, 97.5%, 48.2%, and 96.4%, respectively. The weights of each model, calculated using the weighting formula, are 0.261, 0.333, 0.081, and 0.325, respectively. Finally, the prediction model is integrated. The R-value of the integrated model on the test set is... 2 The results showed an accuracy of 0.9926, an RMSE of 0.38 MPa, a MAE of 0.31 MPa, and a MAPE of 6.7%, indicating that the model has high prediction accuracy and engineering applicability.

[0030] Step 5, Practical Engineering Prediction Application: Samples with different clay mineral composition contents, water-cement ratio, water-solid ratio, curing agent dosage, and curing age were preset, normalized, and input into the prediction model to obtain the predicted strength. Then, samples with the same variables were prepared and subjected to indoor tests to obtain the actual strength. The specific information is shown in Table 1. The difference between the predicted value and the actual value is about 5%, which verifies the reliability and applicability of the method of the present invention.

[0031] Table 1. Relevant information for specimen strength prediction

[0032] The predicted values ​​in the table are obtained from the prediction of the multilayer perceptron neural network, while the actual values ​​are the laboratory strength of the compound sample with the corresponding doping amount.

[0033] In summary, this embodiment, through system data preparation, reasonable model construction and optimization integration, achieves accurate prediction of the unconfined compressive strength of cast-in-place fluidized solidified bottom mud, provides intelligent support for engineering mix design and optimization, significantly improves resource utilization efficiency, and saves test costs and time.

Claims

1. A method for predicting the unconfined compressive strength of cast-in-place fluidized solidified sediment with different clay mineral compositions based on machine learning, characterized in that... Includes the following steps: Step 1: Collect and organize the existing test data of fluidized solidified sediment, obtain the composition information and strength of each test material, and prepare multiple solidified sediment samples with different clay mineral compositions and mix ratios. Determine their unconfined compressive strength through standard tests. Step 2: Normalize all sample data and divide them into training and test sets according to a preset ratio; Step 3: Based on the key factors affecting unconfined compressive strength, construct various machine learning prediction models, input the data input variables into the above models, and the output of the models is the predicted strength; Step 4: Train and optimize the machine learning model using the training set, assign weights according to the model accuracy, and finally integrate the prediction model; Step 5: Use the optimal model to predict the unconfined compressive strength of the cast-in-place solidified slurry.

2. The method for predicting the unconfined compressive strength of cast-in-place fluidized solidified sediment with different clay mineral compositions based on machine learning, as described in claim 1, is characterized in that... The test material composition information in step 1 and the input variables in step 3 are selected based on the degree of synergistic change between the eigenvalues ​​and the unconfined compressive strength, including but not limited to the content of each component in the sediment, the type and content of clay minerals, the amount of curing agent, the curing time, the water-cement ratio, and the water-solid ratio.

3. The method for predicting the unconfined compressive strength of cast-in-place fluidized solidified sediment with different clay mineral compositions based on machine learning, as described in claim 1, is characterized in that... The various machine learning prediction models mentioned in step 3 include, but are not limited to, at least one of the following: random forest algorithm, support vector machine algorithm, gradient boosting decision tree algorithm, and multilayer perceptron neural network algorithm.

4. The method for predicting the unconfined compressive strength of cast-in-place fluidized solidified sediment with different clay mineral compositions based on machine learning, as described in claim 1, is characterized in that... Before inputting the data input variables into the model as described in step 3, a normalization preprocessing step is also included for the input variables. The normalization equation is: ； in, X norm For X, the normalized data is... X max The maximum value of the experimental data. X min The minimum value of the experimental data. X This refers to data from a specific experiment.

5. The method for predicting the unconfined compressive strength of cast-in-place fluidized solidified sediment with different clay mineral compositions based on machine learning, as described in claim 1, is characterized in that... The model optimization described in step 4 includes tuning the key hyperparameters of the model. The tuning steps are as follows: among the multiple models initially trained, select the top 3 models in terms of prediction accuracy, and based on each model, use the coefficient of determination R... 2 To improve the accuracy of the indicators, a weighted ensemble strategy is used to enhance the final prediction accuracy. The weight allocation formula is as follows: ； in, w i Let be the weights of the i-th model; R i For the i-th model, R 2 For the accuracy of the indicator.

6. The method for predicting the unconfined compressive strength of cast-in-place fluidized solidified sediment with different clay mineral compositions based on machine learning, as described in claim 2, is characterized in that... The clay mineral types include montmorillonite, illite, kaolinite, and chlorite.

7. The method for predicting the unconfined compressive strength of cast-in-place fluidized solidified sediment with different clay mineral compositions based on machine learning, as described in claim 2, is characterized in that... The degree of coordinated change is evaluated by calculating the Pearson correlation coefficient matrix, and the calculation formula is as follows: ； in, X i , Y i These are the i-th actual values ​​of the two variables, respectively; and , where n is the sample mean of the two variables; n is the sample size; r ranges from [-1, 1]. The closer r is to 1 or -1, the stronger the correlation; the closer r is to 0, the weaker the correlation. Specifically, features with a correlation with unconfined compressive strength in [-1, -0.5] and [0.5, 1] ​​are selected as input variables.

Images