Method, system, equipment and medium for predicting 28-day strength of cement based on chemical components
By using machine learning-based methods and constructing a predictive model using cement chemical composition data, the problems of lag and accuracy in predicting the 28-day strength of cement were solved, enabling real-time quality control and optimization of cement production.
Patent Information
- Application Number
- CN202511791126.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-01
- Publication Date
- 2026-02-27
AI Technical Summary
Existing technologies for predicting the 28-day strength of cement suffer from problems such as long cycles, high costs, and severe lag. Furthermore, traditional empirical formulas are unable to capture the highly nonlinear intrinsic relationship of cement strength, resulting in insufficient prediction accuracy and generalization ability, which cannot meet the needs of intelligent production in the modern cement industry.
By collecting chemical composition characteristics and strength data of historical cement samples, a training dataset is constructed and preprocessed. Then, machine learning regression models such as gradient boosting decision trees, support vector regression, or artificial neural networks are used to establish a mapping relationship from chemical composition to strength, thereby achieving fast and accurate strength prediction.
It enables immediate prediction of cement strength after 28 days, reduces waiting time, improves the ability to monitor and adjust production quality in real time, and enhances prediction accuracy and generalization ability.
Smart Images
Figure CN121583364A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of building materials, in particular to a method, system, device and medium for predicting the 28-day strength of cement based on chemical composition. BACKGROUND
[0002] Cement is the most critical cementitious material in construction engineering. Its mechanical properties, especially the 28-day compressive strength and flexural strength, are the core indicators for evaluating cement quality, designing concrete mix proportion, and ensuring the safety of engineering structures. Currently, the determination of the 28-day strength of cement requires physical experiments in accordance with national standards (such as GB / T17671). Although this method is accurate and reliable, it has some problems such as long cycle (must be cured for 28 days), high cost (consumes a large amount of manpower, materials, and energy), and serious lag. This leads to serious delays in quality control and adjustment during cement production, making it impossible to achieve real-time optimization, and also affecting the efficiency of concrete mix proportion design.
[0003] To solve this problem, some prediction methods based on mathematical statistics and empirical formulas have been proposed in the prior art. For example, linear or nonlinear regression equations between the main components of cement (such as C3S, C2S, C3A, C4AF) and strength are established for prediction. However, the strength of cement is the result of complex coupling of multiple chemical components, mineral compositions, fineness, particle size distribution, and other factors. Simple empirical formulas cannot capture the highly nonlinear internal relationships, and the prediction accuracy and generalization ability are generally poor, which cannot meet the needs of intelligent production in modern cement industry.
[0004] Therefore, there is an urgent need for a new method that can quickly, accurately, and reliably predict the 28-day strength of cement to overcome the drawbacks of traditional experimental methods and traditional empirical formulas. SUMMARY
[0005] The present application provides a method, system, device and medium for predicting the 28-day strength of cement based on chemical composition to solve the above problems.
[0006] The present application is achieved by the following technical solutions:
[0007] A method for predicting the 28-day strength of cement based on chemical composition, comprising:
[0008] S1, collect the chemical composition characteristics of historical cement samples, the corresponding 28-day compressive strength data and flexural strength data measured by standard experiments, construct a training data set, and perform cleaning, normalization or standardization processing on the data in the training data set, wherein the chemical composition characteristics include chemical composition data and corresponding content data;
[0009] S2, input the chemical composition features in the pretreated training data set as input features, and the corresponding compressive strength data and flexural strength data as actual labels;
[0010] S3, constructing a prediction model based on a machine learning regression model, training the prediction model through the pretreated training data set to learn the mapping relationship from the chemical composition features to the strength target;
[0011] S4, inputting the chemical composition features of the cement sample to be predicted into the trained prediction model, and outputting the predicted 28-day compressive strength and flexural strength values.
[0012] As optimization, the chemical composition data includes calcium oxide, silicon dioxide, aluminum oxide, iron oxide, magnesium oxide, sulfur trioxide, alkali content and loss on ignition.
[0013] As optimization, the machine learning regression model is any one of gradient boosting decision tree, support vector regression SVR or artificial neural network ANN; the gradient boosting decision tree includes XGBoost, LightGBM or CatBoost.
[0014] As optimization, the machine learning regression model is a multi-output regression model, and the output data includes the prediction of 28-day compressive strength and 28-day flexural strength, or the machine learning regression model is a single-output regression model, and the output data includes the prediction of 28-day compressive strength or 28-day flexural strength.
[0015] As optimization, the machine learning regression model is an ensemble learning model, which improves the prediction accuracy and generalization ability by combining multiple weak learners.
[0016] As optimization, in the training process of the prediction model, the hyperparameters of the prediction model are optimized by using k-fold cross-validation and grid search or random search method to avoid overfitting and determine the best model parameters.
[0017] The application also discloses a system for predicting 28-day strength of cement based on chemical composition, which is used to execute the method for predicting 28-day strength of cement based on chemical composition.
[0018] The data input module is used for inputting the chemical composition features of the cement sample to be predicted.
[0019] The data processing module is used for performing pretreatment operation on the data input by the data input module, and the pretreatment operation includes cleaning, normalizing or standardizing the chemical composition features.
[0020] A pre-training model storage module is configured to store a pre-trained prediction model, which is the model trained in step S3 of claim 1.
[0021] A prediction calculation module is configured to call the prediction model and perform prediction calculation according to the input data processed by the data processing module.
[0022] A result output module is configured to output and display the 28-day compressive strength and flexural strength values obtained through prediction calculation.
[0023] As an optimization, a model updating module is further included, which is configured to receive newly added chemical composition features of cement samples and corresponding measured 28-day compressive strength and flexural strength data, re-train and update the prediction model, so as to optimize the prediction accuracy of the prediction model.
[0024] The application further discloses an electronic device, including at least one processor and a memory in communication connection with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method for predicting 28-day strength of cement based on chemical composition as described above.
[0025] The application further discloses a storage medium storing a computer program, and the computer program is executed by a processor to implement the method for predicting 28-day strength of cement based on chemical composition as described above.
[0026] Compared with the prior art, the application has the following advantages and beneficial effects:
[0027] The application utilizes a machine learning algorithm, can only according to the chemical composition data of cement, and immediately after the completion of cement production, the long-term mechanical properties of the cement are predicted with high precision, so that the long waiting period is saved, and real-time monitoring and prospective adjustment of production quality are realized. BRIEF DESCRIPTION OF DRAWINGS
[0028] The accompanying drawings, which are included to provide a further understanding of the embodiments of the application and constitute a part of this application, do not limit the embodiments of the application. In the drawings:
[0029] Figure 1 A flowchart of the method for predicting 28-day strength of cement based on chemical composition. DETAILED DESCRIPTION
[0030] In order to make the purpose, technical scheme and advantages of the application clearer, further detailed description of the application will be given below in combination with embodiments and drawings, and the illustrative embodiments of the application and the description thereof are only used to explain the application, and do not limit the application.
[0031] The embodiment 1 provides a method for predicting the 28-day strength of cement based on chemical composition, as shown in the following formula (I): Figure 1
[0032] S1, collecting chemical composition features of historical cement samples, corresponding 28-day compressive strength data and flexural strength data measured by standard experiments, constructing a training data set, and performing cleaning, normalization or standardization processing on the data in the training data set, and retaining reasonable abnormal data caused by cement production process fluctuations (such as abnormal values of ignition loss caused by calcination temperature fluctuations) during the cleaning process, so as to improve the adaptability of the model to actual production scenarios, wherein the chemical composition features include chemical composition data and corresponding content data.
[0033] In some embodiments, the chemical composition data includes calcium oxide CaO, silicon dioxide , aluminum oxide , iron oxide , magnesium oxide MgO, sulfur trioxide , alkali content , and ignition loss.
[0034] S2, taking the pre-processed chemical composition features in the training data set as input features X, and taking the corresponding compressive strength data and flexural strength data as actual labels Y1.
[0035] S3, constructing a prediction model based on a machine learning regression model, training the prediction model through the pre-processed training data set, so as to learn the mapping relationship from the chemical composition features to the strength target.
[0036] The input features X are input into the prediction model, the prediction model outputs the predicted labels Y2, and the loss function is calculated through the city-level labels Y1 and the predicted labels Y2, and the prediction model is adjusted through the loss function.
[0037] In some embodiments, the machine learning regression model is any one of gradient boosting decision tree, support vector regression SVR or artificial neural network ANN; the gradient boosting decision tree includes XGBoost, LightGBM or CatBoost.
[0038] The machine learning regression model is a multi-output regression model, and the output data includes the prediction of the 28-day compressive strength and the 28-day flexural strength, or the machine learning regression model is a single-output regression model, and the output data includes the prediction of the 28-day compressive strength or the 28-day flexural strength, that is, independent single-output regression models are established for the compressive strength and the flexural strength, respectively.
[0039] The machine learning regression model is an ensemble learning model, which improves the prediction accuracy and generalization ability by combining multiple weak learners.
[0040] In some embodiments, during the training process of the prediction model, the hyperparameters of the prediction model are optimized by using k-fold cross-validation and grid search or random search methods to avoid overfitting and determine the best model parameters.
[0041] More specifically, a phased hyperparameter optimization strategy is used during the training process:
[0042] S3.1, the first stage (coarse adjustment stage): for different machine learning regression models (gradient boosting decision tree / XGBoost / LightGBM / CatBoost, support vector regression SVR, artificial neural network ANN), the core hyperparameters (such as the maximum depth of gradient boosting decision tree, learning rate, kernel function parameter of SVR, and the number of hidden layer neurons of ANN) are quickly screened out 3-5 groups of better performance hyperparameter combinations in the preset hyperparameter space using random search method;
[0043] S3.2, the second stage (fine adjustment stage): the hyperparameter combinations screened out in the first stage are fine-tuned using 5-10-fold cross-validation combined with grid search method, and the data set is divided according to the cement production batch (each batch of data as a sub-sample set) during the cross-validation process, to ensure that the optimized model can adapt to the composition fluctuations of different production batches;
[0044] S3.3, hyperparameter constraint adjustment: according to the error requirement of cement strength prediction (such as compressive strength prediction error ≤3%, flexural strength prediction error ≤5%), the fine-tuned hyperparameter combination is verified for constraint, if the cross-validation error corresponding to a certain group of hyperparameters exceeds the preset threshold, the search range of the hyperparameter is adjusted reversely and S32 is executed again, until the optimal hyperparameter combination that meets the error constraint is obtained.
[0045] When the machine learning regression model is a multi-output regression model, a double-objective error weight distribution is used in the hyperparameter optimization process of S3: the weight of compressive strength prediction error is set to 0.6-0.7, and the weight of flexural strength prediction error is set to 0.3-0.4 (based on the priority of compressive strength to structural safety in cement engineering), the comprehensive error of cross-validation is calculated by weighting, to ensure that the optimized prediction model prioritizes the prediction accuracy of core strength indicators.
[0046] When the machine learning regression model is an ensemble learning model, the hyperparameter optimization of S3 further includes weak learner combination strategy optimization: for the correlation of different chemical composition features (such as the correlation coefficient of aluminum oxide and iron trioxide ≥ 0.6), the feature allocation proportion of weak learners is adjusted, so that features with high correlation are processed by the same group of weak learners, and the model complexity caused by feature redundancy is reduced.
[0047] In the fine-tuning stage of S3.2, the optimal number of folds for 5-10 fold cross-validation is selected, which needs to combine the total amount of cement sample data, production batch distribution, and chemical composition fluctuation amplitude three core features, and make progressive decisions according to the following priority rules, to ensure that the number of folds matches the data characteristics and avoid verification bias caused by improper number of folds (such as missing batch characteristics for small number of folds, and insufficient sample size for large number of folds):
[0048] First priority: Determine the fold number basic interval according to the total amount of training data set.
[0049] The total amount of pre-processed training data set (denoted as N) is the primary basis for determining the basic fold range, ensuring that the sample size of each fold meets the minimum requirement for model learning (in the cement strength prediction scenario, the sample size of each fold needs to be ≥ 10 groups to avoid model bias caused by too small single-fold data):
[0050] Total training data set N (groups) Base fold range Core basis N≤100 5-6 folds If the number of folds is greater than 6, the single-fold sample size is less than or equal to 16 groups, which is prone to random fluctuations due to small sample size, resulting in distorted validation errors; 5-6 folds can ensure that the single-fold sample size is 17-20 groups, covering the fluctuations of basic components 100＜N≤300 7-8 folds When the data volume is moderate, 7-8 folds can balance the "representativeness of single-fold samples" and "efficiency of validation times": the single-fold sample size is 14-43 groups, which can cover the characteristics of multiple batches and avoid the validation time redundancy caused by too many folds (such as 9-10 folds) N＞300 9-10 folds Under large sample size, 9-10 folds can reduce random errors through more rounds of validation: the single-fold sample size is greater than or equal to 30 groups, which can fully capture the component fluctuations of different production cycles (such as the CaO content difference caused by monthly raw material changes), and improve the stability of the validation results
[0051] 2. Second priority: Modify the number of folds according to the cement production batch distribution.
[0052] Cement samples have strong batch correlation (composition fluctuation ≤ 3% within the same batch, and fluctuation may reach 5%-10% between different batches), which needs to be further modified within the basic fold interval according to the number of batches (denoted as K) and the sample size of single batch (denoted as , i=1,2...K) to ensure that each fold contains complete batch characteristics and avoid splitting single batch data across folds:
[0053] If there is a large batch ( ) (such as a batch of samples accounting for 20%, N=200, and the batch of 40 groups):
[0054] The number of folds needs to be ≤ 8, and the large batch sample needs to be completely allocated to 1-2 folds (such as 7-fold validation, splitting the 40-group large batch into 2 folds, each with 20 groups, to avoid splitting into 3 folds and causing batch characteristics to be split):
[0055] Example: N=200 (100 For the large batch (17.5%), choose 7 folds (instead of 8 folds), and distribute the 35 large batches into 2 folds (17-18 groups per fold), and evenly distribute the remaining 165 groups into 5 folds (33 groups per fold), ensuring that all folds contain complete batch characteristics.
[0056] If all batch sample sizes n_i < 10% of the total sample size (small batch dispersion distribution):
[0057] The number of folds can be taken as the upper limit in the basic interval (e.g., N=250, basic folds 7-8, choose 8 folds), and more folds are used to cover multiple batch combinations to verify the adaptability of the model to dispersed batches.
[0058] Third priority: determine the number of folds based on the fluctuation range of core chemical components.
[0059] The fluctuation range (denoted as CV, coefficient of variation = standard deviation / mean x 100%) of the core chemical components (CaO, , which has a weight of ≥20% on strength, determines the component difference range that the model needs to cover. The larger the fluctuation, the more folds are needed to verify the stability of the model:
[0060] If the core component CV ≥ 5% (high fluctuation, such as a 6% fluctuation in CaO content due to changes in raw material origin):
[0061] On the basis of the interval + batch correction, choose the upper limit value (e.g., basic interval 7-8 folds, still conform to 7-8 folds after correction, choose 8 folds), and verify more times to ensure that the model still maintains accuracy under high component fluctuation.
[0062] If the core component CV < 3% (low fluctuation, such as stable raw material supply under fluctuation of 2%):
[0063] On the basis of the interval + batch correction, choose the lower limit value (e.g., basic interval 7-8 folds, choose 7 folds), while ensuring verification accuracy, reduce the cost of calculation (e.g., 8 folds more than 7 folds, 30 minutes of single round training of cement model, can save 30 minutes of optimization time).
[0064] To avoid the adaptation bias of the number of folds due to special data distribution (such as extreme abnormal batches, component mutation samples), the following verification needs to be performed after determining the number of folds, if not satisfied, adjust the number of folds in the opposite direction:
[0065] Single-fold sample component coverage rate verification:
[0066] In any one fold sample, the core chemical components (CaO, The value range of ) should cover more than 85% of the total value range of the training dataset; if the coverage of a certain fold is less than 85%, the number of folds should be reduced (e.g., from 8 folds to 7 folds), and the samples should be redistributed to ensure that each fold contains the complete component fluctuation range.
[0067] Folding error stability verification:
[0068] After performing cross-validation at the selected number of folds, calculate the coefficient of variation of the validation error (e.g., MAPE) for each fold. ;like (If the fluctuation in fold error is too large, it indicates that the choice of fold number has led to uneven data distribution.) Then adjust the fold number (e.g., 5 fold → 6 fold, 8 fold → 7 fold) until... This ensures that the verification results are stable.
[0069] S4. Input the chemical composition characteristics of the cement sample to be predicted into the trained prediction model, and output the predicted 28-day compressive strength and flexural strength values.
[0070] Let's take a specific case as an example.
[0071] Case 1:
[0072] Data Collection: 100 sets of historical data were collected from a company. Each set of data included eight chemical components of cement (CaO, ...). , , ,MgO, , (Loss on ignition) and measured 28-day compressive strength (MPa) and flexural strength (MPa).
[0073] Data preprocessing: Missing values were checked and outliers were removed. Then, Z-Score standardization was applied to standardize all features and the target variable.
[0074] Model selection and training: The XGBoost regression algorithm was selected. The dataset was randomly divided into a training set (80 groups) and a test set (20 groups) in an 8:2 ratio. Using the training set, the main hyperparameters of XGBoost (such as learning rate, maximum depth, subsampling ratio, etc.) were tuned through 5-fold cross-validation and grid search to train the optimal model.
[0075] Model evaluation: Evaluate the model performance on the test set. The coefficient of determination between the predicted results and the true values (...). The accuracy of the model is above 0.93, and the mean absolute percentage error (MAPE) is less than 5%, indicating that the model has extremely high prediction accuracy.
[0076] Prediction application: After a new variety of cement is produced, the laboratory provides its chemical composition percentage: CaO = 59.8%, Fe2O3 = 3.6%, MgO = 2.1%, and loss on ignition = 2.3%. Inputting this data into the trained model instantly outputs the prediction results: 28-day compressive strength is 57.8 MPa, and 28-day flexural strength is 8.4 MPa. , , , Fe2O3 = 3.6%, MgO = 2.1%, and loss on ignition = 2.3%. Inputting this data into the trained model instantly outputs the prediction results: 28-day compressive strength is 57.8 MPa, and 28-day flexural strength is 8.4 MPa. , , Fe2O3 = 3.6%, MgO = 2.1%, and loss on ignition = 2.3%. Inputting this data into the trained model instantly outputs the prediction results: 28-day compressive strength is 57.8 MPa, and 28-day flexural strength is 8.4 MPa.
[0077] Next, specific cases are used to illustrate the method of the present application.
[0078] Case 1:
[0079] Data collection: 100 sets of historical data are collected from a company, each set including 8 chemical components of cement (CaO, , , , Fe2O3, MgO, , , and loss on ignition) and measured 28-day compressive strength (MPa) and flexural strength (MPa).
[0080] Data preprocessing: missing value inspection and outlier removal are performed on the data. Subsequently, Z-Score standardization method is used to standardize all features and target variables.
[0081] Model selection and training: XGBoost regression algorithm is selected. The data set is randomly divided into training set (80 groups) and test set (20 groups) in the ratio of 8:2. Using the training set, the main hyperparameters of XGBoost (such as learning rate, maximum depth, subsampling ratio, etc.) are optimized through 5-fold cross-validation and grid search (GridSearch), and the optimal model is trained.
[0082] Model evaluation: the model performance is evaluated on the test set. The determination coefficient (R²) of the prediction results and the true values reaches more than 0.93, and the mean absolute percentage error (MAPE) is less than 5%, indicating that the model has extremely high prediction accuracy.
[0083] Prediction application: After a new variety of cement is produced, the laboratory provides its chemical composition percentage: CaO = 59.8%, Fe2O3 = 3.6%, MgO = 2.1%, and loss on ignition = 2.3%. Inputting this data into the trained model instantly outputs the prediction results: 28-day compressive strength is 57.8 MPa, and 28-day flexural strength is 8.4 MPa. , , Fe2O3 = 3.6%, MgO = 2.1%, and loss on ignition = 2.3%. Inputting this data into the trained model instantly outputs the prediction results: 28-day compressive strength is 57.8 MPa, and 28-day flexural strength is 8.4 MPa. , , Fe2O3 = 3.6%, MgO = 2.1%, and loss on ignition = 2.3%. Inputting this data into the trained model instantly outputs the prediction results: 28-day compressive strength is 57.8 MPa, and 28-day flexural strength is 8.4 MPa.
[0084] The fold selection process of the present application is illustrated based on this specific case (100 groups of historical data (N=80 groups of training set)).
[0085] First priority: N=80≤100, base fold 5-6 fold;
[0086] Second priority: Assuming that the 80 groups of data contain 5 production batches (K=5), the single batch sample size is 12-20 groups, and it is necessary to ensure that the fold is ≤6 fold (to avoid splitting large batches);
[0087] Third priority: CV of core component CaO=4.2% (3%<CV<5%, medium fluctuation), base interval 5-6 fold, select the middle value 5 fold or 6 fold;
[0088] Effectiveness verification:
[0089] If 6 folds are selected, the single fold sample size is 13-14 groups, and the CaO value range is checked: a fold only covers 82% of the total range (not reaching 85%), then adjust to 5 folds;
[0090] The single fold sample size is 16 groups when 5 folds are selected, the CaO coverage rate is 90%, the CV_error of the fold interval MAPE is 8%≤10%, and the final optimal fold number is determined to be 5 folds.
[0091] Similar to the steps of Case 1, Case 2 uses an artificial neural network (ANN) model. The network structure is an input layer (8 neurons), two hidden layers (64 and 32 neurons respectively, using ReLU activation function), and an output layer (2 neurons, corresponding to compressive and flexural strength respectively). After training and optimization, the prediction effect on the test set is also greater than 0.92.
[0092] Embodiment 2 discloses a system for predicting the 28-day strength of cement based on chemical composition, for performing the method for predicting the 28-day strength of cement based on chemical composition described in Embodiment 1, comprising:
[0093] A data input module for inputting the chemical composition characteristics of the cement sample to be predicted;
[0094] A data processing module for performing preprocessing operations on the data input by the data input module, the preprocessing operations including cleaning, normalizing or standardizing the chemical composition characteristics;
[0095] A pre-trained model storage module for storing a pre-trained prediction model, the prediction model being the model trained in step S3 of claim 1;
[0096] a prediction calculation module, configured to invoke the prediction model and perform prediction calculation according to the input data processed by the data processing module;
[0097] a result output module, configured to output and display the 28-day compressive strength and flexural strength values obtained by the prediction calculation.
[0098] In some embodiments, a model updating module is further included, configured to receive newly added chemical composition features of cement samples and corresponding measured 28-day compressive strength and flexural strength data, retrain and update the prediction model to optimize the prediction accuracy of the prediction model.
[0099] Embodiment 3 discloses an electronic device, comprising at least one processor, and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method for predicting 28-day strength of cement based on chemical composition according to Embodiment 1.
[0100] Embodiment 4 discloses a storage medium, storing a computer program, wherein the computer program is executed by a processor to implement the method for predicting 28-day strength of cement based on chemical composition according to Embodiment 1.
[0101] The above detailed description is further detailed for the purpose, technical solutions and beneficial effects of the present application, and it should be understood that the above description is only a specific embodiment of the present application and is not used to limit the protection scope of the present application, and any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application should be included in the protection scope of the present application.
Claims
1. A method for predicting the 28-day strength of cement based on chemical composition, characterized in that, include: S1. Collect the chemical composition characteristics of historical cement samples, the corresponding 28-day compressive strength data and flexural strength data measured by standard experiments, construct a training dataset, and clean, normalize or standardize the data in the training dataset. The chemical composition characteristics include chemical composition data and corresponding content data. S2. Use the chemical composition features in the preprocessed training dataset as input features, and the corresponding compressive strength data and flexural strength data as actual labels; S3. Construct a prediction model based on a machine learning regression model. Train the prediction model using the preprocessed training dataset to learn the mapping relationship from chemical composition features to intensity targets. S4. Input the chemical composition characteristics of the cement sample to be predicted into the trained prediction model, and output the predicted 28-day compressive strength and flexural strength values.
2. The method for predicting the 28-day strength of cement based on chemical composition according to claim 1, characterized in that, The chemical composition data includes calcium oxide, silicon dioxide, aluminum oxide, ferric oxide, magnesium oxide, sulfur trioxide, alkali content, and loss on ignition.
3. The method for predicting the 28-day strength of cement based on chemical composition according to claim 1, characterized in that, The machine learning regression model is any one of gradient boosting decision tree, support vector regression (SVR), or artificial neural network (ANN); the gradient boosting decision tree includes XGBoost, LightGBM, or CatBoost.
4. The method for predicting the 28-day strength of cement based on chemical composition according to claim 1, characterized in that, The machine learning regression model is a multi-output regression model, and the output data includes predictions of 28-day compressive strength and 28-day flexural strength. Alternatively, the machine learning regression model is a single-output regression model, and the output data includes predictions of either 28-day compressive strength or 28-day flexural strength.
5. The method for predicting the 28-day strength of cement based on chemical composition according to claim 1, characterized in that, The machine learning regression model is an ensemble learning model, which improves prediction accuracy and generalization ability by combining multiple weak learners.
6. The method for predicting the 28-day strength of cement based on chemical composition according to claim 1, characterized in that, During the training process of the prediction model, k-fold cross-validation and grid search or random search methods are used to optimize the hyperparameters of the prediction model in order to avoid overfitting and determine the optimal model parameters.
7. A system for predicting the 28-day strength of cement based on chemical composition, for performing the method for predicting the 28-day strength of cement based on chemical composition as described in any one of claims 1-6, characterized in that, include: The data input module is used to input the chemical composition characteristics of the cement sample to be predicted; Data processing module: used to perform preprocessing operations on the data input by the data input module, the preprocessing operations including cleaning, normalizing or standardizing the chemical component characteristics; A pre-trained model storage module is used to store a pre-trained prediction model, wherein the prediction model is the model trained in step S3 of claim 1. The prediction calculation module is used to call the prediction model and perform prediction calculations based on the input data processed by the data processing module. Results output module: Used to output and display the predicted 28-day compressive strength and flexural strength values.
8. The system for predicting the 28-day strength of cement based on chemical composition according to claim 7, characterized in that, It also includes a model update module, which is used to receive the chemical composition characteristics of newly added cement samples and the corresponding measured 28-day compressive strength and flexural strength data, and to retrain and update the prediction model to optimize the prediction accuracy of the prediction model.
9. An electronic device, characterized in that, The method includes at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a method for predicting the 28-day strength of cement based on chemical composition as described in any one of claims 1-6.
10. A storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for predicting the 28-day strength of cement based on chemical composition as described in any one of claims 1-6.