A method and system for accurate evaluation of ready-mixed concrete performance based on the coupling effects of complex factors
By using a multi-dimensional influencing factor library, SHAP contribution value analysis, and Stacking ensemble learning, combined with Bootstrap resampling, the problem of coupling relationships between factors in ready-mixed concrete was solved, achieving high-precision and reliable performance evaluation and supporting engineering quality control.
Patent Information
- Application Number
- CN202611471731.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-21
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies struggle to effectively capture the complex coupling relationships between factors in ready-mixed concrete, resulting in insufficient accuracy in performance prediction and failing to meet the demands of modern engineering for refined quality control of concrete.
We employ a multi-dimensional influencing factor library, SHAP contribution value analysis, Stacking ensemble learning strategy, and Bootstrap resampling to construct a set of key influencing factors and a set of coupled features. By combining Bootstrap resampling to calculate reliability and prediction interval, we construct a comprehensive performance index (CPI).
It significantly improves the accuracy and reliability of concrete performance prediction, provides a reliable range assessment, supports engineering decisions, and facilitates quality control.
Smart Images

Figure CN123069286A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of concrete performance evaluation and machine learning technology, and in particular relates to a method and system for accurate evaluation of ready-mixed concrete performance based on the coupling effects of complex factors. Background Technology
[0002] Ready-mixed concrete, as the most fundamental and widely used building material in construction engineering, directly affects the safety, durability, and service life of engineering structures. The performance of ready-mixed concrete is influenced by a multitude of factors, including raw material quality, production process, environmental conditions, and curing regime. These factors do not act independently but exhibit complex coupled and interactive effects. For example, the water-cement ratio and fly ash content jointly affect concrete strength development; ambient temperature and curing age couple to influence the hydration reaction rate; and mixing process and admixture dosage synergistically affect concrete workability. Traditional single-factor analysis methods or simple linear regression models struggle to capture these complex coupling relationships, resulting in insufficient performance prediction accuracy and failing to meet the demands of modern engineering for refined control of concrete quality.
[0003] In recent years, machine learning technology has been widely used in the field of concrete performance prediction. Existing technologies mostly employ single machine learning models, such as support vector machines, random forests, or backpropagation neural networks, directly using raw material mix proportions and environmental parameters as input for strength prediction. However, existing methods have the following shortcomings: First, the feature engineering process is relatively crude, often directly using raw features without fully considering the coupling and interaction effects between factors, resulting in the loss of a large amount of valuable information; second, the generalization ability of a single model is limited, and overfitting or underfitting is prone to occur when the data distribution is complex or the sample size is insufficient; third, the prediction results only provide single-point estimates, lacking a quantitative assessment of the prediction reliability and uncertainty, making it difficult for engineers to judge the credibility of the prediction results; fourth, there is a lack of a systematic evaluation system, making it impossible to integrate prediction accuracy, reliability, and uncertainty into a comprehensive indicator that can directly guide engineering decisions.
[0004] Therefore, there is an urgent need to develop a precise evaluation method for ready-mixed concrete performance that can fully consider the coupling effects of multiple factors, has high prediction accuracy and reliable quantitative assessment, so as to provide a scientific basis for concrete production quality control and engineering applications. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention provides a method for accurately evaluating the performance of ready-mixed concrete based on the coupled effects of complex factors, comprising the following steps: Establish a multi-dimensional influencing factor database, which includes raw material factors, environmental factors, process factors, and maintenance factors; The importance of each influencing factor was ranked by using a machine learning model combined with SHAP contribution value analysis. Key influencing factors were then selected based on their cumulative contribution rate, and a set of key influencing factors was constructed. The key influencing factors in the set of key influencing factors are coupled by product and ratio to construct a candidate coupling feature pool. The candidate coupling feature pool is then screened using SHAP contribution value analysis to construct a key coupling feature set. Based on the set of key influencing factors and the set of key coupling features, a Stacking performance evaluation model is constructed using a Stacking ensemble learning strategy, and the concrete performance prediction results are output. The Stacking performance evaluation model is modeled N times based on Bootstrap resampling. The concrete performance prediction result is determined based on the final predicted value obtained from the N repeated modeling. The reliability is calculated based on the output value of the first base learner, and the prediction interval is calculated based on the final predicted value. A comprehensive performance index (CPI) is constructed based on the reliability and the prediction range, and the comprehensive performance grade of concrete is determined based on the comprehensive performance index (CPI).
[0006] Optionally, the raw material factors include one or more of cement, fly ash, mineral powder, sand ratio, water, crushed stone, aggregate, water-cement ratio, and admixtures; the environmental factors include one or more of ambient temperature, ambient humidity, and wind speed; the process factors include one or more of mixing time, mixing speed, and transportation time; and the curing factors include one or more of curing age, curing temperature, and curing humidity.
[0007] Optionally, the machine learning model is a tree model, which includes at least one of XGBoost, Random Forest, and LightGBM; the SHAP contribution values are sorted from largest to smallest, and the preceding features whose cumulative contribution rate reaches a preset cumulative contribution rate threshold are retained as key influencing factors to construct a set of key influencing factors.
[0008] Optionally, the product coupling is the product of the values of two key influencing factors, and the ratio coupling is the division of the values of two key influencing factors; the candidate coupling feature pool contains product coupling features and ratio coupling features corresponding to all pairwise combinations of key influencing factors; the step of using SHAP contribution value analysis to screen the candidate coupling feature pool includes: calculating the SHAP contribution value of each coupling feature in the candidate coupling feature pool, sorting them from largest to smallest contribution value, screening key coupling features based on cumulative contribution rate, and constructing a key coupling feature set.
[0009] Optionally, the Stacking performance evaluation model adopts a two-layer structure: the first layer is a base learner layer, which includes multiple heterogeneous regression models, including XGBoost, Random Forest (RF), Support Vector Regression (SVR), and Backpropagation (BP) neural network; the second layer is a meta-learner layer, which adopts one of Ridge Regression, Support Vector Regression, or Linear Regression; the Stacking performance evaluation model is trained using a five-fold cross-validation strategy; the concrete performance prediction results include one of compressive strength, flexural strength, or durability index.
[0010] Optionally, the number of Bootstrap resampling times N ranges from 100 to 1000.
[0011] Optionally, the first layer base learner of the Stacking performance evaluation model has K base models; the outputs of all K base models in each modeling under N Bootstrap resampling are statistically analyzed, and the total mean and total variance of all outputs are calculated; the standard deviation is calculated based on the total variance, and the reliability is calculated based on the standard deviation.
[0012] Optionally, the final prediction results of the Stacking model obtained from N repeated modeling are used to form a prediction set. All predicted values in the set are sorted from smallest to largest, and the 2.5 percentile and 97.5 percentile are taken to form the 95% prediction interval.
[0013] Optionally, the formula for calculating the comprehensive performance index CPI is: ; in, R represents reliability, U represents the uncertainty index corresponding to the prediction interval width, and W represents the prediction interval width. The maximum allowable interval width is preset; weight. , Determined through expert weighting, analytic hierarchy process, entropy weighting, or data-driven methods; Concrete performance is classified into grades based on the Comprehensive Performance Index (CPI): Grade A is when CPI > 90; Grade B is when 75 ≤ CPI ≤ 90; and Grade C is when 60 ≤ CPI < 75.
[0014] This invention also provides a precise performance evaluation system for ready-mixed concrete based on the coupled effects of complex factors, used to implement the aforementioned precise performance evaluation method for ready-mixed concrete, comprising: The data input module is used to input raw material parameters, environmental parameters, process parameters, and curing parameters; The key factor identification module is used to perform SHAP contribution value analysis, screen key influencing factors, and construct a set of key influencing factors; The coupling analysis module is used to construct a candidate coupling feature pool and perform SHAP contribution value analysis to filter key coupling features; The performance evaluation module is used to build and run the Stacking ensemble learning model and output concrete performance prediction results. The predictive reliability assessment module is used to calculate the reliability based on the output of the first-layer base learner, using the Bootstrap resampling output. The prediction interval generation module is used to generate a 95% prediction interval based on the Bootstrap resampling results, quantifying the prediction uncertainty. The results visualization module is used to output and display the ranking of key influencing factors, coupling characteristics, performance prediction results, reliability, prediction range, and comprehensive performance grade evaluation results of concrete.
[0015] Compared with the prior art, the present invention has the following advantages and technical effects: This invention introduces coupling feature engineering, combining product coupling and ratio coupling, to explicitly incorporate the interaction effects between factors into the model input. This effectively captures the complex coupling relationships among various factors, including raw materials, environment, process, and curing, thus improving the accuracy of performance prediction. It employs Stacking ensemble learning to integrate the advantages of multiple heterogeneous base learners, enhancing the model's generalization ability and robustness. By combining Bootstrap resampling with base learner reliability analysis, it provides uncertainty quantification results in both reliability indices and prediction intervals, thus providing a credibility assessment for the prediction results. A comprehensive performance index (CPI) is constructed to organically integrate prediction reliability and uncertainty, forming a graded evaluation result that can directly guide engineering decisions. This facilitates engineers in quickly determining the concrete quality grade and taking corresponding control measures. The systematic evaluation platform covers a complete process including data input, feature selection, coupling analysis, performance prediction, reliability assessment, and result visualization. Each module functions independently yet works collaboratively, offering convenient operation and significant engineering application value. Attached Figure Description
[0016] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of the SHAP-based key influencing factor screening process according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the coupling feature construction and filtering process in an embodiment of the present invention. Figure 3 This is a structural diagram of the Stacking performance evaluation model according to an embodiment of the present invention; Figure 4 This is a structural diagram of the ready-mixed concrete performance accurate evaluation system according to an embodiment of the present invention. Detailed Implementation
[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0018] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.
[0019] Example 1 This embodiment provides a method for accurately evaluating the performance of ready-mixed concrete based on the coupled effects of complex factors, including the following steps: Establish a multi-dimensional influencing factor database, which includes raw material factors, environmental factors, process factors, and maintenance factors; The importance of each influencing factor was ranked by using a machine learning model combined with SHAP contribution value analysis. Key influencing factors were then selected based on their cumulative contribution rate, and a set of key influencing factors was constructed. The key influencing factors in the set of key influencing factors are coupled by product and ratio to construct a candidate coupling feature pool. The candidate coupling feature pool is then screened using SHAP contribution value analysis to construct a key coupling feature set. Based on the set of key influencing factors and the set of key coupling features, a Stacking performance evaluation model is constructed using a Stacking ensemble learning strategy, and the concrete performance prediction results are output. The Stacking performance evaluation model is modeled N times based on Bootstrap resampling. The concrete performance prediction result is determined based on the final predicted value obtained from the N repeated modeling. The reliability is calculated based on the output value of the first base learner, and the prediction interval is calculated based on the final predicted value. A comprehensive performance index (CPI) is constructed based on the reliability and the prediction range, and the comprehensive performance grade of concrete is determined based on the comprehensive performance index (CPI).
[0020] Specifically, the following steps are included: Step 1: Establish a multi-dimensional influencing factor database.
[0021] Considering that the performance of ready-mixed concrete is affected by various factors such as raw material composition, construction environment, production process, and curing conditions, this invention provides a multi-dimensional influencing factor library, divided into raw material factors, environmental factors, process factors, and curing factors, each containing several specific parameters. According to the "Specification for Mix Proportioning Design of Ordinary Concrete" JGJ55-2011, raw material factors include cement, fly ash, mineral powder, sand ratio, water, crushed stone, aggregate, water-cement ratio, and admixtures. Referring to the work of Akber et al., environmental factors include one or more of ambient temperature, ambient humidity, and wind speed. Process factors characterize the potential impact of concrete production and transportation on the uniformity of mixing and subsequent mechanical properties; therefore, process factors include one or more of mixing time, mixing speed, and transportation time. Curing factors characterize the influence of curing conditions on the degree of cement hydration and performance development during concrete hardening; therefore, curing factors include one or more of curing age, curing temperature, and curing humidity.
[0022] This embodiment takes the prediction of compressive strength of ready-mixed concrete as an example to construct a database of influencing factors. Raw material factors include cement dosage, fly ash content, mineral powder content, sand ratio, water dosage, maximum coarse aggregate particle size, water-cement ratio, and water-reducing agent dosage; environmental factors include ambient temperature, relative humidity, and wind speed; process factors include mixing time, mixing speed, and transportation time; curing factors include curing age, curing temperature, and curing humidity. Historical production data containing the above parameters and corresponding compressive strengths are obtained as model training data.
[0023] Step 2: Screening of key influencing factors (e.g.) Figure 1 (As shown).
[0024] The preferred machine learning model is a tree model (such as XGBoost, Random Forest, LightGBM) for SHAP contribution value analysis; the contribution values are sorted from largest to smallest, and the cumulative contribution rate threshold is set to 85%~95%, retaining the preceding features that have reached the cumulative contribution threshold as key influencing factors.
[0025] This embodiment sorts the SHAP contribution values and uses a cumulative contribution rate of 90% as the screening criterion to obtain a set of key influencing factors. The key influencing factors obtained in this embodiment include, but are not limited to, water-cement ratio, curing age, cement dosage, fly ash content, curing temperature, and water-reducing agent dosage.
[0026] The specific algorithm is described below: XGBoost (Extreme Gradient Boosting) is an ensemble learning algorithm based on Gradient Boosting Decision Tree (GBDT). This algorithm iteratively constructs multiple regression trees in each round, using the residual predictions from the previous round as the new learning objective. It combines objective function optimization and regularization strategies to continuously improve model prediction accuracy while suppressing overfitting. This invention uses XGBoost as one of the key influencing factors for analysis and as a Stacking base learner. Its principle is as follows: ; ; In the formula: Represents the loss function; Represents the t-th tree; Represents the number of leaf nodes; Represents the complexity coefficient of the leaf node; Represents the L2 regularization coefficient; This represents the weight of the j-th leaf node.
[0027] Random Forest (RF) is an ensemble learning algorithm based on the Bagging concept. It constructs multiple training subsets through Bootstrap resampling and trains a decision tree on each subset. During decision tree creation, some feature factors are randomly selected for training to improve the model's generalization ability and resistance to overfitting. Finally, the predictions from all decision trees are averaged to obtain the model's final output. This invention uses Random Forest for key influencing factor analysis, and its principle is as follows: Let the prediction result of the k-th decision tree be: ; The final prediction result of the random forest model is: ; In the formula: The number of decision trees to construct.
[0028] LightGBM (Light Gradient Boosting Machine) is an ensemble learning algorithm based on Gradient Boosting Decision Tree (GBDT). It continuously improves the model's prediction accuracy by progressively building multiple decision trees and learning from the residuals of the previous model. It is suitable for regression prediction tasks with large-scale data and high-dimensional features. This invention uses LightGBM as a key influencing factor for analysis, and its principle is as follows: ; ; ; In the formula: The loss function; Represents the t-th decision tree; In the model regularization term, Represents the number of leaf nodes. Represents the complexity coefficient of the leaf node. Represents the L2 regularization coefficient. Represents the weight of the j-th leaf node; the final prediction result of the model. middle, The number of decision trees.
[0029] Step 3: Construction and screening of coupled features.
[0030] like Figure 2 As shown, product coupling is the product of the values of two key influencing factors, and ratio coupling is the division of the values of two key influencing factors; the candidate coupling feature pool contains all pairwise product coupling features and ratio coupling features; the SHAP contribution value analysis calculates the interaction effect strength between features, sorts them from largest to smallest contribution value, and selects coupling features with significant interaction effects based on the cumulative contribution rate to be included in the key coupling feature set.
[0031] Based on the SHAP contribution value analysis results, coupling features with significant interactions were selected as the key coupling feature set. In this embodiment, the key coupling features selected include, but are not limited to, water-cement ratio × curing age, fly ash content / cement content, and curing temperature × curing age.
[0032] SHAP theory is a model interpretability analysis method based on Shapley values in cooperative game theory. It quantitatively evaluates the importance of each influencing factor by calculating the marginal contribution of each feature to the model's prediction results. SHAP can uniformly explain the prediction process of different machine learning models and reflect the positive or negative contribution of each influencing factor to the prediction results. This invention utilizes the SHAP contribution value analysis method to calculate the contribution of different influencing factors to the prediction results of ready-mixed concrete performance. The principle is as follows: The formula for calculating the SHAP value of a single feature is: ; In the formula: For the first SHAP contribution value of each feature; It is the set of all input features; Not including the first Any subset of features; For set The number of features contained therein; The total number of input features; To use only the feature set The prediction results of the time model; To join the Model prediction results after considering each feature; for Remove the first The entire subset of features following each feature.
[0033] The formula for calculating the prediction result that includes multiple input features is as follows: ; In the formula: The number of input features; The predicted mean of all training samples; For the first The contribution value of the i-th feature to the model prediction result, i.e., the i-th feature The SHAP value corresponding to each influencing factor.
[0034] To evaluate the overall contribution of different factors, the absolute values of the SHAP values across all samples are averaged to obtain the feature importance. ; In the formula: For the first The importance measure of a feature is such that a larger value indicates a greater impact of that feature on the prediction results during the model prediction process. This represents the total number of samples. For the first In the nth sample The SHAP contribution value corresponding to each feature.
[0035] Step 4: Stacking Ensemble Learning Performance Evaluation Modeling (e.g.) Figure 3 (As shown).
[0036] The Stacking performance evaluation model employs a two-layer structure: the first layer is a base learner layer, consisting of multiple heterogeneous regression models with different learning mechanisms. These base learners include, but are not limited to, gradient boosting tree models, Bagging models, support vector regression models, and artificial neural network models, with XGBoost, Random Forest, SVR, and BP neural networks preferably used as the first-layer base learners. The second layer is a meta-learner layer, employing one of Ridge Regression, SVR, or Linear Regression. The model training uses a five-fold cross-validation strategy, and the predicted concrete performance results include, but are not limited to, one of the following performance indicators: compressive strength, flexural strength, and durability.
[0037] In this embodiment, a Stacking performance evaluation model is constructed using a set of key influencing factors and a set of key coupling features. After the model is trained, predictions are made on the test samples, and the predicted values of concrete compressive strength are output. Compared with a single model, the Stacking model can fully integrate the predictive advantages of multiple models, improving the model's prediction accuracy and generalization ability.
[0038] Backpropagation (BP) neural networks are a typical type of feedforward neural network. They continuously adjust network weights and biases through backpropagation, gradually approximating the true values of the model's predictions. A typical BP neural network consists of an input layer, hidden layers, and an output layer. Neurons in each layer are connected by weights to fit the complex nonlinear relationship between input features and output results. This invention uses a BP neural network as one of the first-layer base learners in a stacking model, and its principle is as follows: In the formula: Input features; Output for hidden layer; For input layer weights, First hidden layer bias; For activation functions; Output to the output layer; This is the loss function.
[0039] Support Vector Regression (SVR) is a regression algorithm based on statistical learning theory. It constructs an optimal hyperplane to control prediction error within a given tolerance range and achieves good generalization ability while minimizing model complexity. This invention uses SVR as one of the candidate base learners in the Stacking model, and its principle is as follows: ; In the formula: This is the basic prediction function for the SVR model; The input feature vector; This is the model weight vector; For bias.
[0040] To improve the model's generalization ability, SVR establishes an optimization objective function by minimizing structural risk: ; In the formula: For penalty parameters; This represents the number of training samples; For the first The slack variables corresponding to each sample.
[0041] Step 5: Calculation of concrete performance prediction results, reliability, and prediction range.
[0042] The Bootstrap resampling is performed N times, resulting in N Stacking performance evaluation models, where N ranges from 100 to 1000; the calculation method for the concrete performance prediction results is as follows: Let the first The output of each model is There are a total of If a model outputs, then: ; in This is the result of concrete performance prediction.
[0043] The first layer of the Stacking performance evaluation model contains K base models. The reliability is calculated as follows: Let the first During the Bootstrap resampling, its first layer... The output of each model is The output will be: ; ; ; in Output the mean of the first layer model. Output the variance for the first layer of the overall model. For reliability.
[0044] The prediction interval is calculated as follows: Let the final prediction result of the Stacking model after the nth Bootstrap resampling be... There are a total of The prediction results constitute a set. = , will set After sorting the elements in ascending order, the 2.5th percentile and the 97.5th percentile form the 95% prediction interval. = .
[0045] In this embodiment, the Bootstrap resampling technique is used for 200 repeated modeling iterations. Each time, a subset with replacement is drawn from the original training set with the same sample size, and a complete Stacking model is trained on this subset, resulting in a total of 200 independent Stacking models.
[0046] The final prediction result of the model is obtained by calculation according to the formula. In this example, the prediction of compressive strength is taken as the standard.
[0047] The prediction reliability of each test sample is calculated according to the formula. The higher the reliability, the higher the stability of the prediction result of the first base learner and the better the model prediction stability.
[0048] The 95% prediction interval is calculated based on the multiple final prediction values obtained from Bootstrap, which is used to characterize the uncertainty of the model's prediction results.
[0049] Stacking is a multi-layer ensemble learning method that constructs multiple base learners with different learning capabilities, enabling different models to learn input data features from different perspectives. A meta-learner then performs secondary calculations on the outputs of each base learner, improving prediction accuracy and stability. This invention uses a Stacking model to establish a predictive model for ready-mixed concrete performance. The first layer is the base learner layer, preferably using XGBoost, Random Forest, SVR, or BP neural networks. The second layer is the meta-learner layer, employing one of Ridge Regression, Support Vector Regression (SVR), or Linear Regression. Its principle is as follows: ; ; ; In the formula: For the first The predicted output of each base learner; Output the matrix formed by the first layer model; This represents the final prediction result of the Stacking model; The second-level meta-learner uses one of the following: Ridge regression, Support Vector Regression (SVR), or Linear Regression.
[0050] Step 6: Overall performance rating assessment.
[0051] Overall performance indicators The calculation formula is: Where R represents reliability, and U represents the uncertainty index corresponding to the prediction interval width. To calculate the interval width, The maximum allowable interval width is used to calculate the score; the narrower the interval, the higher the score. The weights... , It can be determined through expert weighting, analytic hierarchy process (AHP), entropy weighting, or data-driven methods.
[0052] In this embodiment, the Analytic Hierarchy Process (AHP) is used to determine the weights: (Reliability weight) (Uncertainty index weight). The uncertainty index U is calculated based on the prediction interval width. The narrower the range, the higher the score. Substitute into the formula CPI = 0.6×R×100 + 0.4×U to calculate the comprehensive performance index.
[0053] The comprehensive performance index (CPI) is calculated based on the prediction reliability and prediction range, and the comprehensive performance is evaluated according to the preset level standards. A higher CPI indicates a more reliable prediction result and better overall performance.
[0054] The overall performance level is divided into three levels: Grade A: >90, predicts stability, high reliability, and excellent performance; Grade B: 75 ≤ ≤ 90, performance meets requirements, but prediction range is wide; Grade C: 60 ≤ <75 indicates low prediction reliability; re-examination or supplementary data is recommended.
[0055] This embodiment also provides a corresponding accurate performance evaluation system for ready-mixed concrete, with a module structure as follows: Figure 4 As shown, it includes: The data input module is used to input raw material parameters, environmental parameters, process parameters, and curing parameters; The key factor identification module is used to perform SHAP contribution value analysis, screen key influencing factors, and construct a set of key influencing factors; The coupling analysis module is used to construct a candidate coupling feature pool and perform SHAP contribution value analysis to filter key coupling features; The performance evaluation module is used to build and run the Stacking ensemble learning model and output concrete performance prediction results. The predictive reliability assessment module is used to calculate the reliability based on the output of the first-layer base learner, using the Bootstrap resampling output. The prediction interval generation module is used to generate a 95% prediction interval based on the Bootstrap resampling results, quantifying the prediction uncertainty. The results visualization module is used to output and display the ranking of key influencing factors, coupling characteristics, performance prediction results, reliability, prediction range, and comprehensive performance grade evaluation results of concrete.
[0056] The system is developed using a B / S architecture, allowing users to access and use it via a browser. The data input module supports both batch import from Excel and single-sample form input. The key factor identification module incorporates a SHAP calculation engine, enabling interactive adjustment of screening thresholds and real-time updates to the key factor set. The coupling analysis module automatically generates coupling features and performs interaction effect analysis, outputting a heatmap visualization. The performance evaluation module provides a visual model configuration interface, supporting base learner selection and parameter tuning. The prediction reliability evaluation module and prediction interval generation module perform calculations automatically in the background. The results visualization module integrates a dashboard display interface, intuitively presenting various evaluation results in chart form, and supports one-click export of evaluation reports in PDF format.
[0057] The present invention achieves the following effects: First, this invention introduces coupled feature engineering to explicitly incorporate the interaction effects between factors into the model input, effectively capturing the complex coupling relationships among various factors such as raw materials, environment, process, and maintenance, thus significantly improving the accuracy of performance prediction. Compared to traditional methods that directly use original features, the introduction of coupled features can uncover deeper-level influence patterns.
[0058] Second, this invention adopts a stacking ensemble learning strategy, which combines the advantages of multiple heterogeneous base learners. Compared with a single model, it has stronger generalization ability and robustness, and can adapt to concrete performance datasets with different distribution characteristics, thus avoiding the bias of a single model.
[0059] Third, this invention innovatively combines Bootstrap resampling with base learner reliability analysis, and provides uncertainty quantification results in two dimensions: reliability index and prediction interval. This makes the prediction results no longer isolated point estimates, but interval assessments with credibility, which better meet the decision-making needs of actual engineering.
[0060] Fourth, the CPI comprehensive performance index system constructed in this invention organically integrates prediction reliability and uncertainty, forming a graded evaluation result that can directly guide engineering decisions, making it easier for engineers to quickly determine the quality grade of concrete and take corresponding control measures.
[0061] Fifth, the systematic evaluation platform provided by this invention covers the entire process from data input, feature selection, coupling analysis to performance prediction, reliability assessment and result visualization. Each module is independent yet works collaboratively, is easy to operate, and has good engineering application value.
[0062] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for accurately evaluating the performance of ready-mixed concrete based on the coupled effects of complex factors, characterized in that, Includes the following steps: Establish a multi-dimensional influencing factor database, which includes raw material factors, environmental factors, process factors, and maintenance factors; The importance of each influencing factor was ranked by using a machine learning model combined with SHAP contribution value analysis. Key influencing factors were then selected based on their cumulative contribution rate, and a set of key influencing factors was constructed. The key influencing factors in the set of key influencing factors are coupled by product and ratio to construct a candidate coupling feature pool. The candidate coupling feature pool is then screened using SHAP contribution value analysis to construct a key coupling feature set. Based on the set of key influencing factors and the set of key coupling features, a Stacking performance evaluation model is constructed using a Stacking ensemble learning strategy, and the concrete performance prediction results are output. The Stacking performance evaluation model is modeled N times based on Bootstrap resampling. The concrete performance prediction result is determined based on the final predicted value obtained from the N repeated modeling. The reliability is calculated based on the output value of the first base learner, and the prediction interval is calculated based on the final predicted value. A comprehensive performance index (CPI) is constructed based on the reliability and the prediction range, and the comprehensive performance grade of concrete is determined based on the comprehensive performance index (CPI).
2. The method for accurate evaluation of ready-mixed concrete performance based on the coupled effects of complex factors as described in claim 1, characterized in that, The raw material factors include one or more of cement, fly ash, mineral powder, sand ratio, water, crushed stone, aggregate, water-cement ratio, and admixtures; the environmental factors include one or more of ambient temperature, ambient humidity, and wind speed; the process factors include one or more of mixing time, mixing speed, and transportation time; and the curing factors include one or more of curing age, curing temperature, and curing humidity.
3. The method for accurate evaluation of ready-mixed concrete performance based on the coupled effects of complex factors as described in claim 1, characterized in that, The machine learning model is a tree model, which includes at least one of XGBoost, Random Forest, and LightGBM. The SHAP contribution values are sorted from largest to smallest, and the preceding features whose cumulative contribution rate reaches the preset cumulative contribution rate threshold are retained as key influencing factors to construct a set of key influencing factors.
4. The method for accurate evaluation of ready-mixed concrete performance based on the coupled effects of complex factors as described in claim 1, characterized in that, The product coupling is the product of the values of two key influencing factors, and the ratio coupling is the division of the values of two key influencing factors; the candidate coupling feature pool contains the product coupling features and ratio coupling features corresponding to all pairwise combinations of the key influencing factors. The step of using SHAP contribution value analysis to screen the candidate coupling feature pool includes: calculating the SHAP contribution value of each coupling feature in the candidate coupling feature pool, sorting them from largest to smallest contribution value, screening key coupling features based on cumulative contribution rate, and constructing a key coupling feature set.
5. The method for accurately evaluating the performance of ready-mixed concrete based on the coupled effects of complex factors as described in claim 1, characterized in that, The Stacking performance evaluation model adopts a two-layer structure: the first layer is the base learner layer, which includes multiple heterogeneous regression models, including XGBoost, Random Forest (RF), Support Vector Regression (SVR), and Backpropagation Neural Network. The second layer is the meta-learner layer, which uses one of ridge regression, support vector regression, or linear regression; the Stacking performance evaluation model is trained using a five-fold cross-validation strategy; the concrete performance prediction results include one of compressive strength, flexural strength, or durability index.
6. The method for accurate evaluation of ready-mixed concrete performance based on the coupled effects of complex factors according to claim 1, characterized in that, The number of Bootstrap resampling times N ranges from 100 to 1000.
7. The method for accurate evaluation of ready-mixed concrete performance based on the coupled effects of complex factors as described in claim 6, characterized in that, The Stacking performance evaluation model has K base models in the first layer base learner. The outputs of all K base models in each modeling under N Bootstrap resampling are statistically analyzed, and the total mean and total variance of all outputs are calculated. The standard deviation is calculated based on the total variance, and the reliability is calculated based on the standard deviation.
8. The method for accurate evaluation of ready-mixed concrete performance based on the coupled effects of complex factors as described in claim 6, characterized in that, The final prediction results of the Stacking model obtained from N repeated modeling are used to form a prediction set. All prediction values in the set are sorted from smallest to largest, and the 2.5 percentile and 97.5 percentile are taken to form the 95% prediction interval.
9. The method for accurate evaluation of ready-mixed concrete performance based on the coupled effects of complex factors as described in claim 1, characterized in that, The formula for calculating the comprehensive performance index (CPI) is as follows: ; in, R represents reliability, U represents the uncertainty index corresponding to the prediction interval width, and W represents the prediction interval width. The maximum allowable interval width is preset; weight. , Determined through expert weighting, analytic hierarchy process, entropy weighting, or data-driven methods; Concrete performance is classified into grades based on the Comprehensive Performance Index (CPI): Grade A is when CPI > 90; Grade B is when 75 ≤ CPI ≤ 90; and Grade C is when 60 ≤ CPI < 75.
10. A precise performance evaluation system for ready-mixed concrete based on the coupled effects of complex factors, used to implement the precise performance evaluation method for ready-mixed concrete as described in any one of claims 1-9, characterized in that, include: The data input module is used to input raw material parameters, environmental parameters, process parameters, and curing parameters; The key factor identification module is used to perform SHAP contribution value analysis, screen key influencing factors, and construct a set of key influencing factors; The coupling analysis module is used to construct a candidate coupling feature pool and perform SHAP contribution value analysis to filter key coupling features; The performance evaluation module is used to build and run the Stacking ensemble learning model and output concrete performance prediction results. The predictive reliability assessment module is used to calculate the reliability based on the output of the first-layer base learner, using the Bootstrap resampling output. The prediction interval generation module is used to generate a 95% prediction interval based on the Bootstrap resampling results, quantifying the prediction uncertainty. The results visualization module is used to output and display the ranking of key influencing factors, coupling characteristics, performance prediction results, reliability, prediction range, and comprehensive performance grade evaluation results of concrete.
