CN115497574A - A method and system for predicting compressive strength of HPC based on model fusion - Google Patents

Abstract

The invention discloses a HPC compressive strength prediction method and a HPC compressive strength prediction system based on model fusion, which comprises the steps of collecting relevant parameter data of high-performance concrete; exploratory data analysis and data cleaning; processing abnormal values of concrete data and transforming the data; concrete data characteristic engineering; constructing a concrete compressive strength prediction model; optimizing parameters of a concrete compressive strength prediction model; fusing a concrete prediction compressive strength prediction model; performing interpretability analysis on a concrete compressive strength prediction model based on SHAP; by utilizing the method and the system, the defects that the traditional neural network model is difficult to train and has high requirements on data quantity are overcome; meanwhile, the comprehensive average of the results of multiple models is more reliable than the prediction result of a single model; meanwhile, the method has the advantages of short test period, high precision and low test cost, and has stronger engineering feasibility compared with the traditional empirical formula method and test method.

Description

A HPC compressive strength prediction method and system based on model fusion

technical field

The invention belongs to the technical field of large-scale caisson construction, and in particular relates to an HPC compressive strength prediction method and system based on model fusion.

Background technique

HPC (High Performance Concrete) has been widely used in the construction of long-span bridges due to its outstanding characteristics such as high strength and high durability. The compressive strength of concrete is an important indicator of concrete quality evaluation, which largely reflects the safety performance of building structures. Therefore, it is of great significance to study the accurate prediction method of the compressive strength of high performance concrete for the precise control of construction projects and the scientific evaluation of engineering projects.

At present, the methods for compressive strength prediction mainly include: formula method based on experience, test method and method based on statistical machine learning.

Among them, the experience-based formula method is based on artificial experience, through the establishment of complex mathematical models to fit the various parameters of concrete, so as to establish a calculation model for compressive strength. This type of method is highly dependent on manual experience, the iterative calculation process is complex, the fitting accuracy is extremely limited, and it is not suitable for the calculation of the compressive strength of materials such as high-performance concrete with many types of mixed materials and extremely complex formula content.

Based on the test method, various test instruments and equipment are used to monitor the structure of the high-performance concrete material before and after forming, so as to obtain the compressive strength. This type of method has high test accuracy and high reliability of test results; however, this type of method has a long test cycle, high test cost, and a high risk factor when testing in a complex construction environment on site. It is widely used for compressive strength testing in laboratory environment.

The method based on statistical machine learning is based on machine learning. Through data-driven methods, without too many assumptions, the compressive strength prediction model is directly established. It has low cost, short test cycle, and high accuracy of test results. Therefore, It has high research and application value.

So far, some studies have used machine learning methods to predict the compressive strength of concrete, such as based on AdaBoost algorithm, based on random forest and intelligent algorithm, based on BP neural network or RBF neural network, based on support vector machine (SVM), based on linear Regression (LR), methods based on deep learning (DL), etc.; however, the above methods still have some defects. For example, the above methods are based on the output of a single model to predict the compressive strength, which lacks certain reliability; Neural network or deep learning methods especially rely on a large amount of experimental data, and are not suitable for high-risk and difficult data collection scenarios such as civil engineering. At the same time, the model training of such methods is difficult, and it is easy to fall into local optimum or model overfitting However, the SVM or LR-based methods are extremely susceptible to the influence of outliers, so that the prediction accuracy is quite different from the actual results, and it is difficult for the LR method to fit the complex linear relationship between the various components of HPC. At the same time, the existing prediction methods are generally a black-box model, which lacks interpretability and cannot clearly know the specific impact of each data sample on the compressive strength, which is not conducive to specific guidance for actual engineering projects.

Contents of the invention

Therefore, in view of the problems that the current HPC (high performance concrete) compressive strength prediction method is highly dependent on manual experience, requires a large amount of data, the model training process is complicated, the accuracy of model prediction results is not high, and the model prediction results lack interpretability. The invention proposes a method for accurate prediction of the compressive strength of ordinary concrete, especially HPC.

A kind of HPC compressive strength prediction method based on model fusion of one of object of the present invention comprises the following steps:

S1. According to the collected historical parameter data of HPC, the first prediction model and the second prediction model are respectively trained, and the first prediction model and the second prediction model for predicting the compressive strength of HPC are respectively obtained after training; the HPC i.e. high performance concrete;

S2. Composite calculation is performed on the trained first prediction model and the second prediction model to obtain an HPC compressive strength prediction fusion model, and the HPC compressive strength prediction fusion model outputs a final HPC compressive strength prediction result.

A further technical solution includes, step S3 is also included after the step S2:

S3. Using the first algorithm to interpret and analyze the fusion model for predicting the HPC compressive strength, and obtain the contribution value of the component content of each HPC parameter to the HPC compressive strength output by the fusion model for predicting the HPC compressive strength; The above contribution value is used to measure whether the content of each component of HPC has the effect of enhancing the compressive strength of HPC on the predicted value of compressive strength of HPC, which is used to guide the actual concrete mix design.

A further technical solution includes: the first algorithm is an interpretability algorithm based on SHAP. The contribution value is represented by Shapley Value and abbreviated as Ψ, which is defined as follows:

In the formula:

S is the feature subset input by the model to be explained, that is, the concrete parameter set input to the fusion model of HPC compressive strength prediction;

x _j is the jth characteristic variable of the sample to be explained; that is, the jth concrete parameter;

p is the total number of features, that is, the total number of concrete parameters;

val _x (S) indicates the prediction result of the model for sample x when S is the input feature, that is, the compressive strength result output by the model, where x is the sample, and the elements of x are recorded as x _i , and x _i is the i-th The value corresponding to the characteristic variable;

The SHAP value indicates the importance of the jth feature to the model output, that is, the marginal contribution, and the ShapleyValue is the mean value of each marginal contribution. The model interpretation results of the fusion model for HPC compressive strength prediction are defined as follows:

In the formula:

g is the compressive strength prediction model to be explained, that is, the fusion model of HPC compressive strength prediction;

z'∈{0,1} ^M is a combination vector, representing whether the feature z _j (j∈[1,M]) exists, where z _j is the jth concrete input to the prediction model of the fused HPC compressive strength parameter, z' is used to identify whether z ₁ ~ z _M exists in the parameter set input to the fusion model of HPC compressive strength prediction;

M is the number of combined features, that is, the number of concrete parameters input to the fusion model for HPC compressive strength prediction;

为特征j的特征归因Shapley Value，即第j个参数对HPC抗压强度预测的融合模型的预测结果，即对抗压强度的贡献值；

Attributing the Shapley Value to the characteristic of feature j, that is, the prediction result of the fusion model of the jth parameter to the HPC compressive strength prediction, that is, the contribution value to the compressive strength;

Ψ ₀ is the average prediction result of the fusion model for HPC compressive strength prediction, that is, the mean value of the compressive strength prediction results.

Shapley Value measures the contribution of the feature to the overall prediction result. When Ψ _j > 0, it indicates that the feature has a positive effect on improving the predictive value of compressive strength, that is, it has the effect of enhancing the compressive strength of HPC.

The SHAP global feature importance is the average of the sum of the absolute value of the Shapley Value of each feature, that is

A further technical solution includes: in the step S2, the method of obtaining the fusion model for predicting the HPC compressive strength includes performing compound calculations on the first prediction model and the second prediction model by using a weighted average method.

A further technical solution includes: the method for compounding the first prediction model and the second prediction model by using the weighted average method includes:

minimize e(Loss)stw ₁ +w ₂ ＝1 and w ₁ ≥0,w ₂ ≥0

In the formula:

w ₁ represents the weight of the first prediction model;

w ₂ represents the weight of the second prediction model;

Loss is the loss function of the fusion model H(x) for HPC compressive strength prediction; its calculation method is as follows:

In the formula:

N is the sample size, that is, the total number of concrete sample data collected;

为第i条样本对应的混凝土实际抗压强度；

is the actual compressive strength of concrete corresponding to the i-th sample;

为HPC抗压强度预测的融合模型H(x)对第i条样本的混凝土抗压强度输出的预测值；

is the predicted value of the concrete compressive strength output of the i-th sample by the fusion model H(x) predicted by HPC compressive strength;

The expression of H(x) is as follows:

In the formula:

H(x): indicates the final predicted HPC compressive strength output by the fusion model of HPC compressive strength prediction;

w ₁ , w ₂ : represent the weights of the first prediction model and the second prediction model respectively;

h ₁ , h ₂ : represent the compressive strength of concrete predicted by the first prediction model and the second prediction model, respectively.

A further technical solution includes: the first prediction model predicts the compressive strength of HPC based on the AdaBoost algorithm.

A further technical solution includes: the second prediction model predicts the compressive strength of HPC based on the CatBoost algorithm.

A further technical solution includes: when using the weighted average method to fuse the first prediction model and the second prediction model, the weight of the second prediction model based on the CatBoost algorithm is greater than the weight of the first prediction model based on the AdaBoost algorithm.

A further technical solution includes: before the step S2, it also includes using a Bayesian optimization method to optimize the hyperparameters of the first prediction model and the first prediction model; and perform cross-validation on the model after parameter adjustment; the The hyperparameters include the tree and the depth, and after tuning the hyperparameters, the tuned first prediction model and the second prediction model for predicting the HPC compressive strength are obtained.

Both the AdaBoost model and the CatBoost model are tree-based integrated models. The base model of both is a tree (Decision Tree), that is, many base models (that is, many trees) together form the integrated model AdaBoost and CatBoost. AdaBoost and CatBoost are as a whole Based on the Boosting integrated learning framework. The trees of AdaBoost model and CatBoost model are the number of Decision Tree trees; the depth of AdaBoost model and the depth of CatBoost model are the number of layers of AdaBoost model and the number of layers of CatBoost model;

A further technical solution includes: Step S1 also includes obtaining new characteristic parameters for the collected historical parameter data of the HPC by means of feature construction, which is used to expand the data set and make the predicted compressive strength of the HPC more accurate; the feature The construction method is to combine engineering experience with mathematical calculation of different concrete parameters to obtain the ratio relationship of different concrete parameters.

A kind of HPC compressive strength prediction system based on model fusion that realizes the second object of the present invention, including a model training module and a model composite operation module;

The model training module is used to train the first model and the second model respectively according to the historical parameter data of the collected HPC, and respectively obtain the first prediction model and the second prediction model for predicting the compressive strength of the HPC that have been trained;

The model composite operation module is used to perform composite operations on the HPC compressive strength output by the first prediction model for predicting the HPC compressive strength and the second prediction model that have been trained to obtain a fusion model for predicting the HPC compressive strength, HPC The fusion model of compressive strength prediction outputs the final prediction result of HPC compressive strength.

Further, in the model composite operation module, a weighted average method is used to perform composite operations on the HPC compressive strength output by the first prediction model and the second prediction model.

Further, it also includes a parameter tuning module, which is used to tune the hyperparameters of the first prediction model and the second prediction model for predicting the compressive strength of HPC after training respectively, so as to obtain the first prediction model and the second prediction model after parameter optimization. The second predictive model is to perform cross-validation on the optimized compressive strength predictive model at the same time; the cross-validation method includes five-fold cross-validation;

Further, it also includes a model interpretation and analysis module, which is used to interpret and analyze the fusion model of the HPC compressive strength prediction by using the first algorithm, so as to obtain the influence of the component content of each HPC parameter on the HPC compressive strength.

Further, it also includes an outlier processing module, which is used to detect outliers in the historical parameter data of the collected concrete; the method for outlier detection includes using an algorithm based on the combination of K-Means++ clustering and isolated forest .

The K-Means++ algorithm steps are as follows:

a) Initialize an empty set M to store the initial cluster center;

b) randomly select the first cluster center μ ^(j ) from the initial sample, and assign it to the set M;

c) For each sample x ⁽ⁱ⁾ (the sample does not belong to the set M), find the minimum square distance d(x ⁽ⁱ⁾ , M) ² between it and all initial cluster centers in the set M;

随机选择下一个质心μ^(p)；d) Based on weighted probability distribution

Randomly select the next centroid μ ^(p) ;

e) Repeat steps b and c above until K cluster centers are selected;

f) Based on the above set M, continue to use the classic K-Means algorithm;

g) Select the K-Means model with the best performance according to SSE, that is, the residual sum of squares, so as to obtain the best cluster center;

Combined with the K cluster centers obtained above, the outlier detection of the data set is performed using the isolation forest algorithm.

The isolated forest algorithm is an outlier detection algorithm based on division and ensemble learning. If you directly use the isolated forest algorithm for outlier detection without the previous cluster analysis, you will face a large amount of calculation, a long running cycle, and the artificiality of the division process. strong question. Data clustering analysis based on the K-Means++ algorithm can greatly improve the detection efficiency of the isolation forest algorithm.

Beneficial effect:

(1) The concrete compressive strength prediction model based on integrated learning is established, and the integrated model is further integrated. This group decision-making modeling method not only improves the prediction accuracy of the model, but also overcomes the traditional neural network model. It is difficult to train and has a high demand for data volume; at the same time, compared with the prediction results of a single model, the comprehensive average of multi-model results is more reliable;

(2) The HPC compressive strength prediction method based on the statistical machine learning method has a short test period, high precision, and low test cost, and is more engineering feasible than the traditional empirical formula method and test method;

(3) The present invention combines model fusion with SHAP interpretability algorithm for the prediction of concrete compressive strength, overcomes the unpredictability and black-box nature of the traditional modeling process, and combines SHAP interpretability analysis as The development of concrete engineering provides convenience and is conducive to a more accurate understanding of the specific impact of each component on the compressive strength;

(4) The outlier processing process of concrete parameter data adopts the combination of K-Means++ clustering and isolation forest, which overcomes the high computational complexity and artificial dependence of data division faced by the direct use of isolation forest method. ;

(5) The present invention realizes the expansion of the data set by means of feature engineering, and newly constructs proportional features such as water-cement ratio and water-cement ratio, which is beneficial to avoid model overfitting.

Description of drawings

Fig. 1 is the modeling flowchart of HPC compressive strength prediction model of the present invention;

Fig. 2 is HPC compressive strength prediction model model fitting effect figure one of the present invention;

Fig. 3 is HPC compressive strength prediction model model fitting effect figure two of the present invention;

Fig. 4 is the schematic diagram of the fusion process of the HPC compressive strength prediction model model of the present invention;

Fig. 5 is a schematic diagram 1 of the single-sample interpretation result of the SHAP model interpretability algorithm of the present invention;

Fig. 6 is the second schematic diagram of the single-sample interpretation result of the SHAP model interpretability algorithm of the present invention;

FIG. 7 is a schematic diagram of the interpretation results of the SHAP model interpretability algorithm of the present invention on the entire data set.

detailed description

The following specific implementation methods are used to explain the technical solutions of the claims of the present invention, so that those skilled in the art can understand the claims. The protection scope of the present invention is not limited to the following specific implementation structures. The protection scope of the present invention includes the technical solution of the claims of the present invention made by those skilled in the art and is different from the following specific embodiments.

As shown in Figure 1, this embodiment includes the following steps:

其中水泥含量、粉煤灰含量、矿渣含量、减水剂含量、粗/细骨料含量、水含量的单位均为(kg/m³)，即每立方米混凝土中对应成分的质量；养护期的单位是day(天数)、温度单位是℃(摄氏度)、坍落度单位是mm(毫米)，均可在配置混凝土时称重或测量得到；Step 1. Collect high-performance concrete related parameter data. Collect concrete related parameter data at concrete factories, mixing stations, etc., and the concrete related parameter data includes but not limited to cement content, fly ash content, slag content, water reducer content, Coarse/fine aggregate content, water content, curing period, temperature, slump, constitute a sample set

The units of cement content, fly ash content, slag content, superplasticizer content, coarse/fine aggregate content, and water content are (kg/m ³ ), that is, the mass of the corresponding components in each cubic meter of concrete; curing period The unit of temperature is day (number of days), the unit of temperature is ℃ (degree Celsius), and the unit of slump is mm (millimeter), which can be obtained by weighing or measuring when configuring concrete;

从而构建实验数据集

数据构建完成后，将数据集存放于本地磁盘或者关系型数据库中；At the same time, the actual compressive strength corresponding to each sample is used as the target variable

To build the experimental data set

After the data construction is completed, store the dataset in a local disk or a relational database;

Step 2. Exploratory data analysis and data cleaning

Conduct initial exploration and understanding of datasets using visualization and statistical analysis. Combined with the visual analysis results, the missing values, repeated values, abnormal values, etc. in the collected high-performance concrete related parameter data are processed, and the data distribution is understood at the same time. For the characteristic variables with few missing values (within 10%) and the corresponding characteristic extreme values are not much different, the missing values are filled with the mean value; for the characteristic variables with few missing values and the corresponding characteristic extreme values are greatly different, the missing values Median filling is used; for the characteristic variables with a missing value ratio of about 10%-50%, the missing values are predicted and filled based on the decision tree algorithm; the characteristic variables with a missing value ratio of more than 50% are eliminated; the data Centralized duplicate values are deleted;

Step 3. Concrete data outlier processing and data transformation

Step 3.1, data transformation

Perform normalization processing on the eigenvariables with large differences in eigenvalue extreme values. The normalization processing described in this embodiment is processed by the Robust Scaler method that is robust to outliers. The steps of the method are as follows:

分位数，其中移除

分位数(即中位数)，然后存储相应分位数；a) Calculate the data to be processed

quantile, where remove

Quantile (ie median), and then store the corresponding quantile;

分位数与

分位数的差值；b) Calculate the IQR, which is defined as

quantile with

difference in quantiles;

c) Use IQR to scale the feature variables to achieve a unified scale;

According to the data visualization results, logarithmic transformation is performed on the data sets or feature variables that do not satisfy the inductive bias of the current algorithm, that is, the logarithm is taken after adding 1 to the value of the corresponding feature variable, so that the distribution of each feature variable is closer to normal distribution, to avoid the skewness of the data distribution from adversely affecting the prediction results of the model;

Step 3.2, outlier processing

The outliers of the data set are processed by a combination of boxplot and clustering+isolation forest. Among them, the boxplot is used for the comparison of the effect before and after the outlier processing and the confirmation of the processing benefit; the method adopted in this embodiment Based on K-Means++ clustering algorithm + outlier detection of isolated forest, K-Means++ algorithm produces more consistent results than traditional K-Means by placing initial centroids far away from each other.

Step 4. Concrete Data Feature Engineering

Since the original data set collected in step 1 only contains the numerical content of each component that makes up HPC, there is no specific content ratio relationship; perform mathematical calculations) to augment the data set. For example, calculate the ratio relationship between the original feature "water content" and "cement content", and then obtain the water-cement ratio; through the ratio relationship between "water content" and "(cement content + slag content + fly ash content) referred to as gel", Then the water-binder ratio is obtained; through feature engineering, the features obtained by the new structure are shown in Table 1 below:

Table 1

Step 5. Construction and evaluation of concrete compressive strength prediction model

Based on the data set after the above preprocessing and feature engineering, the training set and test set are divided into a ratio of 8:2. Based on the Boosting framework, the first prediction model based on the AdaBoost algorithm and the second prediction model based on the CatBoost algorithm are respectively established. At the same time, combined with 5-fold cross-validation, the performance of the first prediction model and the second prediction model were evaluated by using the regression model evaluation index. Among them, the evaluation indicators adopted in this embodiment are respectively defined as follows:

In the formula:

N is the total number of concrete sample data collected;

i is the sample number, that is, which sample;

为模型对第i条样本所预测的抗压强度值，即预测值；

is the compressive strength value predicted by the model for the i-th sample, that is, the predicted value;

为第i条样本所对应的实际抗压强度值，即观测值；

is the actual compressive strength value corresponding to the i-th sample, that is, the observed value;

Step 6. Parameter tuning of concrete compressive strength prediction model

Combined with the evaluation results of the above model performance, the Bayesian optimization method is used to optimize the hyperparameters of the first prediction model and the second prediction model, so as to further improve the performance of the model; among them, some of the key hyperparameters to be adjusted include the base model tree. The depth of the tree and the integrated model tree. The Bayesian optimization process takes the model error as the objective function, and finds the parameter with the smallest corresponding error through the combination of parameters.

Step 7. Fusion of the first prediction model and the second prediction model

Figure 2 shows the fitting effect diagram of the second prediction model based on the CatBoost algorithm on the test set, and Figure 3 shows the fitting effect diagram of the first prediction model based on the AdaBoost algorithm on the test set. It can be seen from the figure that the prediction effect of the second prediction model based on the CatBoost algorithm on the compressive strength is significantly better than that of the first prediction model based on the AdaBoost algorithm;

In order to improve the reliability of model prediction results, group decision-making is used to integrate the results of multiple model predictions, and fusion is performed at the model decision-making level to improve the accuracy of model prediction results, as shown in Figure 4. In this embodiment, The weighted average method fuses the CatBoost model and the AdaBoost model, and gives the CatBoost model a greater weight during the fusion process.

The weighted average model fusion process is as follows:

Taking the first prediction model based on the AdaBoost algorithm and the second prediction model based on the CatBoost algorithm as the base model h(x), record the fusion model of HPC compressive strength prediction as H(x), expressed as follows:

In the formula:

w _i : represents the weight of the i-th base model, in this embodiment, w _i ≥ 0 and satisfies

h _i (x): indicates the compressive strength of HPC predicted by the i-th base model;

T: indicates the number of base models for model fusion, the base models in this embodiment are AdaBoost and CatBoost, so T is equal to 2;

H(x): Indicates the final prediction result of the compressive strength of concrete.

Among them, the determination process of the fusion weight w _i of the base model is as follows:

Combined with RMSE, the loss function of the fusion model is defined as follows:

In the formula:

N is the sample size, that is, the total number of concrete sample data collected;

为第i条样本对应的混凝土实际抗压强度；

is the actual compressive strength of concrete corresponding to the i-th sample;

为HPC抗压强度预测的融合模型H(x)对第i条样本的混凝土抗压强度的预测值；

is the prediction value of the concrete compressive strength of the i-th sample by the fusion model H(x) predicted by HPC compressive strength;

Therefore, the final optimization objective of the fusion model is defined as follows:

minimize e(Loss)stw ₁ +w ₂ ＝1 and w ₁ ≥0,w ₂ ≥0 Equation (8)

In the formula:

Loss is the loss function of the integrated model H(x);

s.t is the abbreviation of constraints;

w ₁ and w ₂ are the fusion weights of the base models CatBoost and AdaBoost;

minimize means minimize;

By solving the constrained minimum optimization problem shown in Equation (8), the fusion weight w of the base model is obtained.

Step 8. Interpretability analysis of SHAP-based concrete compressive strength prediction model

Through the aforementioned model construction and evaluation, model parameter adjustment and model fusion, a concrete compressive strength prediction model with good predictive ability is obtained. Further, this embodiment interprets and analyzes the prediction results of the model in combination with the interpretability algorithm of the SHAP model, so as to better understand the influence of each eigenvalue on the compressive strength of concrete predicted by the model. The influence is represented by Shapley Value and abbreviated as Ψ, defined as follows:

In the formula:

S is the feature subset input by the model to be explained, that is, the concrete parameter set input to the fusion model of HPC compressive strength prediction;

x _j is the jth characteristic variable of the sample to be explained; that is, the jth concrete parameter;

p is the total number of features, that is, the total number of concrete parameters;

val _x (S) indicates the prediction result of the model to be explained for sample x when S is the input feature, that is, the compressive strength result output by the model to be explained, where x is the sample, and the elements of x are denoted as x _i , x _i is the value corresponding to the i-th feature variable;

The SHAP value is the importance of the jth feature to the output of the model to be explained, that is, the marginal contribution, and the Shapley Value is the mean value of each marginal contribution. The model interpretation result of the model to be explained is defined as follows:

In the formula:

G is the model to be explained, in the present embodiment, the fusion model H(x) of the HPC compressive strength prediction after the fusion of CatBoost and AdaBoost;

z'∈{0,1} ^M is a combination vector, representing whether the feature z _j (j∈[1,M]) exists, where z _j is the first input to the fusion model of HPC compressive strength prediction in this embodiment j concrete parameters, z' is used to identify whether z ₁ ~ z _M exists in the parameter set input to the fusion model for HPC compressive strength prediction;

M is the number of combined features, that is, the number of input parameters input to the model g to be explained;

为特征j的特征归因Shapley Value，即第j个参数对待解释的模型g的预测结果，本实施例中即对抗压强度的贡献值；

Be the characteristic attribution Shapley Value of feature j, that is, the prediction result of the model g to be explained by the jth parameter, which is the contribution value to the compressive strength in this embodiment;

Ψ ₀ is the average prediction result of the model g to be explained, which is the average value of the prediction results of compressive strength output by the fusion model of HPC compressive strength prediction in this embodiment.

Shapley Value measures the contribution of the feature to the overall prediction result. When Ψ _j > 0, it indicates that the feature has a positive effect on the prediction value, that is, it has the effect of enhancing the compressive strength.

其中，本实施例利用SHAP算法进行模型可解释性分析的部分图表示例见图5～7；The SHAP global feature importance is the average of the sum of the absolute value of the Shapley Value of each feature, that is

Among them, some chart examples of using the SHAP algorithm for model interpretability analysis in this embodiment are shown in Figures 5-7;

Figure 5 shows the local SHAP interpretation of the first sample, and Figure 6 shows the local SHAP interpretation of the tenth sample; in the figure, the average prediction result of the model is 35.25Mpa, and the compressive strength prediction result of the model for the first sample is 76.17Mpa . The compressive strength prediction result of the model for the 10th sample is 38.73Mpa. Based on the model prediction results and actual observation results, as well as the value of each parameter, it can provide guidance for the design of concrete mix ratio.

Figure 7 is a summary of SHAP’s global interpretation. The Y-axis arranges the characteristic factors that affect the compressive strength from top to bottom according to the degree of contribution; among them, the curing period, water-cement ratio, and cement content are the first three factors that affect the compressive strength of HPC. Strength has a significant impact on factors, followed by water, water-binder ratio and so on. The X-axis is the average influence value of each factor on the prediction results of the compressive strength prediction model. In the current experimental results, as the curing period increases, the compressive strength will increase by an average of 8Mpa.

It should be understood that the sequence numbers of the steps in the above embodiments do not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiment of the present application.

The embodiment of the present application also provides an embodiment of the system, including a model training module and a model composite operation module;

The model training module is used to train the first model and the second model respectively according to the historical parameter data of the collected HPC, and respectively obtain the first prediction model and the second prediction model for predicting the compressive strength of the HPC after training;

The model composite operation module is used to perform composite operations on the HPC compressive strength output by the first prediction model for predicting the HPC compressive strength and the second prediction model that have been trained to obtain a fusion model for predicting the HPC compressive strength, HPC compressive strength The fusion model of strength prediction outputs the final prediction result of HPC compressive strength.

In the model composite operation module, the weighted average method is used to perform composite operations on the HPC compressive strength output by the first prediction model and the second prediction model.

In another embodiment, a model interpretation and analysis module is also included, which is used to interpret and analyze the fusion model of the HPC compressive strength prediction by using the first algorithm, so as to obtain the influence of the component content of each HPC parameter on the HPC compressive strength .

The content not described in detail in this specification belongs to the prior art known to those skilled in the art.

Claims (10)

1. a method for predicting HPC compressive strength based on model fusion, is characterized in that, comprises the steps: S1. The first prediction model and the second prediction model are respectively trained according to the historical parameter data of the collected HPC, and the first prediction model and the second prediction model for predicting the HPC compressive strength after training are respectively obtained; S2. Composite calculation is performed on the trained first prediction model and the second prediction model to obtain an HPC compressive strength prediction fusion model, and the HPC compressive strength prediction fusion model outputs a final HPC compressive strength prediction result. 2. the HPC compressive strength prediction method based on model fusion as claimed in claim 1, is characterized in that, also comprises step S3 after described step S2: S3. Using the first algorithm to interpret and analyze the fusion model for predicting HPC compressive strength, and obtain the contribution value of the component content of each HPC parameter to the HPC compressive strength output by the fusion model for predicting HPC compressive strength. 3. the HPC compressive strength prediction method based on model fusion as claimed in claim 2, is characterized in that, described first algorithm is based on SHAP interpretability algorithm. 4. the HPC compressive strength prediction method based on model fusion as claimed in claim 1, is characterized in that, in described step S2, the method for obtaining the fusion model of HPC compressive strength prediction comprises adopting weighted average method to first prediction The model is compounded with the second predictive model. 5. the HPC compressive strength prediction method based on model fusion as claimed in claim 4, is characterized in that, described first predictive model predicts the compressive strength of HPC based on AdaBoost algorithm; Described second predictive model is based on CatBoost The algorithm predicts the compressive strength of the HPC. 6. the HPC compressive strength prediction method based on model fusion as claimed in claim 5, is characterized in that, when adopting weighted average method to carry out model compound operation to first prediction model and second prediction model, based on the second of CatBoost algorithm The weight of the prediction model is greater than the weight of the first prediction model based on the AdaBoost algorithm. 7. the HPC compressive strength prediction method based on model fusion as claimed in claim 6, is characterized in that, the calculation method of the weight of each prediction model in the described weighted average method comprises: by solving the band constraints shown in the following formula The minimum optimization problem of , the weights of the first prediction model and the second prediction model are obtained: minimize e(Loss)stw ₁ +w ₂ ＝1and w ₁ ≥0,w ₂ ≥0 In the formula: w ₁ and w ₂ correspond to the weights of the first prediction model and the second prediction model respectively; Loss is the loss function of the following H(x);

In the formula: H(x): indicates the final predicted HPC compressive strength output by the fusion model of HPC compressive strength prediction; w ₁ , w ₂ : represent the weights of the first prediction model and the second prediction model respectively; h ₁ (x), h ₂ (x): represent the HPC compressive strength predicted by the first prediction model and the second prediction model, respectively. 8. a HPC compressive strength prediction system based on model fusion of method as claimed in claim 1, is characterized in that, comprises: model training module and model composite operation module; The model training module is used to train the first model and the second model respectively according to the historical parameter data of the collected HPC, and respectively obtain the first prediction model and the second prediction model for predicting the compressive strength of the HPC that have been trained; The model composite operation module is used to perform composite operations on the trained first prediction model and the second prediction model to obtain a fusion model of HPC compressive strength prediction, and the fusion model of HPC compressive strength prediction outputs the final HPC compressive strength strength predictions. 9. the HPC compressive strength prediction system based on model fusion as claimed in claim 8, is characterized in that, also comprises model interpretation analysis module, is used for utilizing the first algorithm to explain the fusion model of described HPC compressive strength prediction The contribution value of the component content of each HPC parameter to the HPC compressive strength output by the fusion model of HPC compressive strength prediction is obtained. 10. The HPC compressive strength prediction system based on model fusion as claimed in claim 8, it is characterized in that, in the described model composite operation module, adopt weighted average method to the HPC compressive strength of the first prediction model and the second prediction model output The compressive strength is compounded.

Images