A strength detection method and system for building concrete
The integration of ultrasonic and rebound testing with nonlinear models and cross-validation enhances concrete strength detection accuracy and efficiency, addressing the limitations of existing methods.
Patent Information
- Application Number
- CN202411449704.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-17
- Publication Date
- 2024-12-13
- Estimated Expiration
- 2044-10-17
AI Technical Summary
Existing concrete strength detection methods are labor-intensive, prone to damage, and lack accuracy and intelligence, especially in complex structures, and struggle with internal defects and chemical changes.
A method combining ultrasonic testing and rebound testing with nonlinear models, using support vector, polynomial, and radial basis function vector machine models to predict concrete strength, incorporating an attention mechanism for feature weighting and cross-validation to enhance accuracy.
Improves the accuracy and efficiency of concrete strength detection by reducing human intervention and errors, capturing nonlinear relationships, and optimizing model performance through cross-validation.
Smart Images
Figure CN118961888B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of building technology, and in particular to a strength detection method and system for building concrete. Background Art
[0002] The stability and bearing capacity of a building depend to a large extent on the strength of concrete. As one of the important contents of laws and regulations on the quality supervision of construction projects, concrete strength testing is an important means of evaluating the quality of concrete materials. Different raw material ratios, construction processes and maintenance conditions will affect the final strength of concrete. Through strength testing, the quality of concrete can be intuitively understood, providing an objective basis for quality evaluation for material suppliers and construction units, and promoting the continuous improvement of material quality. The existing technology mainly uses a thin-walled hollow drill with a water cooling device to drill core samples on concrete components, or drills holes, grinds grooves, embeds anchors and installs a pull-out instrument on the hardened concrete surface for pull-out tests to detect the strength of concrete, and obtains the strength of concrete through rebound method and ultrasonic method.
[0003] However, core sampling and pull-out testing methods can easily damage concrete, and in some complex structures or large projects, the sampling and testing process may be difficult, time-consuming and labor-intensive. The rebound method and ultrasonic method are relatively simple detection methods. When there is a large difference in the quality of the inside and outside of the concrete, such as internal defects, chemical corrosion, fire or frostbite, the accuracy and efficiency of the strength test will be low. In addition, these methods are relatively low in intelligence and require manual operation and more human judgment, making it difficult to achieve fully automated and intelligent detection. Summary of the invention
[0004] The present invention provides a strength detection method and system for building concrete, so as to solve the problem that it is difficult to perform intelligent detection on building concrete and improve the accuracy of strength detection.
[0005] Conduct ultrasonic testing and rebound testing on concrete to obtain an estimated value of concrete strength;
[0006] Acquire the sound velocity data of the concrete through ultrasonic testing, and use a nonlinear model to predict the strength according to the sound velocity data to obtain an initial strength prediction value; wherein the nonlinear model is obtained by taking the model with the highest comprehensive score in the model set, and the model set is obtained by establishing a support vector machine model, a polynomial vector machine model, and a radial basis function vector machine model according to preset concrete strength data; the comprehensive score is calculated by considering various properties of concrete as different feature modes and allowing the model to give higher weights to preset features in the decision-making process through an attention mechanism;
[0007] A cross validation is performed based on the estimated concrete strength value and the initial strength prediction value to obtain a concrete strength prediction value.
[0008] The ultrasonic detection of the present invention can provide the sound velocity data inside the concrete, reflecting the density and integrity of the concrete, while the rebound detection infers its strength by measuring the hardness of the concrete surface. This fusion of multi-source data helps to improve the accuracy of concrete strength prediction. The nonlinear relationship between concrete strength and sound velocity can be better captured by using a nonlinear model to predict strength based on the sound velocity data of ultrasonic detection. In addition, by establishing a model set and selecting the model with the highest comprehensive score as the final nonlinear model, the optimal selection of the model is achieved; and the calculation of the comprehensive score takes into account the different attribute characteristics of concrete, and gives a higher weight to the preset features through the attention mechanism, which helps to screen out the model that best suits the current data set. Since cross-validation helps to discover possible overfitting or underfitting problems of the model, and can optimize the prediction results by adjusting the model parameters or structure, the reliability of the concrete strength prediction value can be further improved by using the concrete strength estimated value and the initial strength prediction value for cross-validation.
[0009] Compared with the prior art, the present invention conducts preliminary testing on concrete through the ultrasonic rebound method to obtain an estimated value of concrete strength, and then predicts the strength of concrete through a nonlinear model to obtain an initial strength prediction value, thereby cross-validating the results obtained based on these two methods to obtain a more accurate concrete strength prediction value. The automated detection and prediction process reduces human intervention and errors, improves the accuracy and efficiency of the prediction, and can therefore solve the problem of difficulty in conducting intelligent detection of building concrete and improve the accuracy of strength detection.
[0010] As a preferred solution, the model set is obtained by establishing a support vector machine model, a polynomial vector machine model, and a radial basis function vector machine model according to preset concrete strength data, specifically:
[0011] Test the preset concrete samples to obtain basic data including sound velocity data and actual strength value;
[0012] Performing data cleaning and standardization processing on the basic data to obtain a data set;
[0013] Finding the best hyperplane in the original space of the data set by a linear kernel function to establish the support vector machine model;
[0014] Capturing the nonlinear relationship between the features in the data set by a polynomial kernel function, and establishing the polynomial vector machine model;
[0015] Based on the data set, the radial basis function vector machine model is established by searching for the best hyperplane in the high-dimensional feature space according to the similarity metric through the RBF kernel function;
[0016] The model set is composed of the support vector machine model, the polynomial vector machine model and the radial basis function vector machine model.
[0017] In this preferred solution, the linear kernel function is suitable for the case where the data itself is close to linearly separable; the polynomial kernel function can capture the nonlinear relationship between features; and the RBF kernel function further enhances the generalization ability of the model by finding the best hyperplane in the high-dimensional feature space. Integrating these models can combine their respective advantages, thereby improving the accuracy and stability of subsequent intensity predictions. Since multiple kernel functions are used, even if one of the kernel functions performs poorly on a specific data set, other kernel functions may still provide useful information. This diversity helps to reduce the overfitting or underfitting problems that may exist in a single model and enhance the robustness and adaptability of the entire model set.
[0018] As a preferred solution, the optimal hyperplane is found in the original space of the data set by a linear kernel function to establish the support vector machine model, specifically:
[0019] A regularization term is introduced to control the complexity of the model, and an actual plane with the shortest distance to the nearest point of the two types of data points is searched in the original space of the data set to obtain a first optimal hyperplane; wherein the two types of data points are data points in the data set;
[0020] The support vector machine model is established according to the first optimal hyperplane.
[0021] This preferred solution can effectively control the complexity of the model and avoid overfitting the model on the training data by introducing regularization terms. This enables the model to show better generalization performance when facing unseen test data, that is, it can more accurately classify or predict new data. By maximizing the classification interval, an optimal decision boundary can be found that can not only accurately classify the training data, but also maintain good classification performance for new data.
[0022] As a preferred solution, the nonlinear relationship between the features in the data set is captured by a polynomial kernel function, and the polynomial vector machine model is established, specifically:
[0023] Mapping the data set from the original space to a high-dimensional polynomial feature space;
[0024] Capturing the polynomial combination mode in the data set by the polynomial kernel function to obtain nonlinear interaction data;
[0025] According to the nonlinear interaction data, searching in the high-dimensional polynomial feature space for an actual plane that is shortest to the nearest points of the two types of data points, to obtain a second optimal hyperplane; wherein the two types of data points are data points in the data set;
[0026] The polynomial vector machine model is established according to the second optimal hyperplane.
[0027] In this preferred embodiment, since the polynomial kernel function can express the complex interactions and combinations between features, the polynomial kernel function can better capture these interactions in concrete strength detection, thereby improving the prediction accuracy and robustness of the model.
[0028] As a preferred solution, based on the data set, the RBF kernel function is used to find the best hyperplane in the high-dimensional feature space according to the similarity metric, and the radial basis function vector machine model is established, specifically:
[0029] Calculating the Euclidean distances between different data points in the data set to obtain a Euclidean distance set;
[0030] According to the Euclidean distance set, similarity measures of different data points in the data set in the high-dimensional feature space are calculated by the RBF kernel function to obtain a similarity measure set;
[0031] According to the similarity measure set, searching in the high-dimensional polynomial feature space for an actual plane that is shortest to the nearest points of the two types of data points, to obtain a third optimal hyperplane; wherein the two types of data points are data points in the data set;
[0032] The radial basis function vector machine model is established according to the third optimal hyperplane.
[0033] The RBF kernel function of the preferred solution calculates the Euclidean distance between data points and defines the similarity based on these distances in high-dimensional space, which enables the model to capture the local similarity between data points well; in concrete strength testing, this means that data points of similar materials or proportions will be more closely linked together, which helps to improve the prediction accuracy of the model. By searching for the third optimal hyperplane that can maximize the classification interval in the high-dimensional feature space, the radial basis function vector machine model can maintain a good fit to the training data while maintaining good prediction capabilities for new data, which helps to reduce overfitting and improve the generalization ability of the model.
[0034] As a preferred solution, the comprehensive score is calculated by treating various properties of concrete as different feature modes and allowing the model to give higher weights to preset features in the decision-making process through the attention mechanism. Specifically, it is:
[0035] The preset properties of concrete are regarded as different eigenmodes, and several eigenmodes are obtained;
[0036] Based on the concrete strength data and similarity formula, the plurality of feature modes are integrated by multi-feature fusion technology, and the key factors affecting the concrete similarity are captured by applying the attention mechanism so that the model gives a higher weight to the preset features in the decision-making process, and the similarity score between the strength prediction result of the vector machine model and the concrete strength data is calculated;
[0037] Calculating a mean square error and an F1 score based on the strength prediction result of the vector machine model and the concrete strength data;
[0038] The comprehensive score is calculated based on the similarity score, the mean square error and the F1 score.
[0039] This preferred solution regards the various preset properties of concrete as different feature modes and integrates them through multi-feature fusion technology, which can make full use of the complementarity between different features and improve the feature expression ability of the similarity score. The introduction of the attention mechanism allows the model to automatically give higher weights to preset features in the decision-making process. This mechanism enables the model to more accurately capture the key factors affecting the similarity of concrete, thereby improving the accuracy and reliability of the prediction. By calculating the similarity score, mean square error and F1 score, and assigning weights to these indicators, the final comprehensive score is obtained to evaluate the model performance, which can more comprehensively reflect the performance of the model in all aspects and provide a more comprehensive reference for the final selection of the model.
[0040] As a preferred solution, the similarity formula is specifically:
[0041] ;
[0042] in, is the concrete batch, i is the concrete resource, t is the time, m is the modal type of concrete, is the weight of the mth mode, is the multimodal attention mechanism function, It is a concrete batch The high-level indicator feature vector of the mth mode at time t, is the high-level indicator eigenvector of the mth mode of concrete resource i at time t, λ is the time decay factor, is the mth modal fusion coefficient.
[0043] As a preferred solution, the comprehensive score is calculated based on the similarity score, the mean square error and the F1 score, specifically:
[0044] Based on a comprehensive score formula, weights are assigned to the similarity score, the mean square error, and the F1 score to calculate the comprehensive score;
[0045] The comprehensive score formula is:
[0046] ;
[0047] in, It is The similarity score between the predicted strength result and the actual strength result of concrete samples, It is The mean square error between the predicted strength results and the actual strength results of concrete samples, It is The F1 score of the predicted strength results and the actual strength results of the concrete samples, , and are the feature weights of similarity, mean square error and F1 score respectively, is the total number of concrete samples.
[0048] As a preferred solution, a cross-validation is performed based on the estimated concrete strength value and the initial strength prediction value to obtain a concrete strength prediction value, specifically:
[0049] According to the weighted average of the estimated concrete strength value and the initial strength predicted value, the predicted concrete strength value is predicted using a prediction model;
[0050] Among them, the prediction model is obtained by adjusting the adaptive model according to a set of performance indicators between the predicted value and the concrete strength estimated value, the predicted value is obtained by training and running the adaptive model using a training set and a test set, and the training set and the test set are obtained by dividing the concrete strength estimated value and the initial strength predicted value in a preset manner.
[0051] This preferred solution can combine the advantages of the estimated concrete strength value and the initial strength predicted value by weighted averaging, and reduce the error that may be caused by a single prediction method. Since the prediction model is obtained by adaptively adjusting the performance index set between the predicted value and the estimated concrete strength value, this means that the model can be dynamically optimized according to the actual prediction effect to better adapt to data changes and the needs of the prediction task. This method enables the model to maintain a high prediction accuracy when running for a long time or facing new data.
[0052] The present invention also provides a strength detection system for building concrete, comprising an estimation module, a prediction module and a verification module;
[0053] The estimation module is used to perform ultrasonic testing and rebound testing on concrete to obtain an estimated value of concrete strength;
[0054] The prediction module is used to obtain the sound velocity data of the concrete through ultrasonic testing, and use a nonlinear model to predict the strength according to the sound velocity data to obtain an initial strength prediction value; wherein the nonlinear model is obtained by taking the model with the highest comprehensive score in the model set, and the model set is obtained by establishing a support vector machine model, a polynomial vector machine model, and a radial basis function vector machine model according to preset concrete strength data; the comprehensive score is calculated by considering various properties of concrete as different feature modes and allowing the model to give higher weights to preset features in the decision-making process through an attention mechanism;
[0055] The verification module is used to perform cross-validation based on the estimated concrete strength value and the initial strength prediction value to obtain a concrete strength prediction value. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1 It is a flow chart of a strength detection method for building concrete provided in an embodiment of the present application;
[0057] Figure 2 It is a structural schematic diagram of a strength detection system for building concrete provided in an embodiment of the present application. DETAILED DESCRIPTION
[0058] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0059] In the description of the present application, it should be understood that the terms "first", "second" and "third" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first", "second" and "third" may explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise specified, "several" means two or more.
[0060] A strength detection method for building concrete provided in an embodiment of the present application is mainly used in situations where intelligent detection of concrete strength is required to improve the accuracy and efficiency of strength detection, thereby ensuring the technical effect of concrete application strength.
[0061] Embodiment 1:
[0062] See also Figure 1 The embodiment of the present application provides a strength detection method for building concrete, including S1 to S3, and the specific implementation steps are as follows:
[0063] S1. Perform ultrasonic testing and rebound testing on concrete to obtain an estimated value of concrete strength.
[0064] Step S1 of the embodiment of the present application is specifically as follows:
[0065] The measurement area is drawn on the concrete by random sampling, and the measurement area number is marked, and the measurement points are evenly distributed within the specified range; the distance between the measurement points is not less than 20mm, and the distance between the measurement points and the edge of the component is not less than 30mm;
[0066] Use a rebound hammer to test the rebound value of each measuring area in a manner perpendicular to the surface to obtain a rebound value set;
[0067] Eliminate abnormal values that exceed the preset range in the rebound value set, and calculate the average value of the remaining rebound values to obtain the actual rebound value .
[0068] At the measuring point corresponding to the rebound test, the transmitting and receiving transducers of the ultrasonic detector are controlled to be on the same axis to perform a sound velocity test and obtain a sound velocity value set;
[0069] Take the average value of the sound speed value set as the representative value of the sound speed in the measurement area to get the actual sound speed value ;
[0070] According to the actual rebound value and the actual sound speed Calculate the estimated value of concrete strength ;
[0071] Among them, the estimated value of concrete strength is:
[0072] ;
[0073] in, , and is a coefficient determined by calibration tests, is the actual rebound value, is the actual speed of sound.
[0074] S2. Acquire the sound velocity data of concrete through ultrasonic testing, and use a nonlinear model to predict the strength based on the sound velocity data to obtain an initial strength prediction value; wherein, the nonlinear model is obtained by taking the model with the highest comprehensive score in the model set, and the model set is obtained by establishing a support vector machine model, a polynomial vector machine model, and a radial basis function vector machine model based on the preset concrete strength data; the comprehensive score is calculated by treating various properties of concrete as different feature modes, and using the attention mechanism to allow the model to give higher weights to preset features in the decision-making process.
[0075] Step S2 of the embodiment of the present application includes S2.1 to S2.6; specifically, it is:
[0076] S2.1. Evenly arrange measurement areas in a square grid shape on the side of the preset concrete sample in the pouring direction, and arrange several pairs of measurement points in each measurement area;
[0077] Use an ultrasonic detector to measure the straight-line distance between each pair of measuring points and record the propagation time of the ultrasonic wave. Calculate the sound velocity data based on the straight-line distance and propagation time.
[0078] The sound velocity data and the actual strength value corresponding to the preset concrete sample are defined as basic data;
[0079] The basic data is cleaned and standardized to obtain the data set.
[0080] S2.2, introduce regularization terms to control the complexity of the model and define the optimal hyperplane; the hyperplane refers to the plane that maximizes the distance (i.e., interval) from all points in the two types of data; the two types of data refer to the two types of data in the data set divided according to the intensity;
[0081] Assume that the hyperplane can be represented as ,in is the normal vector of the hyperplane, is the bias term, is a data point, then the distance formula from the point to the hyperplane is: ; Among them, the operator symbol represents the magnitude of a vector;
[0082] All points in the data set are on the correct side of the preset and the corresponding distance Under the condition that is at least positive, the interval is maximized by the preset optimization objective function, and the Lagrange multiplier method is used to solve it to obtain the optimal normal vector of the first optimal hyperplane and the bias term ;
[0083] According to the normal vector and the bias term Establish a support vector machine model; the decision function of the support vector machine model is ,and The function is used to judge the positive and negative, and the category to which the data point belongs is determined based on the output of the function.
[0084] By introducing a regularization term, this embodiment S2.2 can effectively control the complexity of the model and avoid overfitting the model on the training data. This enables the model to show better generalization performance when facing unseen test data, that is, it can more accurately classify or predict new data. By maximizing the classification interval, an optimal decision boundary can be found that can not only accurately classify the training data, but also maintain good classification performance for new data.
[0085] S2.3, mapping the data set from the original space to a high-dimensional polynomial feature space;
[0086] The polynomial kernel function is used to capture the polynomial combination mode in the data set to obtain nonlinear interaction data;
[0087] According to the nonlinear interaction data, the actual plane with the shortest distance to the nearest point of the two types of data points is searched in the high-dimensional polynomial feature space to obtain the second optimal hyperplane; wherein the two types of data points are the two types of data points in the data set divided according to the intensity;
[0088] Establishing a polynomial vector machine model according to the second optimal hyperplane;
[0089] Among them, the polynomial kernel function is:
[0090] ;
[0091] in, and are two points in the data set, is the kernel coefficient, is a constant term, is the degree of the polynomial, yes and The dot product between .
[0092] In this embodiment S2.3, since the polynomial kernel function can express the complex interactions and combinations between features, in concrete strength detection, the polynomial kernel function can better capture these interactions, thereby improving the prediction accuracy and robustness of the model.
[0093] S2.4, calculating the Euclidean distance between different data points in the data set to obtain a Euclidean distance set;
[0094] According to the Euclidean distance set, the similarity measure of different data points in the data set in the high-dimensional feature space is calculated by the RBF kernel function to obtain the similarity measure set;
[0095] According to the similarity measure set, searching in the high-dimensional polynomial feature space for the actual plane that is shortest to the nearest point of the two types of data points, to obtain the third optimal hyperplane; wherein the two types of data points are data points in the data set;
[0096] Establishing a radial basis function vector machine model according to the third optimal hyperplane;
[0097] The model set is composed of a support vector machine model, a polynomial vector machine model and a radial basis function vector machine model;
[0098] Among them, the RBF kernel function is:
[0099] ;
[0100] in, It is a data point and data points The square of the Euclidean distance between is the bandwidth parameter of the kernel.
[0101] The RBF kernel function of this embodiment S2.4 calculates the Euclidean distance between data points and defines the similarity based on these distances in high-dimensional space, which enables the model to capture the local similarity between data points well; in concrete strength testing, this means that data points of similar materials or proportions will be more closely linked together, which helps to improve the prediction accuracy of the model. By searching for the third optimal hyperplane that can maximize the classification interval in the high-dimensional feature space, the radial basis function vector machine model can maintain a good fit to the training data while maintaining good prediction capabilities for new data, which helps to reduce overfitting and improve the generalization ability of the model.
[0102] In the present embodiment S2.2~S2.4, the linear kernel function is suitable for the case where the data itself is close to linearly separable; the polynomial kernel function can capture the nonlinear relationship between features; and the RBF kernel function further enhances the generalization ability of the model by finding the best hyperplane in the high-dimensional feature space. Integrating these models can combine their respective advantages, thereby improving the accuracy and stability of subsequent intensity prediction. Since multiple kernel functions are used, even if one of the kernel functions performs poorly on a specific data set, other kernel functions may still provide useful information. This diversity helps to reduce the overfitting or underfitting problems that may exist in a single model and enhance the robustness and adaptability of the entire model set.
[0103] S2.5, considering the preset properties of concrete as different characteristic modes, and obtaining several characteristic modes;
[0104] Based on concrete strength data and similarity formula, several feature modes are integrated through multi-feature fusion technology, and the key factors affecting concrete similarity are captured by applying the attention mechanism, so that the model gives higher weights to more important features in the decision-making process. The similarity score between the strength prediction result of the vector machine model and the concrete strength data is calculated.
[0105] Based on the strength prediction results of the vector machine model and the concrete strength data, the mean square error and F1 score were calculated;
[0106] Based on the comprehensive score formula, weights are assigned to the similarity score, mean square error, and F1 score to calculate the comprehensive score;
[0107] The similarity formula is:
[0108] ;
[0109] The comprehensive score formula is:
[0110] ;
[0111] The F1 score is:
[0112] ;
[0113] For the parameters Precision and Recall, we have:
[0114] ;
[0115] ;
[0116] in, is the concrete batch, i is the concrete resource, t is the time; m is the modal type, m=1 represents cubic compressive strength, m=2 represents axial compressive strength, and m=3 represents tensile strength; is the weight of the mth mode, is the multimodal attention mechanism function, It is a concrete batch The high-level indicator feature vector of the mth mode at time t, is the high-level indicator eigenvector of the mth mode of concrete resource i at time t, and λ is the time decay factor; is the mth modal fusion coefficient, which is used to adjust the weights of different modalities;
[0117] It is The similarity score between the predicted strength result and the actual strength result of concrete samples, It is The mean square error between the predicted strength results and the actual strength results of concrete samples, It is The F1 score of the predicted strength results and the actual strength results of the concrete samples, , and are the feature weights of similarity, mean square error and F1 score respectively, is the total number of concrete samples;
[0118] Indicates the number of samples correctly predicted by the model as positive. It indicates the number of samples that the model predicts to be positive but are actually negative. Indicates the number of samples that the model predicts to be negative but are actually positive.
[0119] In this embodiment S2.5, the multiple preset attributes of concrete are regarded as different feature modes, and they are integrated through multi-feature fusion technology, which can make full use of the complementarity between different features and improve the feature expression ability of the similarity score. The introduction of the attention mechanism allows the model to automatically give higher weights to the preset features in the decision-making process. This mechanism enables the model to more accurately capture the key factors affecting the similarity of concrete, thereby improving the accuracy and reliability of the prediction. By calculating the similarity score, mean square error and F1 score, and assigning weights to these indicators, the comprehensive score is finally obtained to evaluate the model performance, which can more comprehensively reflect the performance of the model in all aspects and provide a more comprehensive reference for the final selection of the model.
[0120] S2.6. Obtain the sound velocity data of concrete through ultrasonic testing;
[0121] The model with the highest comprehensive score in the model set is taken as the nonlinear model, and the intensity is predicted according to the sound velocity data to obtain the initial intensity prediction value.
[0122] S3. Perform cross validation based on the estimated concrete strength value and the initial strength prediction value to obtain the concrete strength prediction value.
[0123] Step S3 of the embodiment of the present application includes S3.1 to S3.2; specifically, it is:
[0124] S3.1. Standardize and normalize the estimated concrete strength value, the initial strength prediction value and any physical property data or chemical property data that may affect the concrete strength to obtain a comprehensive data set;
[0125] Divide the comprehensive data set into training set and test set according to a preset method;
[0126] The initial adaptive model is trained using the training set data, and the performance of the model is evaluated using the test set to obtain the adaptive model.
[0127] S3.2. One or more adapter modules are set up to capture the specific features and information of concrete, and a self-attention mechanism is introduced to enhance the adapter module's ability to capture key features, and a dynamic feature selection mechanism is added to control the adapter module to automatically determine which features are more important for the current prediction task, and adjust the weight accordingly to selectively transmit information; wherein the adapter module is used to capture and process information related to the performance indicator set;
[0128] Initialize the parameters of the adapter module and insert the adapter module before the preset last few layers of the adaptive model;
[0129] Update the forward propagation logic of the adaptive model to include the processing steps of the adapter module;
[0130] Locate the original parameters that need to be frozen in the adaptive model by setting the "requires_grad" attribute of the original parameters to "False";
[0131] In the adaptive model, only the parameters of the adapter module are adjusted, while the preset original parameters of the adaptive model are kept unchanged, so that the adaptive model is adapted on the performance indicator set to obtain a prediction model.
[0132] The predicted value of concrete strength is obtained by using the prediction model based on the weighted average of the estimated value of concrete strength and the predicted value of initial strength.
[0133] In this embodiment S3.2, by weighted averaging the estimated concrete strength value and the initial strength prediction value, the advantages of both can be combined to reduce the error that may be caused by a single prediction method. Since the prediction model is obtained by adaptively adjusting the performance indicator set between the predicted value and the estimated concrete strength value, this means that the model can be dynamically optimized according to the actual prediction effect to better adapt to the changes in data and the needs of the prediction task. This method enables the model to maintain a high prediction accuracy when running for a long time or facing new data;
[0134] Moreover, for the prediction model, the self-attention mechanism allows the model to consider the information of all other features when processing each feature, so as to more comprehensively understand the internal structure and relevance of the data; the addition of a dynamic feature selection mechanism enables the adapter module to automatically determine which features are more important for the current prediction task and adjust the weights accordingly, which helps to improve the robustness and adaptability of the model. Only adjusting the parameters of the adapter module allows the model to focus on optimizing the most relevant parts of the current prediction task. This targeted optimization strategy helps to improve the performance and accuracy of the model.
[0135] Overall, this embodiment has the following beneficial effects:
[0136] The ultrasonic detection of the present invention can provide the sound velocity data inside the concrete, reflecting the density and integrity of the concrete, while the rebound detection infers its strength by measuring the hardness of the concrete surface. This fusion of multi-source data helps to improve the accuracy of concrete strength prediction. The nonlinear relationship between concrete strength and sound velocity can be better captured by using a nonlinear model to predict strength based on the sound velocity data of ultrasonic detection. In addition, by establishing a model set and selecting the model with the highest comprehensive score as the final nonlinear model, the optimal selection of the model is achieved; and the calculation of the comprehensive score takes into account the different attribute characteristics of concrete, and gives a higher weight to the preset features through the attention mechanism, which helps to screen out the model that best suits the current data set. Since cross-validation helps to discover possible overfitting or underfitting problems of the model, and can optimize the prediction results by adjusting the model parameters or structure, the reliability of the concrete strength prediction value can be further improved by using the concrete strength estimated value and the initial strength prediction value for cross-validation.
[0137] Embodiment 2:
[0138] See also Figure 2 , an embodiment of the present application provides a strength detection system for building concrete, including an estimation module 10, a prediction module 20 and a verification module 30;
[0139] The estimation module 10 is used to perform ultrasonic testing and rebound testing on concrete to obtain an estimated value of concrete strength;
[0140] The prediction module 20 is used to obtain the sound velocity data of concrete through ultrasonic testing, and use a nonlinear model to predict the strength according to the sound velocity data to obtain an initial strength prediction value; wherein the nonlinear model is obtained by taking the model with the highest comprehensive score in the model set, and the model set is obtained by establishing a support vector machine model, a polynomial vector machine model, and a radial basis function vector machine model according to the preset concrete strength data; the comprehensive score is calculated by treating various properties of concrete as different feature modes and allowing the model to give higher weights to the preset features in the decision-making process through an attention mechanism;
[0141] The verification module 30 is used to perform cross-validation based on the estimated concrete strength value and the initial strength prediction value to obtain the concrete strength prediction value.
[0142] In one embodiment, the estimation module 10 is specifically:
[0143] The measurement area is drawn on the concrete by random sampling, and the measurement area number is marked, and the measurement points are evenly distributed within the specified range; the distance between the measurement points is not less than 20mm, and the distance between the measurement points and the edge of the component is not less than 30mm;
[0144] Use a rebound hammer to test the rebound value of each measuring area in a manner perpendicular to the surface to obtain a rebound value set;
[0145] Eliminate abnormal values in the rebound value set that exceed the preset range, and calculate the average value of the remaining rebound values to obtain the actual rebound value .
[0146] At the measuring point corresponding to the rebound test, the transmitting and receiving transducers of the ultrasonic detector are controlled to be on the same axis to perform a sound velocity test and obtain a sound velocity value set;
[0147] Take the average value of the sound speed value set as the representative value of the sound speed in the measurement area to get the actual sound speed value ;
[0148] According to the actual rebound value and the actual sound speed Calculate the estimated value of concrete strength ;
[0149] Among them, the estimated value of concrete strength is:
[0150] ;
[0151] in, , and is a coefficient determined by calibration tests, is the actual rebound value, is the actual speed of sound.
[0152] In one embodiment, the prediction module 20 includes a data set unit, a first model unit, a second model unit, a third model unit, a score unit, and an initial unit; specifically:
[0153] The data set unit is used to evenly arrange measurement areas in a square grid shape on the side of the pouring direction of the preset concrete sample, and to arrange a plurality of pairs of measurement points in each measurement area;
[0154] The data set unit is also used to measure the straight-line distance between each pair of measuring points using an ultrasonic detector, and to record the propagation time of the ultrasonic wave, and to calculate the sound velocity data according to the straight-line distance and the propagation time;
[0155] The data set unit is also used to define the sound velocity data and the actual strength value corresponding to the preset concrete sample as basic data;
[0156] The data set unit is also used to clean and standardize the basic data to obtain the data set.
[0157] The first unit of the model is used to introduce regularization terms to control the complexity of the model and define the optimal hyperplane; the hyperplane refers to the plane that maximizes the distance (i.e., interval) from all points in the two types of data; the two types of data refer to the two types of data in the data set divided according to the intensity;
[0158] The first unit of the model is also used to assume that the hyperplane can be expressed as ,in is the normal vector of the hyperplane, is the bias term, is a data point, then the distance formula from the point to the hyperplane is: ; Among them, the operator symbol represents the magnitude of a vector;
[0159] The first unit of the model is also used to ensure that all points in the data set are on the correct side of the preset and the corresponding distance Under the condition that is at least positive, the interval is maximized by the preset optimization objective function, and the Lagrange multiplier method is used to solve it to obtain the optimal normal vector of the first optimal hyperplane and the bias term ;
[0160] The first unit of the model is also used to calculate the normal vector and the bias term Establish a support vector machine model; the decision function of the support vector machine model is ,and The function is used to judge the positive and negative, and the category to which the data point belongs is determined based on the output of the function.
[0161] The first unit of the model in this embodiment can effectively control the complexity of the model by introducing a regularization term to avoid overfitting the model on the training data. This enables the model to show better generalization performance when facing unseen test data, that is, to more accurately classify or predict new data. By maximizing the classification interval, an optimal decision boundary can be found that can not only accurately classify the training data, but also maintain good classification performance for new data.
[0162] The second unit of the model is used to map the data set from the original space to a high-dimensional polynomial feature space;
[0163] The second unit of the model is also used to capture the polynomial combination mode in the data set through the polynomial kernel function to obtain nonlinear interaction data;
[0164] The second unit of the model is also used to search for an actual plane with the shortest distance to the nearest point of two types of data points in the high-dimensional polynomial feature space according to the nonlinear interaction data, and obtain a second optimal hyperplane; wherein the two types of data points are two types of data points in the data set divided according to the intensity;
[0165] The second model unit is also used to establish a polynomial vector machine model according to the second optimal hyperplane;
[0166] Among them, the polynomial kernel function is:
[0167] ;
[0168] in, and are two points in the data set, is the kernel coefficient, is a constant term, is the degree of the polynomial, yes and The dot product between .
[0169] In the second unit of the model of this embodiment, since the polynomial kernel function can express the complex interactions and combinations between features, the polynomial kernel function can better capture these interactions in concrete strength detection, thereby improving the prediction accuracy and robustness of the model.
[0170] The third unit of the model is used to calculate the Euclidean distance between different data points in the data set to obtain the Euclidean distance set;
[0171] The third unit of the model is also used to calculate the similarity measure of different data points in the data set in the high-dimensional feature space through the RBF kernel function according to the Euclidean distance set to obtain a similarity measure set;
[0172] The third unit of the model is also used to search for an actual plane with the shortest distance to the nearest points of the two types of data points in the high-dimensional polynomial feature space according to the similarity measurement set, so as to obtain a third optimal hyperplane; wherein the two types of data points are data points in the data set;
[0173] The third model unit is also used to establish a radial basis function vector machine model according to the third optimal hyperplane;
[0174] The third model unit is also used to form a model set consisting of a support vector machine model, a polynomial vector machine model and a radial basis function vector machine model;
[0175] Among them, the RBF kernel function is:
[0176] ;
[0177] in, It is a data point and data points The square of the Euclidean distance between is the bandwidth parameter of the kernel.
[0178] The RBF kernel function of the third unit of the model of this embodiment calculates the Euclidean distance between data points and defines the similarity based on these distances in high-dimensional space, which enables the model to capture the local similarity between data points well; in concrete strength detection, this means that data points of similar materials or proportions will be more closely linked together, which helps to improve the prediction accuracy of the model. By searching for the third optimal hyperplane that can maximize the classification interval in the high-dimensional feature space, the radial basis function vector machine model can maintain a good fit to the training data while maintaining good prediction capabilities for new data, which helps to reduce overfitting and improve the generalization ability of the model.
[0179] In the first, second and third units of the model of this embodiment, the linear kernel function is suitable for the case where the data itself is close to linearly separable; the polynomial kernel function can capture the nonlinear relationship between features; and the RBF kernel function further enhances the generalization ability of the model by finding the best hyperplane in the high-dimensional feature space. Integrating these models can combine their respective advantages, thereby improving the accuracy and stability of subsequent intensity predictions. Since multiple kernel functions are used, even if one of the kernel functions performs poorly on a specific data set, other kernel functions may still provide useful information. This diversity helps to reduce the overfitting or underfitting problems that may exist in a single model and enhance the robustness and adaptability of the entire model set.
[0180] Fractional unit, used to treat the preset properties of concrete as different eigenmodes, and obtain several eigenmodes;
[0181] The score unit is also used to integrate several feature modes based on concrete strength data and similarity formula through multi-feature fusion technology, and to capture the key factors affecting concrete similarity by applying the attention mechanism so that the model gives higher weights to more important features in the decision-making process, and calculates the similarity score between the strength prediction result of the vector machine model and the concrete strength data;
[0182] The score unit is also used to calculate the mean square error and F1 score based on the strength prediction results of the vector machine model and the concrete strength data;
[0183] The score unit is also used to assign weights to the similarity score, mean square error and F1 score based on the comprehensive score formula to calculate the comprehensive score;
[0184] The similarity formula is:
[0185] ;
[0186] The comprehensive score formula is:
[0187] ;
[0188] The F1 score is:
[0189] ;
[0190] For the parameters Precision and Recall, we have:
[0191] ;
[0192] ;
[0193] in, is the concrete batch, i is the concrete resource, t is the time; m is the modal type, m=1 represents cubic compressive strength, m=2 represents axial compressive strength, and m=3 represents tensile strength; is the weight of the mth mode, is the multimodal attention mechanism function, It is a concrete batch The high-level indicator feature vector of the mth mode at time t, is the high-level indicator eigenvector of the mth mode of concrete resource i at time t, and λ is the time decay factor; is the mth modal fusion coefficient, which is used to adjust the weights of different modalities;
[0194] It is The similarity score between the predicted strength result and the actual strength result of concrete samples, It is The mean square error between the predicted strength results and the actual strength results of concrete samples, It is The F1 score of the predicted strength results and the actual strength results of the concrete samples, , and are the feature weights of similarity, mean square error and F1 score respectively, is the total number of concrete samples;
[0195] Indicates the number of samples correctly predicted by the model as positive. It indicates the number of samples that the model predicts to be positive but are actually negative. Indicates the number of samples that the model predicts to be negative but are actually positive.
[0196] The score unit of this embodiment regards the various preset attributes of concrete as different feature modes, and integrates them through multi-feature fusion technology, which can make full use of the complementarity between different features and improve the feature expression ability of the similarity score. The introduction of the attention mechanism allows the model to automatically give higher weights to preset features in the decision-making process. This mechanism enables the model to more accurately capture the key factors affecting the similarity of concrete, thereby improving the accuracy and reliability of the prediction. By calculating the similarity score, mean square error and F1 score, and assigning weights to these indicators, the comprehensive score is finally obtained to evaluate the model performance, which can more comprehensively reflect the performance of the model in all aspects and provide a more comprehensive reference for the final selection of the model.
[0197] Initial unit, used to obtain sound velocity data of concrete by ultrasonic testing;
[0198] The initial unit is also used to select the model with the highest comprehensive score in the model set as the nonlinear model, perform intensity prediction based on the sound velocity data, and obtain the initial intensity prediction value.
[0199] In one embodiment, the verification module 30 includes a model unit and a prediction unit; specifically:
[0200] The model unit is used to standardize and normalize the estimated value of concrete strength, the predicted value of initial strength and any physical property data or chemical property data that may affect the strength of concrete to obtain a comprehensive data set;
[0201] The model unit is also used to divide the comprehensive data set into a training set and a test set according to a preset method;
[0202] The model unit is also used to train the initial adaptive model using the training set data and evaluate the performance of the model through the test set to obtain the adaptive model.
[0203] A prediction unit is used to set one or more adapter modules for capturing concrete-specific features and information, introduce a self-attention mechanism to enhance the adapter module's ability to capture key features, and add a dynamic feature selection mechanism to control the adapter module to automatically determine which features are more important for the current prediction task, and adjust the weight accordingly to selectively transmit information; wherein the adapter module is used to capture and process information related to the performance indicator set;
[0204] The prediction unit is also used to initialize the parameters of the adapter module, which is inserted before the preset last few layers of the adaptive model;
[0205] The prediction unit is also used to update the forward propagation logic of the adaptive model to include the processing steps of the adapter module;
[0206] The prediction unit is also used to locate the original parameters that need to be frozen in the adaptive model by setting the "requires_grad" attribute of the original parameters to "False";
[0207] The prediction unit is also used to adjust only the parameters of the adapter module in the adaptive model while keeping the preset original parameters of the adaptive model unchanged, so that the adaptive model is adapted on the performance indicator set to obtain a prediction model.
[0208] The prediction unit is also used to predict the concrete strength prediction value using the prediction model according to the weighted average of the concrete strength estimation value and the initial strength prediction value.
[0209] The prediction unit of this embodiment can combine the advantages of the estimated concrete strength value and the initial strength predicted value by weighted averaging, thereby reducing the error that may be caused by a single prediction method. Since the prediction model is obtained by adaptively adjusting the performance indicator set between the predicted value and the estimated concrete strength value, this means that the model can be dynamically optimized according to the actual prediction effect to better adapt to data changes and prediction task requirements. This method enables the model to maintain a high prediction accuracy when running for a long time or facing new data;
[0210] Moreover, for the prediction model, the self-attention mechanism allows the model to consider the information of all other features when processing each feature, so as to more comprehensively understand the internal structure and relevance of the data; the addition of a dynamic feature selection mechanism enables the adapter module to automatically determine which features are more important for the current prediction task and adjust the weights accordingly, which helps to improve the robustness and adaptability of the model. Only adjusting the parameters of the adapter module allows the model to focus on optimizing the most relevant parts of the current prediction task. This targeted optimization strategy helps to improve the performance and accuracy of the model.
[0211] Overall, this embodiment has the following beneficial effects:
[0212] The ultrasonic detection of the present invention can provide the sound velocity data inside the concrete, reflecting the density and integrity of the concrete, while the rebound detection infers its strength by measuring the hardness of the concrete surface. This fusion of multi-source data helps to improve the accuracy of concrete strength prediction. The nonlinear relationship between concrete strength and sound velocity can be better captured by using a nonlinear model to predict strength based on the sound velocity data of ultrasonic detection. In addition, by establishing a model set and selecting the model with the highest comprehensive score as the final nonlinear model, the optimal selection of the model is achieved; and the calculation of the comprehensive score takes into account the different attribute characteristics of concrete, and gives a higher weight to the preset features through the attention mechanism, which helps to screen out the model that best suits the current data set. Since cross-validation helps to discover possible overfitting or underfitting problems of the model, and can optimize the prediction results by adjusting the model parameters or structure, the reliability of the concrete strength prediction value can be further improved by using the concrete strength estimated value and the initial strength prediction value for cross-validation.
[0213] Embodiment three:
[0214] The embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the strength detection method for building concrete;
[0215] Wherein, the strength detection method of building concrete, if implemented in the form of a software functional unit and used as an independent product, can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.
[0216] The above are preferred embodiments of the present invention. It should be noted that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.
Claims
1. A method for testing the strength of building concrete, characterized in that: include: Conduct ultrasonic testing and rebound testing on concrete to obtain an estimated value of concrete strength; Acquire the sound velocity data of the concrete through ultrasonic testing, and use a nonlinear model to predict the strength according to the sound velocity data to obtain an initial strength prediction value; wherein the nonlinear model is obtained by taking the model with the highest comprehensive score in the model set, and the model set is obtained by establishing a support vector machine model, a polynomial vector machine model, and a radial basis function vector machine model according to preset concrete strength data; the comprehensive score is calculated by considering various properties of concrete as different feature modes and allowing the model to give higher weights to preset features in the decision-making process through an attention mechanism; Perform cross-validation based on the estimated concrete strength value and the initial strength prediction value to obtain a concrete strength prediction value; The comprehensive score is calculated by treating various properties of concrete as different feature modes and allowing the model to give higher weights to preset features in the decision-making process through the attention mechanism. Specifically, it is: The preset properties of concrete are regarded as different eigenmodes, and several eigenmodes are obtained; Based on the concrete strength data and similarity formula, the plurality of feature modes are integrated by multi-feature fusion technology, and the key factors affecting the concrete similarity are captured by applying the attention mechanism so that the model gives a higher weight to the preset features in the decision-making process, and the similarity score between the strength prediction result of the vector machine model and the concrete strength data is calculated; Calculating a mean square error and an F1 score based on the strength prediction result of the vector machine model and the concrete strength data; Based on a comprehensive score formula, weights are assigned to the similarity score, the mean square error, and the F1 score to calculate the comprehensive score; Wherein, the similarity formula is: ； The comprehensive score formula is: ； in, is the concrete batch, i is the concrete resource, t is the time, m is the modal type of concrete, is the weight of the mth mode, is the multimodal attention mechanism function, It is a concrete batch The high-level indicator feature vector of the mth mode at time t, is the high-level indicator eigenvector of the mth mode of concrete resource i at time t, λ is the time decay factor, is the mth modal fusion coefficient; It is The similarity score between the predicted strength result and the actual strength result of concrete samples, It is The mean square error between the predicted strength results and the actual strength results of concrete samples, It is The F1 score of the predicted strength results and the actual strength results of the concrete samples, , and are the feature weights of similarity, mean square error and F1 score respectively, is the total number of concrete samples.
2. A method for testing the strength of building concrete as claimed in claim 1, characterized in that: The model set is obtained by establishing a support vector machine model, a polynomial vector machine model, and a radial basis function vector machine model according to preset concrete strength data, specifically: Test the preset concrete samples to obtain basic data including sound velocity data and actual strength value; Performing data cleaning and standardization processing on the basic data to obtain a data set; Finding the best hyperplane in the original space of the data set by a linear kernel function to establish the support vector machine model; Capturing the nonlinear relationship between the features in the data set by a polynomial kernel function, and establishing the polynomial vector machine model; Based on the data set, the radial basis function vector machine model is established by searching for the best hyperplane in the high-dimensional feature space according to the similarity metric through the RBF kernel function; The model set is composed of the support vector machine model, the polynomial vector machine model and the radial basis function vector machine model.
3. A strength testing method for building concrete as claimed in claim 2, characterized in that: The optimal hyperplane is found in the original space of the data set by a linear kernel function, and the support vector machine model is established, specifically: A regularization term is introduced to control the complexity of the model, and an actual plane with the shortest distance to the nearest point of the two types of data points is searched in the original space of the data set to obtain a first optimal hyperplane; wherein the two types of data points are data points in the data set; The support vector machine model is established according to the first optimal hyperplane.
4. A strength testing method for building concrete as claimed in claim 2, characterized in that: The polynomial kernel function is used to capture the nonlinear relationship between the features in the data set, and the polynomial vector machine model is established, specifically: Mapping the data set from the original space to a high-dimensional polynomial feature space; Capturing the polynomial combination mode in the data set by the polynomial kernel function to obtain nonlinear interaction data; According to the nonlinear interaction data, searching in the high-dimensional polynomial feature space for an actual plane that is shortest to the nearest points of the two types of data points, to obtain a second optimal hyperplane; wherein the two types of data points are data points in the data set; The polynomial vector machine model is established according to the second optimal hyperplane.
5. A method for testing the strength of building concrete as claimed in claim 2, characterized in that: Based on the data set, the RBF kernel function is used to find the best hyperplane in the high-dimensional feature space according to the similarity metric, and the radial basis function vector machine model is established, specifically: Calculating the Euclidean distances between different data points in the data set to obtain a Euclidean distance set; According to the Euclidean distance set, similarity measures of different data points in the data set in the high-dimensional feature space are calculated by the RBF kernel function to obtain a similarity measure set; According to the similarity measure set, searching in the high-dimensional polynomial feature space for an actual plane that is shortest to the nearest points of the two types of data points, to obtain a third optimal hyperplane; wherein the two types of data points are data points in the data set; The radial basis function vector machine model is established according to the third optimal hyperplane.
6. A method for testing the strength of building concrete as claimed in claim 1, characterized in that: A cross-validation is performed based on the estimated concrete strength value and the initial strength prediction value to obtain a concrete strength prediction value, specifically: According to the weighted average of the estimated concrete strength value and the initial strength predicted value, the predicted concrete strength value is predicted using a prediction model; Among them, the prediction model is obtained by adjusting the adaptive model according to a set of performance indicators between the predicted value and the concrete strength estimated value, the predicted value is obtained by training and running the adaptive model using a training set and a test set, and the training set and the test set are obtained by dividing the concrete strength estimated value and the initial strength predicted value in a preset manner.
7. A strength detection system for building concrete, characterized in that: It includes presumption module, prediction module and verification module; The estimation module is used to perform ultrasonic testing and rebound testing on concrete to obtain an estimated value of concrete strength; The prediction module is used to obtain the sound velocity data of the concrete through ultrasonic testing, and use a nonlinear model to predict the strength according to the sound velocity data to obtain an initial strength prediction value; wherein the nonlinear model is obtained by taking the model with the highest comprehensive score in the model set, and the model set is obtained by establishing a support vector machine model, a polynomial vector machine model, and a radial basis function vector machine model according to preset concrete strength data; the comprehensive score is calculated by considering various properties of concrete as different feature modes and allowing the model to give higher weights to preset features in the decision-making process through an attention mechanism; The verification module is used to perform cross-validation based on the estimated concrete strength value and the initial strength prediction value to obtain a concrete strength prediction value; Wherein, the prediction module includes a score unit; The fractional unit is used to regard the preset properties of concrete as different characteristic modes to obtain a plurality of characteristic modes; The score unit is further used to integrate the several feature modes through multi-feature fusion technology based on the concrete strength data and the similarity formula, and to capture the key factors affecting the concrete similarity by applying the attention mechanism so that the model gives a higher weight to the preset features in the decision-making process, and calculate the similarity score between the strength prediction result of the vector machine model and the concrete strength data; The score unit is further used to calculate the mean square error and the F1 score according to the strength prediction result of the vector machine model and the concrete strength data; The score unit is further used to perform weight allocation on the similarity score, the mean square error and the F1 score based on a comprehensive score formula to calculate the comprehensive score; Wherein, the similarity formula is: ； The comprehensive score formula is: ； in, is the concrete batch, i is the concrete resource, t is the time, m is the modal type of concrete, is the weight of the mth mode, is the multimodal attention mechanism function, It is a concrete batch The high-level indicator feature vector of the mth mode at time t, is the high-level indicator eigenvector of the mth mode of concrete resource i at time t, λ is the time decay factor, is the mth modal fusion coefficient; It is The similarity score between the predicted strength result and the actual strength result of concrete samples, It is The mean square error between the predicted strength results and the actual strength results of concrete samples, It is The F1 score of the predicted strength results and the actual strength results of the concrete samples, , and are the feature weights of similarity, mean square error and F1 score respectively, is the total number of concrete samples.