SVR-based concrete strength prediction system and method

Abstract

The invention relates to the technical field of concrete quality control, in particular to a concrete strength prediction system and method based on SVR. According to the invention, multi-dimensional data of rebound test, compressive strength test and carbonization depth value measurement are collected through the data acquisition module, rich input characteristics are provided for the model, and the prediction capability of the model is enhanced; and abnormal data is eliminated by adopting a Grubbs criterion, so that the quality and consistency of input data are ensured, and the influence of noise on model training is reduced. According to the method, a concrete rebound curved surface regression equation and a support vector regression model are established respectively and serve as prediction criteria, data fusion is carried out through a Kalman filter, and the prediction precision and reliability are further improved. Through automatic and standardized data processing procedures including data acquisition, preprocessing, regression prediction, SVR prediction, data fusion and the like, rapid prediction of the concrete strength is realized, and the prediction efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of concrete quality control, and in particular to a concrete strength prediction system and method based on SVR. Background Art

[0002] Concrete quality control technology is a key link to ensure that concrete performance meets design requirements and guarantees project quality and safety, including raw material quality control, mix design, production process control, maintenance environment control and post-testing. Concrete strength is the core content of concrete structure quality control, an important basis for structural design and construction, and one of the most important properties of concrete. Therefore, the innovation and application of its testing methods are particularly important.

[0003] The traditional way to test concrete strength is to use the compressive strength test of standard cubic test blocks as the main basis for assessing concrete strength. However, in actual engineering applications, the results of the compressive strength test of standard cubic test blocks may not accurately reflect the concrete strength of the structural entity due to insufficient number of test blocks retained and differences between the structural entity environment and the storage environment of the retained test blocks. Therefore, in order to avoid the above problems, on-site testing is often used in engineering to test concrete strength. Common testing methods include local damage testing and non-destructive testing.

[0004] In addition, with the rapid development of modern science and technology, machine learning algorithms have become more mature and play an important role in practical engineering applications. Machine learning is a branch of artificial intelligence that uses computer programs to automatically learn patterns and rules from data, and achieves accurate prediction or decision-making of unknown data by continuously optimizing the performance of the prediction model. The support vector machine regression algorithm (SVR) is a branch of SVM, which can effectively solve nonlinear regression problems, convert low-dimensional nonlinear mapping problems to high-dimensional linear, and return the regression results to the space through inverse mapping after completing linear regression in high-dimensional space until the regression results of the entire sample data set are completed. It can solve problems such as limited sample data and nonlinear data relationships, which just meets the current status of concrete compressive strength prediction. At the same time, SVR requires fewer parameter definitions, making modeling easier and reducing computational costs.

[0005] There are many shortcomings in the existing concrete strength testing methods. First, the local damage detection method will cause certain damage to the structure during the detection process, which will cause certain damage to the building structure. In addition, the structure needs to be maintained after the detection, which increases costs and is not conducive to large-scale detection of the structure.

[0006] Secondly, the commonly used non-destructive testing methods have limited detection accuracy, and the results of some non-destructive testing methods require manual interpretation and analysis by professional engineers, which may lead to subjectivity in the interpretation of the results and also limit the interpretability of the data.

[0007] Finally, although existing technologies have begun to attempt to establish AI-based nondestructive testing methods for concrete, there are still many limitations on their use. Existing testing data sets cannot cover all types of concrete and different working conditions, resulting in poor generalization of the model when facing new scenarios and new samples, and prone to problems such as overfitting and underfitting.

[0008] In view of the above problems, it is necessary to propose a concrete strength prediction system and method based on SVR. Summary of the invention

[0009] The purpose of the present invention is to solve the problems existing in the background technology and to propose a concrete strength prediction system and method based on SVR.

[0010] The purpose of the present invention can be achieved through the following technical solutions:

[0011] A concrete strength prediction system and method based on SVR.

[0012] In a first aspect, the present invention provides a concrete strength prediction system based on SVR, including a data acquisition module, a data preprocessing module, a regression prediction module, an SVR prediction module and a data fusion module.

[0013] The data acquisition module collects the result data of the rebound test, compressive strength test and carbonation depth value determination of the concrete specimen.

[0014] Each concrete specimen is numbered and subjected to a compressive strength test. The load at failure is recorded and the compressive strength is calculated.

[0015] Test points were arranged on each test block, and the rebound test and carbonization depth were measured. The average rebound value and the average carbonization depth were calculated.

[0016] The data preprocessing module inputs all the rebound values, compressive strength values ​​and carbonization depth values ​​into the outlier processing program, and removes the abnormal data through the Grubbs criterion, including the abnormal data removal of the rebound values, compressive strength values ​​and carbonization depth values.

[0017] The outlier processing results are renumbered and the aligned data are integrated to generate a data vector, which is then sent to the regression prediction module.

[0018] The data vector is subjected to Z-score normalization, including normalization of rebound value, compressive strength, carbonization depth, age and design strength grade, and the normalization result is sent to the SVR prediction module.

[0019] The regression prediction module fits the relationship between compressive strength, rebound value and carbonization depth through mathematical functions based on the data vector.

[0020] The regression equation of concrete rebound surface is established, and the best fitting parameters are solved by the least squares method.

[0021] Draw the fitting surface image and save the concrete rebound surface regression equation as the first prediction criterion.

[0022] The SVR prediction module builds a support vector regression model, including an input layer, a kernel function, and an output layer.

[0023] Set hyperparameters, including regularization parameters, error tolerance, kernel function parameters, optimization objective function and constraints. Solve the model parameters through the SMO optimization algorithm and train the support vector regression model as the second prediction criterion.

[0024] The data fusion module performs data fusion on the output results of the first prediction criterion and the second prediction criterion.

[0025] The two prediction results are weighted and fused through the Kalman filter to calculate the optimal estimate.

[0026] The filtering results are recorded and output as the final concrete strength prediction value.

[0027] In a second aspect, the present invention provides a concrete strength prediction method based on SVR, comprising the following steps:

[0028] Step 1: Material analysis and data collection;

[0029] The result data of the rebound test, compressive strength test and carbonation depth value determination of each concrete specimen i are collected, including the rebound value Ri, compressive strength fi and carbonation depth hi.

[0030] Step 2: Data preprocessing and cleaning;

[0031] All the collected rebound values ​​Ri, compressive strength fi and carbonization depth hi are input into the outlier processing program, and the outlier data are eliminated by the Grubbs criterion to obtain the outlier processing result, that is, the data vector (Rj, fj, hj)

[0032] Step 3: Data fitting and regression model establishment;

[0033] S301, regression model data fitting;

[0034] According to the data vector (Rj, fj, hj), the relationship between compressive strength, rebound value and carbonation depth is fitted through mathematical functions to obtain the concrete rebound surface regression equation f = k × R b ×h c The parameters k, b and c in .

[0035] The concrete rebound surface with known parameters is regressed into the equation f = k × R b ×h c Recorded as the first prediction criterion.

[0036] S302. Constructing a support vector regression model

[0037] Obtain the concrete age tj and design strength grade Cj corresponding to the concrete specimen number j in the data vector (Rj, fj, hj), and construct a new data vector (tj, Cj, Rj, fj, hj).

[0038] Support vector regression model training is performed based on the new data vector (tj, Cj, Rj, fj, hj).

[0039] The support vector regression model includes an input layer, a kernel function and an output layer.

[0040] The input layer includes 4 nodes, which are used to input Rj, hj, tj and Cj respectively; the kernel function layer; the kernel function maps the input data to a high-dimensional space and assigns a nonlinear relationship between the input data.

[0041] The output layer includes 1 output node, which is used to output the predicted compressive strength fj.

[0042] Set hyperparameters: regularization parameter C, error tolerance ε, and hyperparameter γ.

[0043] Set the insensitive loss coefficient, kernel function, optimization objective function and constraints of the support vector regression model.

[0044] Substitute the new data vector (tj, Cj, Rj, fj, hj), and use the SMO optimization algorithm to solve the model parameters and record the solution results in combination with the optimization objective function and constraints. As the final parameters of the trained support vector regression model.

[0045] The trained support vector regression model is used as the second prediction criterion.

[0046] Step 4: Prediction of concrete strength;

[0047] Take the concrete used to predict the compressive strength and prepare several concrete test blocks.

[0048] The basic data of the concrete test blocks are obtained, including rebound value, carbonation depth, age and design strength grade.

[0049] Inputting the rebound values ​​and carbonation depths of the plurality of concrete test blocks into a first prediction criterion to obtain an output result of the first prediction criterion;

[0050] The rebound values, carbonation depths, ages and design strength grades of the plurality of concrete test blocks are input into a second prediction criterion to obtain an output result of the second prediction criterion.

[0051] Step 5: Prediction data fusion;

[0052] The Kalman filter is used to fuse the output result of the first prediction criterion and the output result of the second prediction criterion to obtain the final concrete strength prediction result.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] 1. The present invention collects multi-dimensional data of rebound test, compressive strength test and carbonization depth value determination through the data acquisition module, provides rich input features for the model, and enhances the prediction ability of the model; adopts the Grubbs criterion to eliminate abnormal data, ensures the quality and consistency of input data, and reduces the influence of noise on model training.

[0055] The present invention establishes a concrete rebound surface regression equation and a support vector regression model respectively, and uses the two as prediction criteria, and performs data fusion through a Kalman filter, thereby further improving the accuracy and reliability of the prediction.

[0056] The present invention realizes rapid prediction of concrete strength and improves prediction efficiency through an automated and standardized data processing process, including steps such as data collection, preprocessing, regression prediction, SVR prediction and data fusion. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings:

[0058] Figure 1 is a flow chart of the method of the present invention;

[0059] Figure 2 Schematic diagram of the test point layout of the concrete specimen.

[0060] Figure 3 It is a schematic diagram of the abnormal value processing procedure of the present invention;

[0061] Figure 4 It is a schematic diagram of the fitting surface of the concrete rebound surface regression equation;

[0062] Figure 5 This is a schematic diagram of the structural topology of the support vector regression model;

[0063] Figure 6 It is a schematic diagram of the Kalman filtering result of the present invention;

[0064] Figure 7 Schematic diagram of probability distribution of Kalman filtering results of the present invention;

[0065] Figure 8 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0066] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] See also Figure 1 As shown, a concrete strength prediction system based on SVR includes a data acquisition module, a data preprocessing module, a regression prediction module, an SVR prediction module and a data fusion module.

[0068] The data acquisition module collects the result data of the rebound test, compressive strength test and carbonation depth value determination of the concrete specimen.

[0069] All concrete test blocks are numbered with the number symbol i, i = 1, 2, 3, ..., n; n is the total number of concrete test blocks. The compressive strength test is performed on each concrete test block i to obtain its compressive strength fi;

[0070] The compressive strength test is specifically:

[0071] The pressure test machine is used to uniformly load the upper and lower pressure surfaces along the central axis of each concrete test block i at a preset loading speed until the test block fails. The load Fi at the time of concrete test block i failure is recorded and the formula is used to calculate the load Fi. Calculate the compressive strength fi of concrete specimen i. Where Ai is the compressive area of ​​concrete specimen i.

[0072] See also Figure 2 As shown, several test points are arranged on each concrete test block i, and rebound test and carbonation depth measurement are performed at each test point to obtain the rebound value and carbonation depth of each test point. The average rebound value of all test points is calculated to obtain the rebound value Ri of the entire concrete test block i; the average carbonation depth of all test points is calculated to obtain the carbonation depth hi of the entire concrete test block i.

[0073] The rebound test is specifically as follows: the rebound hammer is placed vertically at the test point to ensure that the probe of the instrument is in full contact with the surface. Then, the pressure head of the rebound hammer is pressed hard to compress the spring inside and rebound, and the rebound value displayed by the rebound hammer is recorded.

[0074] The carbonization depth value determination is specifically as follows: after completing the rebound value test and the compressive strength test, a hole with a preset depth and a preset diameter is drilled into each test point on the surface of the fractured test block using a drilling tool. The debris and powder in the hole are cleaned, and a phenolphthalein alcohol solution is applied to the inner wall of the hole. When a clear color dividing line appears on the inner wall of the hole, the dividing line is determined to be the dividing line between the carbonized area and the non-carbonized area of ​​the concrete. The carbonization depth is measured by a carbonization depth measuring instrument.

[0075] The data preprocessing module inputs all the collected rebound values ​​Ri, compressive strength fi and carbonization depth hi into the abnormal value processing program, and eliminates abnormal data through the Grubbs criterion.

[0076] See also Figure 3 As shown, the abnormal value processing procedure is specifically as follows:

[0077] STEP1: Start the program;

[0078] Instruction: Clear the program cache and running data in the previous round of operation;

[0079] STEP2: Capture rebound value data;

[0080] Instruction: retrieve all rebound value data and record the data volume n1;

[0081] STEP3: Capture compressive strength data;

[0082] Command: retrieve all compressive strength data and record the data volume n2;

[0083] STEP4: Carbonization depth data capture;

[0084] Command: retrieve all carbonization depth data and record the data volume n3;

[0085] STEP5: Data number distinction;

[0086] Instruction: Use the numbering symbol Ri1 to number the rebound value data, i1∈(1, n1);

[0087] Use the numbering symbol fi2 to number the compressive strength data, i2∈(1,n2);

[0088] The carbonization depth data are numbered using the numbering symbol hi3, i3∈(1, n3);

[0089] STEP6: Extract statistical features of rebound value;

[0090] Instructions: By formula Calculate the mean μ(R) and variance σ(R) of the rebound value data;

[0091] Feedback: Skip to STEP9;

[0092] STEP7: Extract statistical features of compressive strength;

[0093] Instructions: By formula Calculate the mean μ(f) and variance σ(f) of the compressive strength data;

[0094] Feedback: Skip to STEP12;

[0095] STEP8: Extraction of statistical features of carbonization depth;

[0096] Instructions: By formula Calculate the mean μ(h) and variance σ(h) of the carbonization depth data;

[0097] Feedback: Skip to STEP15;

[0098] STEP9: Calculate the upper data volume of the rebound value;

[0099] Instructions: By formula Calculate the Grubbs criterion outlier G(Ri1) of each rebound value Ri1;

[0100] Feedback 1: G(Ri1) is greater than the preset threshold Gmax → execute STEP10;

[0101] Feedback 2: G(Ri1) does not exist and is greater than the preset threshold Gmax → execute STEP18;

[0102] STEP10: Mark the abnormal rebound value data;

[0103] Instruction: Mark the rebound value Ri1 corresponding to the Grubbs criterion abnormal value G(Ri1) greater than the preset threshold value Gmax as abnormal rebound value data, and obtain the number of abnormal rebound value data k1;

[0104] STEP11: Eliminate abnormal rebound value data

[0105] Instruction: Delete the abnormal rebound value data from all rebound values ​​Ri1, and set n1 = n1-k1;

[0106] Feedback: Return to STEP6:

[0107] STEP12: Calculate the upper data volume of compressive strength;

[0108] Instructions: By formula Calculate the Grubbs criterion anomaly G(fi2) for each compressive strength fi2;

[0109] Feedback 1: G(fi2) is greater than the preset threshold Gmax → execute STEP13;

[0110] Feedback 2: G(fi2) does not exist and is greater than the preset threshold Gmax → execute STEP19;

[0111] STEP13: Marking of abnormal compressive strength data;

[0112] Instruction: mark the compressive strength fi2 corresponding to the Grubbs criterion abnormal value G(fi2) greater than the preset threshold Gmax as abnormal compressive strength data, and obtain the number of abnormal compressive strength data k2;

[0113] STEP 14: Eliminate abnormal compressive strength data

[0114] Command: Delete abnormal compressive strength data from all compressive strength fi2, and set n2 = n2-k2;

[0115] Feedback: Return to STEP7:

[0116] STEP15: Calculate the data volume on the upper side of the carbonization depth;

[0117] Instructions: By formula Calculate the Grubbs criterion anomaly value G(hi3) for each carbonization depth hi3;

[0118] Feedback 1: G(hi3) is greater than the preset threshold Gmax → execute STEP16;

[0119] Feedback 2: G(hi3) does not exist and is greater than the preset threshold Gmax → execute STEP20;

[0120] STEP16: Carbonization depth abnormal data marking;

[0121] Instruction: mark the carbonization depth hi3 corresponding to the Grubbs criterion abnormal value G(hi3) greater than the preset threshold value Gmax as carbonization depth abnormal data, and obtain the number k3 of carbonization depth abnormal data;

[0122] STEP 17: Eliminate abnormal data of carbonization depth

[0123] Instruction: Delete abnormal carbonization depth data from all carbonization depths hi3, and set n3 = n3-k3;

[0124] Feedback: Return to STEP8:

[0125] STEP18: Output the rebound value processing result;

[0126] Instruction: Generate end signal 1;

[0127] Feedback: Record all remaining Ri1;

[0128] STEP19: Output the compressive strength processing results;

[0129] Instruction: Generate end signal 2;

[0130] Feedback: Record all remaining fi2;

[0131] STEP20: Output the carbonization depth treatment results;

[0132] Instruction: Generate end signal three;

[0133] Feedback: Record all remaining hi3s;

[0134] STEP21: Data integration and alignment;

[0135] Instruction: Identify the number symbols in the rebound value data Ri1, the compressive strength data fi2 and the carbonization depth data hi3, that is, the items where i1, i2 and i3 are completely consistent, and mark them as aligned data;

[0136] Feedback: Integrate the aligned data and assign a new number j to generate a data vector (Rj, fj, hj) = (Ri1, fi2, hi3); where i1 = i2 = i3; where j = 1, 2, 3, ..., m; where m is the total number of data vectors obtained after integrating the aligned data;

[0137] STEP22: Output and end;

[0138] Instructions: Output all data vectors (Rj, fj, hj); where j = 1, 2, 3, ..., m; Output the average value μ(R) and variance σ(R) of the rebound value data in the last cycle; Output the average value μ(f) and variance σ(f) of the compressive strength data in the last cycle; Output the average value μ(h) and variance σ(h) of the carbonization depth data in the last cycle;

[0139] Feedback: End all calculation processes;

[0140] All output data vectors (Rj, fj, hj) are copied into two copies, one of which is sent to the regression prediction module and the other is input to the normalization processing program.

[0141] The normalization processing procedure is specifically as follows:

[0142] Get the mean μ(X) and standard deviation σ(X) of Rj; Get the mean μ(X) and standard deviation σ(X) of Rj; Calculate the mean μ(X) and standard deviation σ(X) of Rj;

[0143] By formula Z-score standardization is performed to transform Rj, fj and hj into preprocessed data Rj', fj' and hj' with a mean of 0 and a standard deviation of 1.

[0144] Furthermore, the age tj and design strength grade Cj of the concrete specimen corresponding to each data vector (Rj, fj, hj) are collected.

[0145] And through the formula Perform Z-score standardization to transform tj and Cj into preprocessed data tj' and Cj' with a mean of 0 and a standard deviation of 1. Where μ(t) is the age mean and μ(t) is the age variance; where μ(C) is the design strength grade mean and μ(C) is the design strength grade variance.

[0146] All Rj', fj', hj', tj' and Cj' are combined into a new data vector (Rj', fj', hj', tj', Cj') and output to the SVR prediction module.

[0147] It should be noted that when using SVR support vector regression, since SVR is very sensitive to the scale of features, input data normalization is a key step to ensure effective model training. The purpose of standardization is to convert each feature to the same scale range, usually by converting it to a distribution with a mean of 0 and a variance of 1, or scaling it to a fixed range. After standardization, the impact of various input data on the model output results will be more balanced, thereby improving the model effect.

[0148] The regression prediction module fits the relationship between compressive strength, rebound value and carbonization depth through mathematical functions according to the data vector (Rj, fj, hj).

[0149] The concrete rebound surface regression equation is established as follows:

[0150] f=k×R b ×h c Where k is the influence coefficient; b and c are the influence indexes; f is the compressive strength, R is the rebound value; and h is the carbonization depth.

[0151] Substituting the experimental data, the linear regression operation was carried out through the least square method to obtain the specific values ​​of the influence coefficient k and the influence indexes b and c.

[0152] The specific operation process of the least squares method is as follows:

[0153] First, take the logarithmic form of the concrete rebound surface regression equation: lnf = lnk + b × lnR + c × lnh,

[0154] Then, the logarithmic concrete rebound surface regression equation is transformed into lnf cu =y', let lnk = a; let lnR = x, let lnh = z, and we can get the regression equation of the concrete rebound surface after substitution: y' = bx + cz + a.

[0155] Then, let the loss function The partial derivatives of a, b and c are equal to 0, that is

[0156] Expanding the equation yields:

[0157] Substitution

[0158] have to

[0159] Finally, substitute the data vector (Rj, fj, hj) to expand the operation and calculate the specific values ​​of k, b and c.

[0160] For example, substitute the following data Rj, fj and hj:

[0161]

[0162]

[0163]

[0164] The best fitting parameters were obtained: k = 0.0326; b = 1.9143; c = -0.0902.

[0165] That is, f = 0.0326 × R 1.9143 ×h -0.902 .

[0166] See also Figure 4 As shown, the concrete rebound surface regression equation f = 0.0326 × R 1.9143 ×h -0.902 The fitted surface image of

[0167] The concrete rebound surface regression equation is saved as the first prediction criterion.

[0168] The SVR prediction module builds a support vector regression model and performs model training.

[0169] The support vector regression model is specifically:

[0170] See also Figure 5 As shown, the support vector regression model includes an input layer, a kernel function and an output layer.

[0171] The input layer includes 4 nodes, which are used to input Rj, hj, tj and Cj respectively; the kernel function layer; the kernel function maps the input data to a high-dimensional space and assigns a nonlinear relationship between the input data.

[0172] The output layer includes 1 output node, which is used to output the predicted compressive strength fj.

[0173] It should be noted that support vector regression is a regression method based on the support vector machine algorithm, which is mainly used to deal with regression problems. The basic idea of ​​SVR is to map the original data from low-dimensional space to high-dimensional space through a mapping function, and then find the optimal hyperplane in the high-dimensional space to predict the output.

[0174] The specific process of the model training is as follows:

[0175] Set hyperparameters: regularization parameter C, error tolerance ε, and hyperparameter γ.

[0176] It should be noted that the regularization parameter C controls the balance between model complexity and training error. A larger C value will make the model pay more attention to the training error, which may easily lead to overfitting. A smaller C value will make the model smoother and increase generalization ability; the error tolerance ε controls the smoothness of the regression function. A larger ε value means that a larger error can be tolerated, while a smaller ε value means that the regression model will strictly fit the training data; the hyperparameter γ is the training parameter of the kernel function, which controls the influence range of a single training sample. If the γ value is large, the influence range of a single sample is small, which may lead to overfitting; a smaller γ value will make the model smoother, which may lead to underfitting.

[0177] Setting insensitive loss coefficient Where fj is the actual value, i.e. the measured value of the compressive strength of concrete specimen j; is the predicted value, that is, the actual output value of the support vector regression model; ε is preset as the tolerance error threshold.

[0178] Set kernel function: K(x, xw) = exp(-γ×||x1-x2|| 2 ). x and xw are the input values ​​of the kernel function, including Rj, hj, tj and Cj; γ is a hyperparameter. w = 1, 2, 3, ..., N. N is the kernel function node number.

[0179] Set the optimization objective function: Where w is the weight

[0180] Quantity;||w|| 2 is the regularization term of the weight vector; is the penalty error term for setting the optimization goal; where ξ i and is the slack variable, Represents the error degree of the jth sample; where C is the regularization parameter.

[0181] The input data is mapped to the high-order space through the set kernel function K(xi, xj), and the mapping formula is: where α i and is the support vector; K(x1, x2) is the preset kernel function; b is the bias coefficient.

[0182] Set constraints:

[0183] According to the optimization objective function and constraints, the SMO optimization algorithm is used to solve the model parameters and record the solution results. As the final parameters of the trained support vector regression model.

[0184] The trained support vector regression model is used as the second prediction criterion.

[0185] The data fusion module performs data fusion on the output results of the first prediction criterion and the second prediction criterion.

[0186] See also Figure 6 and Figure 7 As shown, the two prediction results are fused by using a Kalman filter, and the first prediction criterion and the second prediction criterion are fused into a prediction result with a small mean and variance.

[0187] The specific process of data fusion is: obtain the output result of the first prediction criterion: f1j={f11, f12, ..., f1m}; obtain the output result of the second prediction criterion: f2j={f21, f22, ..., f2m}.

[0188] Calculate the mean μ(f1j) and μ(f2j) of f1j and f2j respectively; calculate the variance σ of f1j and f2j respectively 2 (f1j) and σ 2 (f2j).

[0189] The Kalman filter is used to perform weighted fusion of the two data sets and calculate the optimal estimate

[0190] The Kalman filter formula is:

[0191] Where K1 and K2 are Kalman gain coefficients; is the error estimate variance of the previous number j, and its update formula is

[0192] Record the filtering results, that is, the final output The final predicted value of concrete strength is determined and output.

[0193] See also Figure 8 As shown, a concrete strength prediction method based on SVR includes the following steps:

[0194] Step 1: Material analysis and data collection;

[0195] The result data of the rebound test, compressive strength test and carbonation depth value determination of each concrete specimen i are collected, including the rebound value Ri, compressive strength fi and carbonation depth hi.

[0196] Step 2: Data preprocessing and cleaning;

[0197] All the collected rebound values ​​Ri, compressive strength fi and carbonization depth hi are input into the outlier processing program, and the outlier data are eliminated by the Grubbs criterion to obtain the outlier processing result, that is, the data vector (Rj, fj, hj)

[0198] Step 3: Data fitting and regression model establishment;

[0199] S301, regression model data fitting;

[0200] According to the data vector (Rj, fj, hj), the relationship between compressive strength, rebound value and carbonation depth is fitted through mathematical functions to obtain the concrete rebound surface regression equation f = k × R b ×h c The parameters k, b and c in .

[0201] The concrete rebound surface with known parameters is regressed into the equation f = k × R b ×h c Recorded as the first prediction criterion.

[0202] S302. Constructing a support vector regression model

[0203] Obtain the concrete age tj and design strength grade Cj corresponding to the concrete specimen number j in the data vector (Rj, fj, hj), and construct a new data vector (tj, Cj, Rj, fj, hj).

[0204] Support vector regression model training is performed based on the new data vector (tj, Cj, Rj, fj, hj).

[0205] The support vector regression model includes an input layer, a kernel function and an output layer.

[0206] The input layer includes 4 nodes, which are used to input Rj, hj, tj and Cj respectively; the kernel function layer; the kernel function maps the input data to a high-dimensional space and assigns a nonlinear relationship between the input data.

[0207] The output layer includes 1 output node, which is used to output the predicted compressive strength fj.

[0208] Set hyperparameters: regularization parameter C, error tolerance ε, and hyperparameter γ.

[0209] Set the insensitive loss coefficient, kernel function, optimization objective function and constraints of the support vector regression model.

[0210] Substitute the new data vector (tj, Cj, Rj, fj, hj), and use the SMO optimization algorithm to solve the model parameters and record the solution results in combination with the optimization objective function and constraints. As the final parameters of the trained support vector regression model.

[0211] The trained support vector regression model is used as the second prediction criterion.

[0212] Step 4: Prediction of concrete strength;

[0213] Take the concrete used to predict the compressive strength and prepare several concrete test blocks.

[0214] The basic data of the concrete test blocks are obtained, including rebound value, carbonation depth, age and design strength grade.

[0215] Inputting the rebound values ​​and carbonation depths of the plurality of concrete test blocks into a first prediction criterion to obtain an output result of the first prediction criterion;

[0216] The rebound values, carbonation depths, ages and design strength grades of the plurality of concrete test blocks are input into a second prediction criterion to obtain an output result of the second prediction criterion.

[0217] Step 5: Prediction data fusion;

[0218] The Kalman filter is used to fuse the output result of the first prediction criterion and the output result of the second prediction criterion to obtain the final concrete strength prediction result.

[0219] It should be understood that the terms “include” and “comprising” used in the specification and claims of the present disclosure indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0220] It should also be understood that the terms used in this disclosure are only for the purpose of describing specific embodiments and are not intended to limit the disclosure. As used in this disclosure and the claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should also be further understood that the term "and / or" used in this disclosure and the claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations;

[0221] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A concrete strength prediction system based on SVR, comprising a data preprocessing module, a regression prediction module and an SVR prediction module, characterized in that: The data preprocessing module inputs all rebound values, compressive strength values ​​and carbonization depth values ​​into the abnormal value processing program, and removes abnormal data through the Grubbs criterion, including the removal of abnormal data of rebound values, compressive strength values ​​and carbonization depth values; Renumber the outlier processing results and integrate the alignment data to generate a data vector; then, send the data vector to the regression prediction module; Performing Z-score normalization processing on the data vector, including normalization of rebound value, compressive strength, carbonization depth, age and design strength grade; and sending the normalization processing result to the SVR prediction module; The regression prediction module fits the relationship between compressive strength, rebound value and carbonization depth through mathematical functions according to the data vector; The regression equation of concrete rebound surface is established, and the best fitting parameters are solved by the least square method. Draw the fitted surface image and save the concrete rebound surface regression equation as the first prediction criterion; The SVR prediction module builds a support vector regression model, including an input layer, a kernel function, and an output layer; Set hyperparameters, including regularization parameters, error tolerance, kernel function parameters, optimization objective function and constraints; solve model parameters through SMO optimization algorithm, and train the support vector regression model as the second prediction criterion.

2. A concrete strength prediction system based on SVR according to claim 1, characterized in that: It also includes a data acquisition module and a data fusion module; The data acquisition module collects the result data of the rebound test, compressive strength test and carbonation depth value determination of the concrete test block; Use the number symbol i to number each concrete test block, perform a compressive strength test, record the load at failure, and calculate the compressive strength fi; Arrange test points on each test block, perform rebound test and carbonization depth measurement, and calculate the average rebound value Ri and the average carbonization depth hi; The data fusion module performs data fusion on the output results of the first prediction criterion and the second prediction criterion; The two prediction results are weighted and fused through the Kalman filter to calculate the optimal estimate; The filtering results are recorded and output as the final concrete strength prediction value.

3. The SVR-based concrete strength prediction system according to claim 1, characterized in that: The outlier processing procedure is specifically as follows: STEP1: Start the program; Instruction: Clear the program cache and running data in the previous round of operation; STEP2: Capture rebound value data; Instruction: retrieve all rebound value data and record the data volume n1; STEP3: Capture compressive strength data; Command: retrieve all compressive strength data and record the data volume n2; STEP4: Carbonization depth data capture; Command: retrieve all carbonization depth data and record the data volume n3; STEP5: Data number distinction; Instruction: Use the numbering symbol Ri1 to number the rebound value data, i1∈(1, n1); Use the number symbol fi2 to number the compressive strength data, i2∈(1,n2); Use the numbering symbol hi3 to number the carbonization depth data, i3∈(1, n3); STEP6: Extract statistical features of rebound value; Instructions: By formula Calculate the mean μ(R) and variance σ(R) of the rebound value data; Feedback: Skip to STEP9; STEP7: Extract statistical features of compressive strength; Instructions: By formula Calculate the mean μ(f) and variance σ(f) of the compressive strength data; Feedback: Skip to STEP12; STEP8: Extraction of statistical features of carbonization depth; Instructions: By formula Calculate the mean μ(h) and variance σ(h) of the carbonization depth data; Feedback: Skip to STEP15; STEP9: Calculate the upper data volume of the rebound value; Instructions: By formula Calculate the Grubbs criterion outlier G(Ri1) of each rebound value Ri1; Feedback 1: G(Ri1) is greater than the preset threshold Gmax → execute STEP10; Feedback 2: G(Ri1) does not exist and is greater than the preset threshold Gmax → execute STEP18; STEP10: Mark the abnormal rebound value data; Instruction: Mark the rebound value Ri1 corresponding to the Grubbs criterion abnormal value G(Ri1) greater than the preset threshold value Gmax as abnormal rebound value data, and obtain the number of abnormal rebound value data k1; STEP11: Eliminate abnormal rebound value data Instruction: Delete the abnormal rebound value data from all rebound values ​​Ri1, and set n1 = n1-k1; Feedback: Return to STEP6: STEP12: Calculate the upper data volume of compressive strength; Instructions: By formula Calculate the Grubbs criterion anomaly G(fi2) for each compressive strength fi2; Feedback 1: G(fi2) is greater than the preset threshold Gmax → execute STEP13; Feedback 2: G(fi2) does not exist and is greater than the preset threshold Gmax → execute STEP19; STEP13: Marking of abnormal compressive strength data; Instruction: mark the compressive strength fi2 corresponding to the Grubbs criterion abnormal value G(fi2) greater than the preset threshold Gmax as abnormal compressive strength data, and obtain the number of abnormal compressive strength data k2; STEP 14: Eliminate abnormal compressive strength data Command: Delete abnormal compressive strength data from all compressive strength fi2, and set n2 = n2-k2; Feedback: Return to STEP7: STEP15: Calculate the data volume on the upper side of the carbonization depth; Instructions: By formula Calculate the Grubbs criterion anomaly value G(hi3) for each carbonization depth hi3; Feedback 1: G(hi3) is greater than the preset threshold Gmax → execute STEP16; Feedback 2: G(hi3) does not exist and is greater than the preset threshold Gmax → execute STEP20; STEP16: Carbonization depth abnormal data marking; Instruction: mark the carbonization depth hi3 corresponding to the Grubbs criterion abnormal value G(hi3) greater than the preset threshold value Gmax as carbonization depth abnormal data, and obtain the number k3 of carbonization depth abnormal data; STEP 17: Eliminate abnormal data of carbonization depth Instruction: Delete abnormal carbonization depth data from all carbonization depths hi3, and set n3 = n3-k3; Feedback: Return to STEP8: STEP18: Output the rebound value processing result; Instruction: Generate end signal 1; Feedback: Record all remaining Ri1; STEP19: Output compressive strength processing results; Instruction: Generate end signal 2; Feedback: Record all remaining fi2; STEP20: Output the carbonization depth treatment results; Instruction: Generate end signal three; Feedback: Record all remaining hi3s; STEP21: Data integration and alignment; Instruction: Identify the number symbols in the rebound value data Ri1, the compressive strength data fi2 and the carbonization depth data hi3, that is, the items where i1, i2 and i3 are completely consistent, and mark them as aligned data; Feedback: Integrate the aligned data and assign a new number j to generate a data vector (Rj, fj, hj) = (Ri1, fi2, hi3); where i1 = i2 = i3; where j = 1, 2, 3, ..., m; where m is the total number of data vectors obtained after integrating the aligned data; STEP22: Output and end; Instructions: Output all data vectors (Rj, fj, hj); where j = 1, 2, 3, ..., m; Output the average value μ(R) and variance σ(R) of the rebound value data in the last cycle; Output the average value μ(f) and variance σ(f) of the compressive strength data in the last cycle; Output the average value μ(h) and variance σ(h) of the carbonization depth data in the last cycle; Feedback: End all calculation processes.

4. A concrete strength prediction system based on SVR according to claim 1 or 3, characterized in that: The specific process of Z-score standardization is as follows: Get the mean μ(X) and standard deviation σ(X) of Rj; Get the mean μ(X) and standard deviation σ(X) of Rj; Calculate the mean μ(X) and standard deviation σ(X) of Rj; By formula Perform Z-score standardization to transform Rj, fj and hj into preprocessed data Rj', fj' and hj' with a mean of 0 and a standard deviation of 1; Furthermore, the age tj and design strength grade Cj of the concrete specimen corresponding to each data vector (Rj, fj, hj) are collected; And through the formula Perform Z-score standardization to transform tj and Cj into preprocessed data tj' and Cj' with a mean of 0 and a standard deviation of 1; Wherein μ(t) is the age mean and μ(t) is the age variance; where μ(C) is the design strength grade mean and μ(C) is the design strength grade variance; All Rj', fj', hj', tj' and Cj' are combined into a new data vector (Rj', fj', hj', tj', Cj') and output to the SVR prediction module.

5. The SVR-based concrete strength prediction system according to claim 1, characterized in that: The specific process of establishing the regression equation of concrete rebound surface is as follows: Let power function f = k × R b ×h c Where k is the influence coefficient; b and c are the influence indexes; f is the compressive strength, R is the rebound value; h is the carbonization depth; Substituting the experimental data, the linear regression operation was carried out through the least square method to obtain the specific values ​​of the influence coefficient k and the influence indexes b and c.

6. The SVR-based concrete strength prediction system according to claim 1, characterized in that: The support vector regression model is specifically: It includes an input layer, a kernel function and an output layer; the input layer includes 4 nodes, which are used to input Rj, hj, tj and Cj respectively; the kernel function layer; the kernel function maps the input data to a high-dimensional space and assigns a nonlinear relationship between the input data; the output layer includes 1 output node, which is used to output the predicted compressive strength fj.

7. The SVR-based concrete strength prediction system according to claim 6, characterized in that: The specific process of training the support vector regression model is: Set hyperparameters: regularization parameter C, error tolerance ε, and hyperparameter γ; Setting insensitive loss coefficient Where fj is the actual value, i.e. the measured value of the compressive strength of concrete specimen j; is the predicted value, i.e., the actual output value of the support vector regression model; ε is preset as the tolerance error threshold; Set kernel function: K(x, xw) = exp(-γ×||x1-x2|| 2 ). x and xw are the input values ​​of the kernel function, including Rj, hj, tj and Cj; γ is a hyperparameter. w = 1, 2, 3, ..., N. N is the kernel function node number; Set the optimization objective function: where w is the weight vector; ||w|| 2 is the regularization term of the weight vector; where is the penalty error term for setting the optimization goal; where ξ i and is the slack variable, Indicates the error degree of the jth sample; Where C is the regularization parameter; The input data is mapped to the high-order space through the set kernel function K(xi, xj), and the mapping formula is: where α i and is the support vector; K(x1, x2) is the preset kernel function; b is the bias coefficient; Set constraints: According to the optimization objective function and constraints, the SMO optimization algorithm is used to solve the model parameters and record the solution results. As the final parameters of the trained support vector regression model.

8. The SVR-based concrete strength prediction system according to claim 2, characterized in that: The specific process of compressive strength test, rebound test and carbonization depth determination is as follows: The pressure test machine is used to uniformly load the upper and lower pressure surfaces along the central axis of each concrete test block i at a preset loading speed until the test block is destroyed; the load Fi when the concrete test block i is destroyed is recorded, and the formula is used to calculate the load Fi. Calculate the compressive strength fi of concrete specimen i; where Ai is the compressive area of ​​concrete specimen i; Arrange several test points on each concrete test block i, perform rebound test and carbonation depth measurement on each test point, and obtain the rebound value and carbonation depth of each test point; calculate the average rebound value of all test points to obtain the rebound value Ri of the entire concrete test block i; calculate the average carbonation depth of all test points to obtain the carbonation depth hi of the entire concrete test block i Place the rebound hammer vertically at the test point, making sure that the probe of the instrument is in full contact with the surface; then, press the pressure head of the rebound hammer hard to compress the spring inside and rebound, and record the rebound value displayed by the rebound hammer; After completing the rebound value test and the compressive strength test, use a drilling tool to drill a hole with a preset depth and a preset diameter at each test point on the surface of the fractured test block; clean up the debris and powder in the hole, and apply phenolphthalein alcohol solution to the inner wall of the hole; when an obvious color dividing line appears on the inner wall of the hole, determine that the dividing line is the dividing line between the carbonized area and the non-carbonized area of ​​the concrete; measure the carbonization depth using a carbonization depth measuring instrument.

9. The SVR-based concrete strength prediction system according to claim 2, characterized in that: The specific process of weighted fusion of the two prediction results through the Kalman filter is: Obtain the output result of the first prediction criterion: f1j={f11, f12, ..., f1m}; Obtain the output result of the second prediction criterion: f2j={f21, f22, ..., f2m}; Calculate the mean μ(f1j) and μ(f2j) of f1j and f2j respectively; calculate the variance σ of f1j and f2j respectively 2 (f1j) and σ 2 (f2j); The Kalman filter is used to perform weighted fusion of the two data sets and calculate the optimal estimate The Kalman filter formula is: Where K1 and K2 are Kalman gain coefficients; is the error estimate variance of the previous number j, and its update formula is Record the filtering results, that is, the final output The final predicted value of concrete strength is determined and output.

10. A concrete strength prediction method based on SVR, characterized in that: The following steps are involved: Step 1: Material analysis and data collection; Collect the result data of the rebound test, compressive strength test and carbonation depth value determination of each concrete specimen i, including the rebound value Ri, the compressive strength fi and the carbonation depth hi; Step 2: Data preprocessing and cleaning; All the collected rebound values ​​Ri, compressive strength fi and carbonization depth hi are input into the outlier processing program, and the outlier data are eliminated by the Grubbs criterion to obtain the outlier processing result, that is, the data vector (Rj, fj, hj) Step 3: Data fitting and regression model establishment; S301, regression model data fitting; According to the data vector (Rj, fj, hj), the relationship between compressive strength, rebound value and carbonation depth is fitted through mathematical functions to obtain the concrete rebound surface regression equation f = k × R b ×h c Parameters k, b and c in; The concrete rebound surface with known parameters is regressed into the equation f = k × R b ×h c Recorded as the first prediction criterion; S302. Constructing a support vector regression model Obtain the concrete age tj and design strength grade Cj corresponding to the concrete specimen number j in the data vector (Rj, fj, hj), and construct a new data vector (tj, Cj, Rj, fj, hj); Train the support vector regression model based on the new data vector (tj, Cj, Rj, fj, hj); The support vector regression model includes an input layer, a kernel function and an output layer; The input layer includes 4 nodes, which are used to input Rj, hj, tj and Cj respectively; the kernel function layer; the kernel function maps the input data to a high-dimensional space and assigns nonlinear relationships between the input data; The output layer includes 1 output node, which is used to output the predicted compressive strength fj; Set hyperparameters: regularization parameter C, error tolerance ε, and hyperparameter γ; Set the insensitive loss coefficient, kernel function, optimization objective function and constraints of the support vector regression model; Substitute the new data vector (tj, Cj, Rj, fj, hj), and use the SMO optimization algorithm to solve the model parameters and record the solution results in combination with the optimization objective function and constraints. As the final parameters of the trained support vector regression model; The trained support vector regression model is used as the second prediction criterion; Step 4: Prediction of concrete strength; Take the concrete used to predict the compressive strength and prepare several concrete test blocks; Obtaining basic data of the several concrete test blocks, including: rebound value, carbonation depth, age and design strength grade; Inputting the rebound values ​​and carbonation depths of the plurality of concrete test blocks into a first prediction criterion to obtain an output result of the first prediction criterion; Inputting the rebound value, carbonation depth, age and design strength grade of the plurality of concrete test blocks into a second prediction criterion to obtain an output result of the second prediction criterion; Step 5: Prediction data fusion; The Kalman filter is used to fuse the output result of the first prediction criterion and the output result of the second prediction criterion to obtain the final concrete strength prediction result.

Images