A prediction method for damage quantification index in the process of freeze-thaw and sulfate erosion of concrete

Abstract

The present invention belongs to the technical field of concrete. In order to solve the lack of comprehensive, dynamic and in-depth evaluation means in the existing methods, a prediction method for the damage quantification index in the process of freeze-thaw and sulfate erosion of concrete is proposed, including: collecting the damage quantification index; constructing a training set and a test set of feature data for the damage quantification index, as well as a training set and a test set of label data; inputting the training set of feature data of the damage quantification index into the CNN-LSTM-Attention model to obtain the predicted value of the damage quantification index, and using the Adam optimization algorithm to update the biases and weights of the output layer of the CNN-LSTM-Attention model; inputting the test set of feature data of the damage quantification index into the optimized CNN-LSTM-Attention model to output the predicted value of the test set of feature data. The method of the present invention realizes fast and accurate prediction.

Description

Technical Field

[0001] The present invention belongs to the field of research on the durability of concrete lining structures, and particularly relates to a method for predicting damage quantification indexes in the process of concrete freeze-thaw-sulfate erosion. Background Art

[0002] Saline soils are widely distributed in China, and the western region is the main concentrated distribution area of saline soils, and the saline soils are rich in sulfates. The environment in this area also has the characteristics of less precipitation and large evaporation, and concrete structures are easily affected by wet-dry cycles, resulting in serious deterioration of mechanical properties. In addition, the cold regions in China have a wide coverage area, and the seasonal cold regions account for a relatively high proportion. Due to special climatic conditions, the climate in cold regions is extremely cold and the seasonal temperature difference is large, and concrete structures are easily damaged by freeze-thaw. In the environment of sulfate saline soil in cold regions in the west, concrete structures are not only affected by freeze-thaw, but also affected by sulfate erosion, and the deterioration of concrete performance is more rapid. Therefore, the damage quantification prediction of concrete has become an important and complex problem.

[0003] At present, some research methods for the durability of concrete materials mainly focus on corrosion tests and partial index monitoring, lacking comprehensive, dynamic and in-depth evaluation methods. In addition, the current concrete damage quantification prediction methods are mostly based on empirical formulas or mathematical models, with many factors to be considered and many parameters required, which is not conducive to realizing rapid and accurate damage prediction. Summary of the Invention

[0004] In view of the above problems, the present invention provides a method for predicting damage quantification indexes in the process of concrete freeze-thaw-sulfate erosion, aiming to comprehensively and accurately evaluate the performance changes of concrete in actual engineering through freeze-thaw sulfate wet-dry cycle tests, combining neural network to monitor multiple damage quantification indexes (mass loss rate, relative compressive strength, relative dynamic elastic modulus), and integrating a convolutional neural network and a long short-term memory network (LSTM) with an attention mechanism (Attention).

[0005] A method for predicting damage quantification indexes in the process of concrete freeze-thaw-sulfate erosion, comprising:

[0006] S1: During the alternate test of freeze-thaw-sulfate wet-dry cycle of concrete specimens, collect the damage quantification indexes of the concrete specimens during the test process;

[0007] Among them, the damage quantification indexes include mass loss rate, relative compressive strength and relative dynamic elastic modulus;

[0008] S2: Perform data preprocessing on the damage quantification indicators in S1 to construct a training set and a test set of feature data for the damage quantification indicators, as well as a training set and a test set of label data for the damage quantification indicators;

[0009] S3: Input the training set of feature data of the damage quantification indicators in S2 into the CNN-LSTM-Attention model to obtain the predicted values based on the training set of feature data of the damage quantification indicators; Use the Adam optimization algorithm to update the biases and weights of the output layer of the CNN-LSTM-Attention model in combination with the predicted values and the training set of label data of the damage quantification indicators to optimize the CNN-LSTM-Attention model;

[0010] Among them, the first layer of the CNN-LSTM-Attention model is a convolutional neural network layer, the second layer is a long short-term memory neural network layer, the third layer is an attention layer, and the fourth layer is an output layer;

[0011] S4: Input the test set of feature data of the damage quantification indicators into the optimized CNN-LSTM-Attention model. The output layer of the CNN-LSTM-Attention model outputs the predicted values based on the test set of feature data of the damage quantification indicators to obtain the prediction results of the damage quantification indicators in the concrete freeze-thaw-sulfate erosion process; Use the test set of label data of the damage quantification indicators to verify the convergence of the CNN-LSTM-Attention model through simulation experiments.

[0012] Preferably, the process of performing data preprocessing on the damage quantification indicators in S2 includes:

[0013] S2.1: Calculate the mean and standard deviation of the damage quantification indicators, and remove the outliers in the damage quantification indicators according to the calculated mean and standard deviation to obtain the damage quantification indicators with missing data points;

[0014] S2.2: Use the interpolation method to fill in the missing data points of the damage quantification indicators in S2.1 to obtain the filled damage quantification indicators, and expand the filled damage quantification indicators to obtain the expanded damage quantification indicators;

[0015] S2.3: Perform normalization processing on the expanded damage quantification indicators in S2.2 to obtain the set of normalized damage quantification indicators;

[0016] S2.4: Divide the set of damage quantification indicators into a training set and a test set of feature data of the damage quantification indicators.

[0017] Preferably, S2.3 includes:

[0018] S2.3.1: Normalize the mass loss rate in the damage quantification index after expanding S2.2 to obtain the normalized mass loss rate. The calculation formula is:

[0019] ;

[0020] where, is the mass loss rate after expansion, is the minimum value in the mass loss rate after expansion, is the maximum value in the mass loss rate after expansion, is the j th normalized mass loss rate;

[0021] Then the set of normalized mass loss rates is :

[0022] ;

[0023] S2.3.2: Normalize the relative compressive strength in the damage quantification index after expanding S2.2 to obtain the normalized relative compressive strength. The calculation formula is:

[0024] ;

[0025] where, is the relative compressive strength after expansion, is the minimum value in the relative compressive strength after expansion, is the maximum value in the relative compressive strength after expansion, is the j th normalized relative compressive strength;

[0026] Then the set of normalized relative compressive strengths is :

[0027] ;

[0028] S2.3.3: Normalize the relative dynamic elastic modulus in the damage quantification index after expanding S2.2 to obtain the normalized relative dynamic elastic modulus. The calculation formula is:

[0029] ;

[0030] where, is the relative dynamic elastic modulus after expansion, is the minimum value in the relative dynamic elastic modulus after expansion, is the maximum value in the relative dynamic elastic modulus after expansion, is the jThe normalized relative dynamic elastic modulus;

[0031] Then the set of normalized relative dynamic elastic moduli is obtained :

[0032] .

[0033] Preferably, S3 includes:

[0034] S3.1: Input the characteristic data training set of the damage quantification index into the convolutional neural network layer of the CNN-LSTM-Attention model for convolution operation to obtain the convolved value, and then ReLU Use the activation function to perform non-linear transformation on the convolved value to obtain the non-linearly transformed value, and perform pooling operation on the non-linearly transformed value to obtain the output result of the convolutional neural network layer;

[0035] S3.2: Input the output result of the convolutional neural network layer in S3.1 into the long short-term memory neural network layer to obtain the output result of the long short-term memory neural network layer; among them, the long short-term memory neural network layer includes an input gate, a forget gate, an output gate, and a cell state;

[0036] S3.3: Input the output result of the long short-term memory neural network layer in S3.2 into the attention layer to obtain the output result of the attention layer;

[0037] S3.4: Input the output result of the attention layer in S3.3 into the output layer, and the output layer outputs the predicted value based on the characteristic data training set of the damage quantification index.

[0038] Preferably, the specific calculation process of the pooling operation in S3.1 is:

[0039] ;

[0040] ;

[0041] ;

[0042] Among them, is the value after pooling operation of the s th mass loss rate, is the value after pooling operation of the s th relative compressive strength, is the value after pooling operation of the s th relative dynamic elastic modulus; is the value after non-linear transformation of the i th mass loss rate, is the value after non-linear transformation of the i th relative compressive strength, is the value after non - linear conversion of the i th relative dynamic elastic modulus, is the value after non - linear conversion of the i+ 1st mass loss rate, is the value after non - linear conversion of the i +(1)th relative compressive strength, is the value after non - linear conversion of the i +(1)th relative dynamic elastic modulus, The function represents taking the maximum value.

[0043] Preferably, in S3.2, the output of the long - short - term memory neural network layer is:

[0044] ;

[0045] ;

[0046] ;

[0047] Among them, is the hyperbolic tangent function, is t the mass loss rate output by the time - step short - term memory neural network layer; is t the relative compressive strength output by the time - step short - term memory neural network layer; is t the relative dynamic elastic modulus output by the time - step short - term memory neural network layer, is t the value of the mass loss rate after passing through the input - output gate at the time step, is t the value of the relative compressive strength after passing through the input - output gate at the time step, is t the value of the relative dynamic elastic modulus after passing through the input - output gate at the time step.

[0048] Preferably, S3.3 includes:

[0049] S3.3.1: Use the weight matrix to perform weighted calculation on the output result of the long - short - term memory neural network layer to obtain the attention score of the damage quantification index:

[0050] ;

[0051] ;

[0052] ;

[0053] Among them, is the hyperbolic tangent function, is the attention score of the mass loss rate at each time step, is the attention score of the relative compressive strength at each time step, is the attention score of the relative dynamic elastic modulus at each time step;

[0054] S3.3.2: Normalize the attention scores of the damage quantification indicators in S3.3.1 through the Softmax function to obtain the attention weights of the normalized damage quantification indicators:

[0055] ;

[0056] ;

[0057] ;

[0058] where, is t the attention weight of the mass loss rate at each time step, is t the attention weight of the relative compressive strength at each time step, is t the attention weight of the relative dynamic elastic modulus at each time step, is the exponential function with e as the base, is the total number of time steps, is the current time step;

[0059] S3.3.3: Based on the attention weights in S3.3.2, perform a weighted sum of the damage quantification indicators output by the long short-term memory neural network layer to obtain the output result of the attention layer:

[0060] ;

[0061] ;

[0062] ;

[0063] where, is t the mass loss rate output by the attention layer at each time step, is t the relative compressive strength output by the attention layer at each time step, is t the relative dynamic elastic modulus output by the attention layer at each time step; is the total number of time steps, is the current time step.

[0064] Preferably, the output result of the attention layer in S3.3 is input into the output layer, and the output layer outputs the predicted value of the feature data training set based on the damage quantification index:

[0065] ;

[0066] ;

[0067] ;

[0068] wherein, is the bias of the output layer, is the weight of the output layer, is the predicted value of the quality loss rate, is the predicted value of the relative compressive strength, the predicted value of the relative dynamic elastic modulus.

[0069] Preferably, the Adam optimization algorithm is adopted to update the bias and weight of the output layer of the CNN-LSTM-Attention model, which specifically includes:

[0070] (1): Determine the first-order moment estimate of the current traversal number and the second-order moment estimate of the current traversal number:

[0071] ;

[0072] ;

[0073] wherein, is the gradient of the current traversal number, is the first-order moment estimate of the current traversal number, is the first-order moment estimate of the previous traversal number, is the second-order moment estimate of the current traversal number, is the exponential decay rate of the momentum term, is the exponential decay rate of the variance term;

[0074] (2): On the basis of (1), correct the deviation to obtain the first-order moment estimate and the second-order moment estimate after eliminating the deviation of the current traversal number:

[0075] ;

[0076] ;

[0077] wherein, is the first-order moment estimate after eliminating the deviation of the current traversal number, is the second-order moment estimate after eliminating the deviation of the current traversal number; is thex power and is the exponential decay rate of the variance term x power;

[0078] (3): Based on (2), update the weights and biases of the output layer of the CNN-LSTM-Attention model, thereby realizing the optimization of the CNN-LSTM-Attention model:

[0079] ;

[0080] ;

[0081] wherein, is the weight of the output layer during the x+ 1st traversal, is the neural network bias of the output layer during the th traversal, is the neural network bias of the output layer during the x th traversal, is the learning rate, is a constant;

[0082] (4): Calculate the loss function of the CNN-LSTM-Attention model MSE , update the weights and biases through the number of traversals, and find the minimized loss function:

[0083] ;

[0084] ;

[0085] ;

[0086] wherein, MSE 1 is the loss function of the CNN-LSTM-Attention model based on the mass loss rate, MSE 2 is the loss function of the CNN-LSTM-Attention model based on the relative compressive strength, MSE 3 is the loss function of the CNN-LSTM-Attention model based on the relative dynamic elastic modulus, is the label data training set of the mass loss rate, is the label data training set of the relative compressive strength, is the labeled data training set of the relative dynamic elastic modulus, the labeled data training set of the mass loss rate, the labeled data training set of the relative compressive strength, and the labeled data training set of the relative dynamic elastic modulus all belong to the labeled data training set of the damage quantification index. n is the number of elements in the feature data training set.

[0087] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0088] 1. A method for predicting the damage quantification index during the freeze-thaw-sulfate erosion process of concrete provided by the present invention can predict the damage quantification index of the next freeze-thaw-sulfate erosion process through this method by detecting the dynamic concrete mass loss rate, relative compressive strength, and relative dynamic elastic modulus values, without the need to conduct a large number of experiments to obtain corresponding parameters. In the process of predicting the concrete damage quantification index, the present invention does not fit the mass loss rate, relative compressive strength, and relative dynamic elastic modulus values or adopt mathematical assumptions, and has high reliability. The method of the present invention can perform damage quantification prediction on various concrete service environments, including both the freeze-thaw sulfate wet-dry cycling environment in this case and the corrosion environment caused by climate change, and has the characteristics of a wide application range. Therefore, the method of the present invention comprehensively solves a series of problems existing in the current prediction technology, such as narrow application range, complex calculation, and high prediction cost.

[0089] 2. The method for predicting the damage quantification index during the freeze-thaw-sulfate erosion process of the concrete lining structure provided by the present invention integrates CNN, Attention, and LSTM, and the established prediction model combines the characteristics of different neural network algorithms. It avoids the problems of existing methods based on empirical formulas or mathematical models, which consider many factors and require many parameters and are not conducive to achieving fast and accurate damage prediction, and provides a fast and effective method for realizing the prediction of concrete damage quantification index. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention, and do not constitute a limitation to the present invention.

[0091] In the drawings:

[0092] Att Figure 1 : The flowchart of the method of the present invention.

[0093] Att Figure 2 : The flowchart of the freeze-thaw-sulfate wet-dry cycling alternating test regime.

[0094] Att Figure 3:Prediction diagrams of the mass loss rate of concrete under freeze-thaw sulfate wet-dry cycles on three neural network models: (a) LSTM, (b) CNN-LSTM, (c) CNN-LSTM-Attention.

[0095] Appendix Figure 4 :Prediction diagrams of the relative compressive strength of concrete under freeze-thaw sulfate wet-dry cycles on three neural network models: (a) LSTM, (b) CNN-LSTM, (c) CNN-LSTM-Attention.

[0096] Appendix Figure 5 :Prediction diagrams of the relative dynamic elastic modulus of concrete under freeze-thaw sulfate wet-dry cycles on three neural network models: (a) LSTM, (b) CNN-LSTM, (c) CNN-LSTM-Attention. Specific implementation manners

[0097] The following is a description of the preferred embodiments of the present invention in conjunction with the appendix Figure 1 - Appendix Figure 5 The preferred embodiments of the present invention will be described. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0098] Embodiment:

[0099] As Figure 1 , a prediction method for damage quantification indexes in the process of concrete freeze-thaw-sulfate erosion includes the following steps:

[0100] S1: Determine the temperature acceleration coefficient, the erosion solution concentration acceleration coefficient, and introduce the freeze-thaw deterioration ratio coefficient; determine the number of sulfate wet-dry cycles according to the temperature acceleration coefficient and the erosion solution concentration acceleration coefficient, and determine the number of freeze-thaw cycles according to the freeze-thaw deterioration ratio coefficient; based on the number of sulfate wet-dry cycles and the number of freeze-thaw cycles, complete the alternate test of freeze-thaw-sulfate wet-dry cycles of concrete specimens, and collect the damage quantification indexes of concrete specimens during the test: mass loss rate, relative compressive strength, and relative dynamic elastic modulus.

[0101] S1.1: Determine the temperature acceleration coefficient and the erosion solution concentration acceleration coefficient, and introduce the freeze-thaw deterioration ratio coefficient.

[0102] S1.1.1: The expression for the average temperature during immersion and drying of concrete specimens in the sulfate wet-dry cycle test is:

[0103] ;

[0104] Where: T 均 is the average test temperature, T 烘箱is the oven temperature, t 1 is the duration of oven drying, T 1 is the annual average temperature of the region under the Celsius scale, t 2 is the sulfate immersion time, t 3 is the air-drying time;

[0105] The expression for the temperature acceleration coefficient is:

[0106] ;

[0107] In the formula: K a is the temperature acceleration coefficient; E / R is the activation energy, taken as 14242; T 1 is the annual average temperature of the region under the Celsius scale.

[0108] S1.1.2: The erosion solution concentration acceleration coefficient is:

[0109] ;

[0110] In the formula: is the erosion solution concentration acceleration coefficient; is the sulfate concentration in the erosion solution; is the average concentration of sulfate in the actual environment; is the erosion ion concentration used in the accelerated test; is the erosion ion concentration in the actual service environment of the concrete.

[0111] S1.1.3: Introduce the freeze-thaw deterioration ratio coefficient :

[0112] The range of the ratio between the number of test freeze-thaw cycles and the number of freeze-thaw cycles under natural conditions is 1:10 to 1:15, and the value is 1:12, that is, one test freeze-thaw cycle is equivalent to 12 freeze-thaw cycles under natural conditions. Therefore, the freeze-thaw deterioration ratio coefficient takes 12.

[0113] S1.2: Determine the number of sulfate wet-dry cycles according to the temperature acceleration coefficient and the erosion solution concentration acceleration coefficient, and determine the number of freeze-thaw cycles according to the freeze-thaw deterioration ratio coefficient.

[0114] Adopt the test regime of freeze-thaw - sulfate wet-dry cycle to simulate 1 year of the actual environment. The freeze-thaw cycle considers the freeze-thaw deterioration acceleration, and the sulfate wet-dry cycle considers the temperature acceleration and the erosion solution concentration acceleration.

[0115] The number of freeze-thaw cycle tests during the test:

[0116] ;

[0117] Number of days of sulfate wet-dry cycles during the test:

[0118] ;

[0119] In this embodiment, Haidong City, Qinghai Province is taken as the research object, and the annual average temperature in this area is 7°C. Referring to the SO4 content in the saline soil at the test site, the actual environmental erosion ion concentration is taken as 2%. 2- Content, the actual environmental erosion ion concentration is taken as 2%.

[0120] The oven temperature is 65°C, the duration of oven drying is 6h, the annual average temperature in the area under the Celsius scale is 7°C, the sulfate immersion time is 16h, and the air-drying time is 0.5h. Then:

[0121] The average test temperature is:

[0122] .

[0123] Temperature acceleration coefficient:

[0124] ;

[0125] Erosion solution concentration acceleration coefficient:

[0126] ;

[0127] The number of annual freeze-thaw cycles in Haidong City is 100 - 130 times. In this embodiment, the average value of 115 times is taken, and it is considered that there is one freeze-thaw cycle per day and night. Therefore, the number of days of freeze-thaw in a year is 115 days. The number of days of sulfate wet-dry cycles = the number of days in a year - the number of days of freeze-thaw cycles = 365 - 115 = 250 days.

[0128] Number of freeze-thaw cycles: 115÷12 = 9.6 times; Number of days of sulfate wet-dry cycles: 250÷18.1÷2.5 = 5.5d.

[0129] Prepare concrete specimens. Considering the actual adverse environmental conditions and the requirements for the surface flatness of the specimens in the subsequent SHPB impact test, it is finally determined that one freeze-thaw - sulfate wet-dry cycle alternation test includes 8 freeze-thaw cycles and 6 sulfate wet-dry cycles. The number of freeze-thaw - sulfate wet-dry cycles is recorded as N , and a total of 7 times are carried out. The test regime is shown in Figure 2 .

[0130] S1.3: Based on the number of days of sulfate wet-dry cycles and the number of freeze-thaw cycles, complete the freeze-thaw - sulfate wet-dry cycle alternation test, and collect the damage quantification indexes during the test: mass loss rate, relative compressive strength, and relative dynamic elastic modulus.

[0131] S1.3.1: On the basis of Step 1.2, determine the freeze-thaw - sulfate wet-dry cycle alternation test regime.

[0132] Before the start of the first freeze-thaw cycle test, soak all specimens in a 5% sodium sulfate solution for 4 days to make them in a fully saturated water state. The freeze-thaw cycle test and the sulfate wet-dry cycle test are carried out alternately, and before the start of the n ( n ≥2)th freeze-thaw cycle test, soak the concrete specimens in a sodium sulfate solution with a mass fraction of 5% for 1 day.

[0133] After the soaking is completed, take out the concrete specimens and place them in the freeze-thaw box in turn, and place wooden strips under each concrete specimen to prevent the bottom of the concrete specimen from being affected by the freezing of the residual water at the bottom of the freeze-thaw box. Keep a certain distance between adjacent concrete specimens to avoid sticking together, so that each surface is evenly affected by the temperature change of the freeze-thaw box.

[0134] One freeze-thaw cycle period includes a spraying section, a cooling section, a low-temperature constant-temperature section, a heating stage, and a high-temperature constant-temperature section. The duration of each stage is: 5 minutes, 2 hours, 2 hours, 0.5 hours, 1.5 hours, totaling 6 hours and 5 minutes. In the spraying section, place the concrete specimens under the nozzles of the freeze-thaw box for continuous water replenishment for 5 minutes. The temperature of the low-temperature constant-temperature section of the freeze-thaw box is set at -20°C to ensure that the concrete specimens can complete the freezing process at this temperature. The high-temperature constant-temperature section is set at +25°C. After the end of the high-temperature constant-temperature section, the freeze-thaw box automatically sprays three times, each time lasting 1 minute, to keep the concrete specimens in a water-retaining state.

[0135] The operation method of the sulfate wet-dry cycle test is: soak the concrete specimens in a 5% sulfate solution. The time from when the concrete specimens start to be put into the solution to the end of the soaking process is 16 hours. After the soaking process is completed, take out the concrete specimens and place them at the designated position to air-dry for 1 hour. After the air-drying process is completed, put the concrete specimens into a blast drying oven, turn on the oven switch, and after the temperature rises to 65°C, maintain the temperature at about 65°C and dry for 6 hours. After drying is completed, cool the concrete specimens for 1 hour, and the surface temperature of the concrete specimens needs to be cooled to room temperature, and then continue the above-mentioned freeze-thaw cycle test and sulfate wet-dry cycle test.

[0136] Since the ettringite in the sulfate erosion products in the concrete specimens may decompose at 70°C, which may change the erosion mechanism. Therefore, it is necessary to ensure that the drying temperature is 65°C ± 5°C, and the drying temperature in this sulfate wet-dry cycle test is 65°C.

[0137] Carry out the air-freezing and air-thawing method 8 times (about 2 days) according to the above freeze-thaw cycle regime; after the freeze-thaw cycle is completed, carry out the sulfate wet-dry cycle 6 times (about 6 days) according to the above sulfate wet-dry cycle steps.

[0138] S1.3.2: Measure the mass of the concrete specimens in the freeze-thaw-sulfate wet-dry cycle test using an electronic balance, and obtain the mass loss rate. :

[0139] ;

[0140] Where: is the mass loss rate of the concrete specimens after N cycles of freeze-thaw-sulfate wet-dry cycle; is the initial mass of the concrete specimens; is the mass of the specimens after N cycles ( N = 1, 2, 3, 4, 5, 6, 7) of freeze-thaw-sulfate wet-dry cycle.

[0141] S1.3.3: Calculate the relative compressive strength of the concrete specimens:

[0142] ;

[0143] ;

[0144] Where, is the relative compressive strength, is the pressure applied by the press to the concrete specimens; is the compressive area of the concrete specimens; is the measured compressive strength of the concrete specimens after N cycles of freeze-thaw-sulfate wet-dry cycle; is the initial measured compressive strength of the concrete specimens.

[0145] S1.3.4: Measure the initial ultrasonic velocity of the concrete specimens and the ultrasonic velocity of the specimens after N cycles ( N = 1, 2, 3, 4, 5, 6, 7) of freeze-thaw-sulfate wet-dry cycle, and then obtain the relative dynamic elastic modulus of the concrete:

[0146] ;

[0147] Where: is the relative dynamic elastic modulus of the concrete specimens after N cycles of freeze-thaw-sulfate wet-dry cycle; is the dynamic elastic modulus of the concrete specimens after N cycles of freeze-thaw-sulfate wet-dry cycle; is the dynamic elastic modulus before the freeze-thaw-sulfate cycle V 0 is the initial ultrasonic velocity of the concrete specimens; is after Ntimes ( N = 1, 2, 3, 4, 5, 6, 7) the ultrasonic velocity of the specimens after freeze-thaw-sulfate wet-dry cycles.

[0148] S2: Preprocess the damage quantification indexes in S1 to construct the feature data training set and feature data test set of the damage quantification indexes, as well as the label data training set and label data test set of the damage quantification indexes. Specifically include:

[0149] S2.1: For each freeze-thaw-sulfate wet-dry cycle test, measure the mass loss rate, relative compressive strength, and relative dynamic elastic modulus of 6 specimens respectively, calculate the mean and standard deviation of the mass loss rate, relative compressive strength, and relative dynamic elastic modulus, and remove the outliers in the damage quantification indexes to obtain the damage quantification indexes with missing data points.

[0150] S2.1.1: Calculate the mean of the mass loss rate, relative compressive strength, and relative dynamic elastic modulus respectively:

[0151] ;

[0152] ;

[0153] ;

[0154] Among them, represents the mean of the mass loss rate, represents the mean of the relative compressive strength, represents the mean of the relative dynamic elastic modulus; represents the i th mass loss rate, represents the i th relative compressive strength; represents the i th relative dynamic elastic modulus;

[0155] S2.1.2: Calculate the standard deviation of the mass loss rate, relative compressive strength, and relative dynamic elastic modulus respectively:

[0156] ;

[0157] ;

[0158] ;

[0159] Among them, represents the standard deviation of the mass loss rate, represents the standard deviation of the relative compressive strength, represents the standard deviation of the relative dynamic elastic modulus;

[0160] S2.1.3: According to principle, determine that the range of the mass loss rate is , the range of the relative compressive strength is , the range of the relative dynamic elastic modulus is . Remove the outliers outside the range to obtain the normal damage quantification index, and calculate the arithmetic mean of the normal damage quantification index to obtain the average mass loss rate , the average relative compressive strength , the average relative dynamic elastic modulus ;

[0161] S2.1.4: On the basis of S2.1.3, the set of arithmetic means of the damage quantification indexes after seven freeze-thaw-sulfate wet-dry cycles is:

[0162] Mass loss rate: , relative compressive strength: , relative dynamic elastic modulus: ;

[0163] S2.2: Use the interpolation method to fill the missing data points of the damage quantification index in S2.1 to obtain the filled damage quantification index, and expand the filled damage quantification index to obtain the expanded damage quantification index;

[0164] S2.2.1: Use the interpolation method to fill the missing data points and expand the data for the mass loss rate:

[0165] ;

[0166] Among them, represents the number of freeze-thaw-sulfate wet-dry cycles, , is the mass loss rate after interpolation filling, and are the known mass loss rates at and , is the number of cycles corresponding to the mass loss rate to be estimated, and are both the number of freeze-thaw-sulfate wet-dry cycles corresponding to the known mass loss rates.

[0167] S2.2.2: Use the interpolation method to fill the missing data points for the relative compressive strength:

[0168] ;

[0169] Among them, represents the number of freeze-thaw-sulfate wet-dry cycles, , is the relative compressive strength after interpolation filling, and are the known relative compressive strengths at and respectively. The relative compressive strength to be estimated corresponds to the number of cycles ; and are the number of freeze-thaw-sulfate wet-dry cycles corresponding to the known relative compressive strengths respectively.

[0170] S2.2.3: Interpolate to fill the missing data points for the relative dynamic elastic modulus:

[0171] ;

[0172] wherein, represents the number of freeze-thaw-sulfate wet-dry cycles, , is the relative dynamic elastic modulus after interpolation filling, and are the known relative dynamic elastic moduli at and respectively. The relative dynamic elastic modulus to be estimated corresponds to the number of cycles ; and are the number of freeze-thaw-sulfate wet-dry cycles corresponding to the known relative dynamic elastic moduli respectively.

[0173] S2.2.4: After the interpolation of the damage quantification index data is completed, expand it (using interpolation method) to obtain the final set of interpolated mass loss rates as , relative compressive strength , and the set of relative dynamic elastic moduli as ; represents the j th mass loss rate, represents the j th relative compressive strength, represents the j th relative dynamic elastic modulus. In this application, takes 70.

[0174] S2.3: Normalize and summarize the damage quantification indexes expanded in S2.2 to obtain the set of normalized damage quantification indexes;

[0175] S2.3.1: Normalize the mass loss rate in the damage quantification indexes expanded in S2.2 to obtain the normalized mass loss rate, and its calculation formula is:

[0176] ;

[0177] Among them, is the expanded mass loss rate, is the minimum value among the expanded mass loss rates, is the maximum value among the expanded mass loss rates, is the j th normalized mass loss rate;

[0178] Then the set of normalized mass loss rates is :

[0179] ;

[0180] S2.3.2: Normalize the relative compressive strength in the damage quantification index expanded in S2.2 to obtain the normalized relative compressive strength, and its calculation formula is:

[0181] ;

[0182] Among them, is the expanded relative compressive strength, is the minimum value among the expanded relative compressive strengths, is the maximum value among the expanded relative compressive strengths, is the j th normalized relative compressive strength;

[0183] Then the set of normalized relative compressive strengths is :

[0184] ;

[0185] S2.3.3: Normalize the relative dynamic elastic modulus in the damage quantification index expanded in S2.2 to obtain the normalized relative dynamic elastic modulus, and its calculation formula is:

[0186] ;

[0187] Among them, is the expanded relative dynamic elastic modulus, is the minimum value among the expanded relative dynamic elastic moduli, is the maximum value among the expanded relative dynamic elastic moduli, is the j th normalized relative dynamic elastic modulus;

[0188] Then the set of normalized relative dynamic elastic moduli is :

[0189] 。

[0190] S2.4: Divide the set of damage quantification indicators normalized in S2.3 into a training set and a test set.

[0191] S2.4.1: Divide the normalized damage quantification indicators into segments of five each, with a fixed length;

[0192] Segments of mass loss rate are:

[0193] ; where n is a positive integer, , is the n th mass loss rate;

[0194] Then the mass loss rate feature data set is:

[0195] ;

[0196] Segments of relative compressive strength are: , where n is a positive integer , is the n th normalized relative compressive strength;

[0197] Then the relative compressive strength feature data set F is:

[0198] ;

[0199] Segments of relative dynamic elastic modulus are:

[0200] ;

[0201] where n is a positive integer, ; is the n th normalized relative elastic modulus;

[0202] Then the relative dynamic elastic modulus feature data set G is:

[0203] ;

[0204] S2.4.2: Use the next value of each damage quantification indicator segment in S2.4.1 as a label to construct a label data set for the damage quantification indicators;

[0205] Mass loss rate label data set :

[0206] ;

[0207] Among them, , represents the prediction label of the mass loss rate;

[0208] Relative compressive strength label dataset :

[0209] ;

[0210] Among them, , represents the prediction label of the relative compressive strength;

[0211] Relative dynamic elastic modulus label dataset :

[0212] ;

[0213] Among them, , represents the prediction label of the relative dynamic elastic modulus;

[0214] Among them, n is a positive integer, .

[0215] S2.4.3: Divide the feature dataset of the damage quantification index V, F, G into a feature data training set and a feature data test set according to 8:2, and divide the label dataset of the damage quantification index , , into a label data training set and a label data test set according to 8:2 respectively;

[0216] The feature data training set of the mass loss rate is denoted as , the feature data training set of the relative compressive strength is denoted as , and the feature data training set of the relative dynamic elastic modulus is denoted as ;

[0217] The feature data test set of the mass loss rate is denoted as , the feature data test set of the relative compressive strength is denoted as , and the feature data test set of the relative dynamic elastic modulus is denoted as ;

[0218] The label data training set of the mass loss rate is denoted as , and the label data training set of the relative compressive strength is denoted as , the labeled data training set of the relative dynamic elastic modulus is expressed as ;

[0219] The labeled data test set of the mass loss rate is expressed as , the labeled data test set of the relative compressive strength is expressed as , the labeled data test set of the relative dynamic elastic modulus is expressed as .

[0220] S3: Input the feature data training set of the S2 damage quantification index into the CNN-LSTM-Attention model to obtain the predicted values of the feature data training set based on the damage quantification index; use the Adam optimization algorithm to update the biases and weights of the output layer of the CNN-LSTM-Attention model by combining the predicted values and the labeled data training set of the damage quantification index to optimize the CNN-LSTM-Attention model;

[0221] Among them, the first layer of the CNN-LSTM-Attention model is the convolutional neural network layer, the second layer is the long short-term memory neural network layer, the third layer is the attention layer, and the fourth layer is the output layer.

[0222] S3.1: Input the feature data training set of the damage quantification index into the convolutional neural network layer of the CNN-LSTM-Attention model for convolution operation to obtain the convolved values, and then perform non-linear transformation on the convolved values through ReLU the activation function to obtain the non-linearly transformed values, and perform pooling operation on the non-linearly transformed values to obtain the output result of the convolutional neural network layer;

[0223] S3.1.1: The feature data training of the damage quantification index V 0.8 , F 0.8 and G 0.8 are input into the convolutional neural network layer for convolution operation:

[0224] ;

[0225] ;

[0226] ;

[0227] Among them, is the value after convolution of the i th mass loss rate, is the value after convolution of the i th relative compressive strength, is thei The value after convolution of the relative dynamic elastic modulus, is the weight of the convolution kernel, b is the bias of the convolution kernel, k is the size of the convolution kernel, j is from 1 all the way to k , is V 0.8 , F 0.8 , G 0.8 the th element in respectively are the th data in;

[0228] S3.1.2: After the convolution operation is completed, the values after convolution are non-linearly transformed through ReLU the activation function:

[0229] ;

[0230] ;

[0231] ;

[0232] Among them, is the value after non-linear transformation of the i th mass loss rate, is the value after non-linear transformation of the i th relative compressive strength, is the value after non-linear transformation of the i th relative dynamic elastic modulus, The function represents taking the maximum value;

[0233] S3.1.3: Perform pooling operation on the values of the mass loss rate after non-linear transformation in S3.1.2. The specific calculation process of the pooling operation is:

[0234] ;

[0235] ;

[0236] ;

[0237] Among them, is the value after pooling operation of the s th mass loss rate, is the value after pooling operation of the s th relative compressive strength, is the sThe value after a relative dynamic elastic modulus pooling operation; is the i value after non - linear conversion of the mass loss rate, is the i value after non - linear conversion of the relative compressive strength, is the i value after non - linear conversion of the relative dynamic elastic modulus, is the i+ value after non - linear conversion of the 1st mass loss rate, is the i +(1) value after non - linear conversion of the relative compressive strength, is the i +(1) value after non - linear conversion of the relative dynamic elastic modulus, The function represents taking the maximum value.

[0238] S3.2: Input the output result of the convolutional neural network layer in S3.1 into the long short - term memory neural network layer to obtain the output result of the long short - term memory neural network layer; among them, the long short - term memory neural network layer includes an input gate, a forget gate, an output gate, and a cell state.

[0239] S3.2.1: Input the output result of the convolutional neural network layer in S3.1 into the forget gate of the long short - term memory neural network layer:

[0240] ;

[0241] ;

[0242] ;

[0243] is t the value after the mass loss rate at the time step is input into the forget gate, is t the value after the relative compressive strength at the time step is input into the forget gate, is t the value after the relative dynamic elastic modulus at the time step is input into the forget gate, is the Sigmoid function, is the weight in the forget gate, are respectively the mass loss rate, the compressive strength, and the relative dynamic elastic modulus output by the long short - term memory neural network layer in the previous time step, is the bias coefficient in the forget gate;

[0244] S3.2.2: Input the output result of the convolutional neural network layer in S3.1 into the input gate of the long short - term memory neural network layer:

[0245] ;

[0246] ;

[0247] ;

[0248] Among them, is t the value of the mass loss rate at the time step after passing through the input gate, is t the value of the relative compressive strength at the time step after passing through the input gate, is t the value of the relative dynamic elastic modulus at the time step after passing through the input gate, is t the weight in the forget gate at the time step, is t the bias coefficient in the forget gate at the time step;

[0249] S3.2.3: Input the result output by the convolutional neural network layer in S3.1 into the candidate memory unit of the cell state:

[0250] ;

[0251] ;

[0252] ;

[0253] is t the value of the mass loss rate output by the candidate memory unit at the time step, is t the value of the relative compressive strength output by the candidate memory unit at the time step, is t the value of the relative elastic modulus output by the candidate memory unit at the time step, is the weight in the forget gate, is the bias coefficient in the forget gate, is the hyperbolic tangent function;

[0254] S3.2.4: Combine S3.2.1 - S3.2.3 to update the damage quantification index based on the cell state:

[0255] ;

[0256] ;

[0257] ;

[0258] Among them, is t the mass loss rate updated based on the cell state at the time step, is t the relative compressive strength after the time step is updated based on the cell state, is t the relative dynamic elastic modulus after the time step is updated based on the cell state;

[0259] S3.2.5: Input the result output by the S3.1 convolutional neural network layer into the input-output gate:

[0260] ;

[0261] ;

[0262] ;

[0263] wherein, is t the value of the mass loss rate of the time step after passing through the input-output gate, is t the value of the relative compressive strength of the time step after passing through the input-output gate, is t the value of the relative dynamic elastic modulus of the time step after passing through the input-output gate, is the weight in the forget gate, is the bias coefficient in the forget gate;

[0264] S3.2.6: Based on S3.2.4 and S3.2.5, the output of the long short-term memory neural network layer is:

[0265] ;

[0266] ;

[0267] ;

[0268] wherein, is the hyperbolic tangent function, is t the mass loss rate output by the time step long short-term memory neural network layer; is t the relative compressive strength output by the time step long short-term memory neural network layer; is t the relative dynamic elastic modulus output by the time step long short-term memory neural network layer, is t the value of the mass loss rate of the time step after passing through the input-output gate, is t the value of the relative compressive strength of the time step after passing through the input-output gate, is t the value of the relative dynamic elastic modulus of the time step after passing through the input-output gate.

[0269] S3.3: Input the output result of the long short-term memory neural network layer in S3.2 into the attention layer to obtain the output result of the attention layer;

[0270] S3.3.1: Use the weight matrix to perform weighted calculation on the output result of the long short-term memory neural network layer to obtain the attention score of the damage quantification index:

[0271] ;

[0272] ;

[0273] ;

[0274] where is the hyperbolic tangent function, is the attention score of the mass loss rate at the time step, is the attention score of the relative compressive strength at the time step, is the attention score of the relative dynamic elastic modulus at the time step;

[0275] S3.3.2: Normalize the attention score of the damage quantification index in S3.3.1 through the Softmax function to obtain the attention weight of the normalized damage quantification index:

[0276] ;

[0277] ;

[0278] ;

[0279] where is t the attention weight of the mass loss rate at the time step, is t the attention weight of the relative compressive strength at the time step, is t the attention weight of the relative dynamic elastic modulus at the time step, is the exponential function with e as the base, is the total number of time steps, is the current time step;

[0280] S3.3.3: Based on the attention weight in S3.3.2, perform weighted summation on the damage quantification index output by the long short-term memory neural network layer to obtain the output result of the attention layer:

[0281] ;

[0282] ;

[0283] ;

[0284] Among them, is t the quality loss rate output by the time-step attention layer, is t the relative compressive strength output by the time-step attention layer, is t the relative dynamic elastic modulus output by the time-step attention layer; is the total time step, is the current time step.

[0285] S3.4: Input the output result of the attention layer in S3.3 into the output layer, and the output layer outputs the predicted value based on the damage quantification index feature data training set:

[0286] ;

[0287] ;

[0288] ;

[0289] Among them, is the bias of the output layer, is the weight of the output layer, is the predicted value of the quality loss rate, is the predicted value of the relative compressive strength, the predicted value of the relative dynamic elastic modulus.

[0290] Adopt the Adam optimization algorithm to update the bias and weight of the output layer of the CNN-LSTM-Attention model, which specifically includes:

[0291] (1): Determine the first-order moment estimate of the current traversal number and the second-order moment estimate of the current traversal number:

[0292] ;

[0293] ;

[0294] Among them, is the gradient of the current traversal number, is the first-order moment estimate of the current traversal number, is the first-order moment estimate of the previous traversal number, is the second-order moment estimate of the current traversal number, is the exponential decay rate of the momentum term, is the exponential decay rate of the variance term;

[0295] (2): Based on (1), correct the bias to obtain the first-moment estimate and second-moment estimate after eliminating the bias of the current traversal times:

[0296] ;

[0297] ;

[0298] where, is the first-moment estimate after eliminating the bias of the current traversal times, is the second-moment estimate after eliminating the bias of the current traversal times; is the x power of the exponential decay rate of the momentum term, is the x power of the exponential decay rate of the variance term;

[0299] (3): Based on (2), update the weights and biases of the output layer of the CNN-LSTM-Attention model, thereby realizing the optimization of the CNN-LSTM-Attention model:

[0300] ;

[0301] ;

[0302] where, is the weight of the output layer at the -th traversal, is the neural network bias of the output layer at the -th traversal, is the neural network bias of the output layer at the x -th traversal, is the learning rate, is a constant used to prevent division by zero errors;

[0303] (4): Calculate the loss function MSE of the CNN-LSTM-Attention model, update the weights and biases through the traversal times, and find the loss function that is minimized:

[0304] ;

[0305] ;

[0306] ;

[0307] Among them, MSE 1 is the loss function of the CNN-LSTM-Attention model based on the mass loss rate, MSE 2 is the loss function of the CNN-LSTM-Attention model based on the relative compressive strength, MSE 3 is the loss function of the CNN-LSTM-Attention model based on the relative dynamic elastic modulus, is the label data training set of the mass loss rate, is the label data training set of the relative compressive strength, is the label data training set of the relative dynamic elastic modulus. The label data training set of the mass loss rate, the label data training set of the relative compressive strength, and the label data training set of the relative dynamic elastic modulus all belong to the label data training set of the damage quantification index, n is the number of elements in the feature data training set, and it is set to traverse 80 times. The finally optimized weights and biases are respectively assigned to the weights and biases of the output layer of S3.4.

[0308] In this application, 8 neurons are set in the output layer of the long short-term memory neural network layer, so there are 8 initial weights of the output layer [0.5159404 ], [0.2228024 ], [0.00190622], [0.2992239 ], [0.5869981 ], [0.31229842], [0.60286933], [0.64401984]; after traversing 80 times, the weights of the output layer are [0.5159404 ], [0.2228024], [0.00190622], [0.29266465 ], [0.6093647 ], [0.35189116], [0.61722505], [0.63815206];

[0309] 1 neuron is set in the output layer, the initial bias is [0.00199428], and after traversing 80 times, the bias of the output layer is [0.04527584].

[0310] S4: The optimization of the CNN-LSTM-Attention model is completed through the above operations. At this time, the weights and biases of the output layer in the CNN-STM-ttention model are determined through optimization. The feature data test set of the damage quantification index is input into the optimized CNN-LSTM-Attention model. The output layer of the CNN-LSTM-Attention model outputs the predicted values based on the feature data test set of the damage quantification index, and the prediction results of the damage quantification index in the concrete freeze-thaw-sulfate erosion process are obtained. Using the label data test set of the damage quantification index, the convergence of the CNN-LSTM-Attention model is verified through simulation experiments.

[0311] Specifically, the feature data test set of the mass loss rate , the feature data test set of the relative compressive strength , and the feature data test set of the relative dynamic elastic modulus are input into the optimized CNN-LSTM-Attention model. The output layer of the CNN-LSTM-Attention model outputs the predicted values of the feature data test set of the damage quantification index.

[0312] After each prediction is completed, the predicted values of the feature data test set of the damage quantification index output by the output layer are added to the feature data test set of the mass loss rate, and the prediction of the damage quantification index after the next freeze-thaw-sulfate wet-dry cycle is started.

[0313] The label data test of the mass loss rate , the label data test set of the relative compressive strength , and the label data test set of the relative dynamic elastic modulus are respectively input into different models (LSTM prediction, CNN-LSTM, CNN-LSTM-Attention) for simulation experiments to verify the convergence of the CNN-LSTM-Attention model. The smaller the calculated MSE, the smaller the convergence and the higher the prediction accuracy.

[0314] Figure 3 , Figure 4 , Figure 5 are respectively the comparison situations of the test values of the concrete durability damage index and the true values of the test sample set of the label data test set of the damage quantification index on three neural network models (LSTM prediction, CNN-LSTM, CNN-LSTM-Attention). In Figure 3 , (a) is LSTM, (b) is CNN-LSTM, and (c) is CNN-LSTM-Attention; in Figure 4In the figure, (a) is LSTM, (b) is CNN-LSTM, and (c) is CNN-LSTM-Attention; Figure 5 In the figure, (a) is LSTM, (b) is CNN-LSTM, and (c) is CNN-LSTM-Attention. The blue line is the actual value of the actual damage index, and the red line is the damage index value predicted by the corresponding model. The following Table 1 shows the actual value and predicted value of the loss quantification index.

[0315] Table 1 The actual and predicted values ​​of loss quantification indicators:

[0316] ;

[0317] Based on the predicted values ​​and true values ​​of the loss quantification indicators in Table 1, the MSE of the predicted values ​​and the true values ​​is calculated.

[0318] The quality loss rate is based on the MSE of LSTM prediction = 0.0116, the MSE of CNN-LSTM prediction = 0.00143, and the MSE of CNN-LSTM-Attention prediction = 0.000874.

[0319] The relative compressive strength is based on the MSE of LSTM prediction = 0.0358, the MSE of CNN-LSTM prediction = 0.00408, and the MSE of CNN-LSTM-Attention prediction = 0.00145.

[0320] The relative dynamic elastic modulus is predicted by LSTM with MSE=0.0524, CNN-LSTM with MSE=0.00267, and CNN-LSTM-Attention with MSE=0.001364. The image intuitively shows the advancement of the CNN-LSTM-Attention model over other models.

[0321] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for predicting quantitative indicators of concrete damage during freeze-thaw-sulfate attack, characterized in that: include: S1: Determine the temperature acceleration coefficient, the corrosion solution concentration acceleration coefficient and the freeze-thaw degradation proportional coefficient, and then determine the number of freeze-thaw cycles and the number of sulfate dry-wet cycle days, so as to complete the freeze-thaw-sulfate dry-wet cycle alternation test of concrete specimens, and collect the quantitative damage indicators of concrete specimens during the test: mass loss rate, relative compressive strength and relative dynamic elastic modulus; S2: Preprocess the damage quantification index in S1 and convert the feature data set of the damage quantification index into V, F, G The label data set of the damage quantification index is divided into a feature data training set and a feature data test set according to the ratio of 8:

2. , , They are divided into a labeled data training set and a labeled data test set at a ratio of 8:2; G is the relative dynamic elastic modulus characteristic data set; F is the relative compressive strength characteristic data set; is the mass loss rate feature data set; represents the mass loss rate label dataset; Represents a relative compressive strength label dataset; Representing a relative dynamic elastic modulus label data set; constructing a feature data training set and a feature data test set of damage quantification indicators and a label data training set and a label data test set of damage quantification indicators; S3: Input the characteristic data training set of the damage quantification index of S2 into the CNN-LSTM-Attention model to obtain the predicted value based on the characteristic data training set of the damage quantification index; Adopt the Adam optimization algorithm, combine the predicted value and the label data training set of the damage quantification index to update the bias and weight of the output layer of the CNN-LSTM-Attention model, and optimize the CNN-LSTM-Attention model; Among them, the first layer of the CNN-LSTM-Attention model is the convolutional neural network layer, the second layer is the long short-term memory neural network layer, the third layer is the attention layer, and the fourth layer is the output layer; S4: The feature data test set of the damage quantification index is input into the optimized CNN-LSTM-Attention model. The output layer of the CNN-LSTM-Attention model outputs the predicted value based on the feature data test set of the damage quantification index, and the prediction result of the damage quantification index of the concrete freeze-thaw-sulfate attack process is obtained. The labeled data test set of the damage quantification index is used to verify the convergence of the CNN-LSTM-Attention model through simulation experiments.

2. The method for predicting quantitative indicators of concrete freeze-thaw-sulfate attack damage according to claim 1, characterized in that: The process of data preprocessing for S2 damage quantification index includes: S2.1: Calculate the mean and standard deviation of the damage quantification index, remove the outliers in the damage quantification index according to the calculated mean and standard deviation, and obtain the damage quantification index with missing data points; S2.2: fill in the missing data points of the damage quantification index in S2.1 by using the interpolation method to obtain the filled damage quantification index, and expand the filled damage quantification index to obtain the expanded damage quantification index; S2.3: normalize the damage quantification index expanded in S2.2 to obtain a normalized damage quantification index set; S2.4: Divide the damage quantification indicator set into a feature data training set and a feature data test set of the damage quantification indicator.

3. The method for predicting quantitative indicators of concrete freeze-thaw-sulfate attack damage according to claim 2, characterized in that: S2.3 includes: S2.3.1: Normalize the mass loss rate in the damage quantification index expanded in S2.2 to obtain the normalized mass loss rate, which is calculated as follows: ； in, is the mass loss rate after expansion, is the minimum value of the mass loss rate after expansion, is the maximum value of the mass loss rate after expansion, It is j The normalized mass loss rate; Then the set of normalized mass loss rates is obtained as : ； S2.3.2: The relative compressive strength in the expanded damage quantification index of S2.2 is normalized to obtain the normalized relative compressive strength, which is calculated as follows: ； in, is the relative compressive strength after expansion, is the minimum value of the relative compressive strength after expansion, is the maximum value of the relative compressive strength after expansion, It is j The normalized relative compressive strength; Then the normalized relative compressive strength set is obtained : ； S2.3.3: Normalize the relative dynamic elastic modulus in the expanded damage quantification index of S2.2 to obtain the normalized relative dynamic elastic modulus, which is calculated as follows: ； in, is the relative dynamic elastic modulus after expansion, is the minimum value of the relative dynamic elastic modulus after expansion, is the maximum value of the relative dynamic elastic modulus after expansion, It is j The normalized relative dynamic elastic modulus; Then the normalized relative dynamic elastic modulus set is obtained : 。 4. The method for predicting quantitative indicators of concrete freeze-thaw-sulfate attack damage according to claim 3, characterized in that: S3 includes: S3.1: Input the feature data training set of the damage quantification index into the convolutional neural network layer of the CNN-LSTM-Attention model, perform convolution operation, obtain the convolved value, and then pass ReLU The activation function performs nonlinear transformation on the convolutional values ​​to obtain nonlinear transformed values, and performs pooling operation on the nonlinear transformed values ​​to obtain the output result of the convolutional neural network layer; S3.2: Input the output result of the convolutional neural network layer of S3.1 into the long short-term memory neural network layer to obtain the output result of the long short-term memory neural network layer; wherein the long short-term memory neural network layer includes an input gate, a forget gate, an output gate and a cell state; S3.3: Input the output result of the long short-term memory neural network layer in S3.2 into the attention layer to obtain the output result of the attention layer; S3.4: The output result of the attention layer in S3.3 is input into the output layer, and the output layer outputs the predicted value based on the training set of the damage quantification indicator feature data.

5. A method for predicting quantitative indicators of concrete freeze-thaw-sulfate attack damage according to claim 4, characterized in that: The specific calculation process of S3.1 pooling operation is: ； ； ； in, It is s The value after the mass loss rate pooling operation, It is s The value after the relative compressive strength pooling operation, It is s The value after the relative dynamic elastic modulus pooling operation; It is i The value of the mass loss rate after nonlinear transformation, It is i The value after nonlinear transformation of relative compressive strength, It is i The value of the relative dynamic elastic modulus after nonlinear transformation, It is i+ 1 value of mass loss rate after nonlinear conversion, It is i +1 value after nonlinear conversion of relative compressive strength, It is i +1 value after nonlinear conversion of relative dynamic elastic modulus, The function represents the maximum value.

6. A method for predicting quantitative indicators of concrete freeze-thaw-sulfate attack damage according to claim 5, characterized in that: In S3.2, the output of the long short-term memory neural network layer is: ； ； ； in, is the hyperbolic tangent function, for t The quality loss rate of the output of the short-term memory neural network layer at the time step; for t The relative compressive strength of the output of the short-term memory neural network layer of the time step; for t The relative dynamic elastic modulus of the output of the short-term memory neural network layer at time step, yes t The time step mass loss rate is the value after the input and output gate, yes t The time step relative compressive strength input and output gate values, yes t The time step relative dynamic elastic modulus input and output gate value; yes t The time step is based on the mass loss rate after the cell state is updated, yes t The time step is based on the relative compressive strength of the cell state after updating, yes t The time step is based on the relative dynamic elastic modulus after the cell state is updated.

7. A method for predicting quantitative indicators of concrete freeze-thaw-sulfate attack damage according to claim 6, characterized in that: S3.3 includes: S3.3.1: Use a weight matrix The results of the LSTM neural network layer output are weighted to obtain the attention score of the damage quantification index: ； ； ； in, is the hyperbolic tangent function, for The attention score of the time step quality loss rate, for The relative strength of attention of the time step, for The attention score of the time step relative to the dynamic elastic modulus; S3.3.2: Normalize the attention score of the damage quantification index in S3.3.1 through the Softmax function to obtain the normalized attention weight of the damage quantification index: ； ； ； in, for t The attention weight of the time step quality loss rate, for t The attention weights of the time steps relative to the compressive strength, for t The attention weight of the time step relative to the dynamic elastic modulus, For e The exponential function with base , is the total time step, is the current time step; S3.3.3: Based on the attention weights in S3.3.2, the damage quantification indicators output by the long short-term memory neural network layer are weighted and summed to obtain the output result of the attention layer: ； ； ； in, for t The quality loss rate of the attention layer output at the time step, for t The relative compressive strength of the attention layer output at each time step, for t The relative dynamic elastic modulus of the attention layer output at time step; is the total time step, is the current time step.

8. The method for predicting quantitative indicators of concrete freeze-thaw-sulfate attack damage according to claim 7, characterized in that: The output result of the S3.3 attention layer is input into the output layer, which outputs the predicted value of the feature data training set based on the damage quantification index: ； ； ； in, is the bias of the output layer, is the weight of the output layer, is the predicted value of mass loss rate, is the predicted value of relative compressive strength, Predicted value of relative dynamic elastic modulus.

9. A method for predicting quantitative indicators of concrete freeze-thaw-sulfate attack damage according to claim 8, characterized in that: The Adam optimization algorithm is used to update the bias and weight of the output layer of the CNN-LSTM-Attention model. Specifically, the following steps are performed: (1): Determine the first-order moment estimate of the current traversal number and the second-order moment estimate of the current traversal number: ； ； in, is the gradient of the current traversal times, is the first-order moment estimate of the current traversal number, is the first-order moment estimate of the last traversal number, is the second-order moment estimate of the current traversal number, is the exponential decay rate of the momentum term, is the exponential decay rate of the variance term; (2): Based on (1), the deviation is corrected to obtain the first-order moment estimate and the second-order moment estimate after eliminating the deviation of the current traversal number: ； ； in, To eliminate the first-order moment estimate after the deviation of the current traversal number, To eliminate the bias of the current traversal times, the second-order moment estimate is obtained; is the exponential decay rate of the momentum term x Second power, is the exponential decay rate of the variance term x Power; (3): Based on (2), update the weight of the output layer of the CNN-LSTM-Attention model and bias , and then optimize the CNN-LSTM-Attention model: ； ； in, It is x+ The weight of the output layer in one pass, It is The neural network bias of the output layer of the traversal, It is x The neural network bias of the output layer of the traversal, is the learning rate, is a constant; (4): Calculate the loss function of the CNN-LSTM-Attention model MSE , update the weight by the number of traversals and bias , find the loss function that minimizes: ； ； ； in, MSE 1 is the loss function of the CNN-LSTM-Attention model based on the quality loss rate, MSE 2 is the loss function of the CNN-LSTM-Attention model based on relative compressive strength. MSE 3 is the loss function of the CNN-LSTM-Attention model based on relative dynamic elastic modulus, is the labeled data training set of mass loss rate, is a label data training set with relative compressive strength, is the label data training set of relative dynamic elastic modulus. The label data training set of mass loss rate, the label data training set of relative compressive strength and the label data training set of relative dynamic elastic modulus all belong to the label data training set of damage quantification index. n is the number of elements in the feature data training set.

Images