A method for predicting the strength of concrete

Abstract

The present application provides a concrete strength prediction method, relating to the field of concrete strength prediction. The concrete strength prediction process provided by the present application includes constructing a concrete strength prediction data set, converting time series data into frequency domain data through discrete Fourier transform, converting the real part, imaginary part and amplitude of each frequency component in the frequency domain data into a high-dimensional feature vector through linear mapping, obtaining frequency domain feature information by using a frequency domain information fusion module in combination with the high-dimensional feature vectors of the real part, imaginary part and amplitude, obtaining time domain feature information of the time series data by using a multilayer perception machine, then better combining the time domain and frequency domain feature information by using an improved self-attention mechanism to obtain a fusion feature vector, and finally splicing all the fusion feature vectors and predicting the concrete strength through a fully connected neural network. The method can effectively combine time domain and frequency domain information and improve the prediction accuracy of the method.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of concrete strength prediction, and particularly relates to a concrete strength prediction method. BACKGROUND

[0002] The concrete strength refers to the compressive capacity of concrete under external force, indicating the structural bearing capacity and stability of concrete, and insufficient concrete strength cannot guarantee the structural safety of buildings, leading to cracks, deformation or even collapse, causing serious economic losses and safety accidents, therefore, concrete strength prediction is crucial, and predicting the strength change of concrete under different conditions in advance can help engineers make accurate construction decisions, ensure engineering quality and safety, and effectively reduce risks and hidden dangers caused by insufficient strength.

[0003] The field of concrete strength prediction mainly uses statistical models based on experimental data and machine learning methods, although statistical models can provide effective prediction results in some cases, they are usually difficult to handle nonlinear and complex multivariate relationships, and machine learning methods establish prediction models by analyzing various influencing factors such as concrete mix and curing conditions, however, they often fail to fully consider the frequency characteristics of time series data composed of influencing factors to better understand the internal structure of data and improve the reliability and accuracy of prediction results.

[0004] Time series data is often composed of different frequency components, low-frequency components usually represent long-term trends or periodic fluctuations, and high-frequency components usually represent short-term changes or sudden fluctuations, therefore, in machine learning methods, by mapping time series data to the frequency domain, combining frequency domain analysis and time domain analysis, the model can better understand the internal structure of data and improve the reliability of prediction results. SUMMARY

[0005] The application provides a concrete strength prediction method, which improves prediction accuracy by combining time domain and frequency domain information, first uses discrete Fourier transform to convert time series data of concrete strength factor data into different frequency components, and enhances the expression ability of frequency components through linear mapping, then proposes a frequency domain information fusion module to effectively extract frequency domain feature information, uses a multilayer perceptron to extract time domain feature information of time series data of concrete strength factor data, improves the self-attention mechanism to fuse time domain and frequency domain feature information, and captures the correlation between them, and finally integrates the fused information through a fully connected neural network to accurately predict the concrete strength.

[0006] The technical method adopted by the application to achieve the above purpose specifically includes the following steps:

[0007] S1. Collect data on factors affecting concrete strength and construct a concrete strength prediction dataset;

[0008] S2. The factor data constitutes time series data. The time series data is converted into different frequency components through discrete Fourier transform to form frequency domain data, and a frequency domain mapping module is constructed.

[0009] S3. Calculate the amplitude based on the real and imaginary parts of the frequency components, and use linear mapping to convert the real, imaginary parts and amplitude into high-dimensional feature vectors;

[0010] S4. A frequency domain information fusion module is proposed, which introduces an adaptive feature adjustment factor to weight and aggregate the real part, imaginary part and amplitude of each frequency component in the frequency domain data to obtain the frequency domain feature information of each frequency component.

[0011] S5. By employing a multilayer perceptron, deep learning is performed on the high-dimensional feature relationships in the time series data of concrete strength factor data to extract time-domain feature information containing global time series information.

[0012] S6. Improve the self-attention mechanism. Construct a time-domain and frequency-domain fusion module using the improved self-attention mechanism. Use time-domain feature information as the query vector and frequency-domain feature information as the key and value vector. Introduce an asymmetric adjustment factor in the attention score. Based on the time-domain feature information, adaptively learn the correlation of frequency-domain feature information. Combine the time-domain and frequency-domain feature information to generate a brand-new fusion feature vector.

[0013] S7. Construct a fully connected neural network, concatenate all fused feature vectors to obtain a prediction vector, and use the fully connected neural network to calculate the predicted value of concrete strength.

[0014] Preferably, in step S1, data on factors affecting concrete strength are collected, including cement content, cement strength, water-cement ratio, pouring height, curing temperature, curing humidity, and curing time. For missing values ​​that are not collected, the average of the data three days before and after the data is used to fill in the missing values ​​and construct a concrete strength prediction dataset.

[0015] Preferably, in step S2, a frequency domain mapping module is constructed by inputting the time series data of the c-th concrete strength factor in the concrete strength prediction dataset. c∈[1,N], where N is the number of collected concrete strength factor data types, X c The data is in the time domain, where T is the length of the time series of the input data. Let X be the data value of the Cth type of concrete strength factor collected at time step t, where t∈[1,T]. c Applying the Discrete Fourier Transform, X cThe mapping is to different frequency components, and the specific mathematical model is:

[0016]

[0017] In the formula, is the kth frequency component of the time series data of the cth concrete strength factor data after discrete Fourier transform, k is the frequency component number, is the exponential form of a complex number, where e is the base of natural logarithm, i is the imaginary unit, and i 2 =-1, π is the circular constant, is the real part of A c，k , cos(·) is the cosine function, is the imaginary part of A c，k , sin(·) is the sine function, and A c is the frequency domain data obtained by arranging the frequency components of the cth concrete strength factor data in ascending order of frequency component number, A c =[A 。，0 , A c，1 ,... A c，T-1 ].

[0018] Preferably, the concrete strength prediction data is subjected to discrete Fourier transform, and by converting the time domain data into frequency domain data, different frequency components in the data are extracted, which can help the model to capture various characteristics such as periodicity, long-term trend and short-term fluctuation in the data, balance the learning of low-frequency and high-frequency components by the model, and thus improve the prediction ability and robustness of the model.

[0019] Preferably, in S3, the amplitude |A c，k | is calculated according to the real part Re(A c，k ) and the imaginary part Im(A c，k ) of the kth frequency component A c，k of the cth concrete strength factor data, and the amplitude can help to understand the strength of the frequency component, and the specific mathematical model is:

[0020]

[0021] In the formula, is the amplitude of the frequency component A c，k ;

[0022] Further, the real part, the imaginary part and the amplitude are linearly mapped to a richer space to represent a high-dimensional vector, and the specific mathematical model is:

[0023]

[0024] and are high-dimensional feature vectors obtained by linear mapping of the real part, imaginary part and amplitude, respectively, d is the dimension of the vector after linear mapping, W Re , W Im and are trainable weight matrices, b Re , b Im and are trainable bias parameters.

[0025] Preferably, by linearly mapping the real part, imaginary part and amplitude of the frequency component, a high-dimensional feature vector is generated, which can effectively capture the feature difference between different frequency components, convert low-dimensional data into a high-dimensional feature space, and improve the expression and learning ability of the model. By introducing trainable weights and bias parameters, the model can automatically optimize these mapping processes according to the data, further improving the prediction accuracy.

[0026] Preferably, in the S4, a frequency domain information fusion module is proposed, which specifically includes the following steps:

[0027] S41, an adaptive feature adjustment factor is introduced, and the adaptive feature adjustment factor is calculated according to the strength of the real part, imaginary part and amplitude high-dimensional feature vector matrix. The specific mathematical model is:

[0028]

[0029] In the formula, are the adaptive feature adjustment factors of the real part, imaginary part and amplitude, respectively, and ||·||F F is the Frobenius norm;

[0030] Further, the real part, imaginary part and amplitude high-dimensional feature vectors are weighted and summed using the factor to dynamically adjust the contribution of the three high-dimensional feature vectors to the final output. The feature vector of the fused overall frequency domain information is obtained, and the specific mathematical model is:

[0031]

[0032] In the formula, and are high-dimensional feature vector matrices composed of the real part, imaginary part and amplitude of the frequency component in the frequency domain data of the cth concrete strength factor according to the frequency component number,

[0033] is the feature vector matrix of the fused real part, imaginary part and amplitude high-dimensional feature vectors;

[0034] S42, a new feature vector is obtained by performing nonlinear mapping on Y1, Y1 is connected with the new feature vector in residual connection, and final frequency domain feature information is obtained by performing nonlinear mapping again, and a specific mathematical model is as follows:

[0035] Y2 = ReLU(Y1W1 + b1);

[0036] Y c = ReLU((Y2 + Y1)W2 + b2);

[0037] In the formula, Y2 is a new feature vector obtained by performing nonlinear mapping, W1, is a trainable weight matrix, b1, is a bias parameter, and ReLU(·) is a ReLU activation function, is frequency domain feature information matrix obtained by combining residual connection and nonlinear mapping.

[0038] Preferably, by introducing an adaptive feature adjustment factor, nonlinear mapping and residual connection, the frequency domain information fusion module can efficiently fuse three high-dimensional feature vectors of real part, imaginary part and amplitude, dynamically adjust the contribution of each feature to the final output, and further improve the representation ability of the model through a deep network structure. The method can effectively capture complex patterns in frequency domain information and improve the robustness of the model in practical applications.

[0039] Preferably, in S5, the time series data of the cth concrete strength factor data is learned using a multilayer perceptron to obtain overall time domain feature information, the multilayer perceptron includes three layers of input layer, hidden layer and output layer, wherein the input layer captures the basic features in X c through linear transformation and activation function, and a specific mathematical model is as follows:

[0040] h1 = ReLU(X c W3 + b3);

[0041] In the formula, X is a basic feature vector of the input layer, d h is the dimension of the input layer basic feature, W3 and b3 are respectively the weight matrix and bias parameter of the input layer, and ReLU(·) is a ReLU activation function;

[0042] The output layer further extracts time domain feature information from the output feature of the hidden layer, and a specific mathematical model is as follows:

[0043] h2 = ReLU(h1W4 + b4);

[0044] In the formula, h2 is a time sequence feature of the hidden layer, respectively are the weight matrix and bias parameter learnable by the hidden layer;

[0045] The hidden layer combines these basic features to extract deeper temporal relationships to obtain temporal features, and the specific mathematical model is:

[0046] h c = h2W out + b out ;

[0047] In the formula, is the time domain feature information containing the cth concrete strength factor data output by the output layer, respectively are the weight matrix and bias parameter learnable by the output layer.

[0048] Preferably, by using a three-layer multilayer perceptron to learn the high-dimensional feature relationship of time series data, a deeper temporal information modeling can be realized, the input layer performs linear transformation on the original data, and the activation function is used to enhance the representation of nonlinear features, and the expression ability of the model is further optimized through the hidden layer and the output layer. This process can extract comprehensive information of time series data to provide more accurate time series feature representation for subsequent analysis and prediction.

[0049] Preferably, in S6, the self-attention mechanism is improved, and a time domain and frequency domain fusion module is constructed, which specifically includes the following steps:

[0050] S61, taking the time domain feature information h c as the query vector, and the frequency domain feature vector matrix Y c The time domain feature information of each frequency component in Y c = [y c，0 , y c，1 ,..., y c，T-1 ], wherein is the frequency domain feature information of the kth frequency component of the cth concrete strength factor data, the query vector and the key vector calculate the similarity through the dot product method to obtain the attention score, and the specific mathematical model is:

[0051]

[0052] In the formula, is the attention score of the kth frequency component of the cth concrete strength factor data, is the transpose of y c，k ;

[0053] S62, the interaction between the time domain feature and the frequency domain feature is asymmetric, an asymmetric adjustment factor is introduced to asymmetrically adjust the attention score, the asymmetric adjustment factor is calculated and multiplied with the attention score s′c，k The specific mathematical model is:

[0054] s c，k = s' c，k + σ c，k ;

[0055] In the formula, is the asymmetric adjustment factor of the kth frequency component of the cth concrete strength factor data, β, is a learnable parameter, is the attention score of the asymmetric adjustment factor of the kth frequency component of the cth concrete strength factor data;

[0056] S63, normalize s c，k to obtain the attention weight, and the specific mathematical model is:

[0057]

[0058] In the formula, is the normalized attention weight of s c，k , and is the attention score of the asymmetric adjustment factor of the ith frequency component of the cth concrete strength factor data;

[0059] According to the weight, the corresponding value vector is weighted and summed to obtain a feature vector after fusion of the frequency domain and time domain feature information, and the specific mathematical model is:

[0060]

[0061] In the formula, is the weighted sum of all frequency domain feature information of the cth concrete strength factor data according to the corresponding attention weight, and the weighted sum is added to the time domain feature information h c to obtain the fusion feature vector.

[0062] Preferably, the time domain feature information and the frequency domain feature information are effectively fused by the improved self-attention mechanism, an asymmetric adjustment factor is introduced to adjust the interaction mode of the time domain and the frequency domain features, thereby improving the flexibility and expression ability of the model. The improved self-attention mechanism uses a query vector to perform a weighting operation on each frequency domain feature information to combine the time domain and frequency domain information, thereby more comprehensively understanding the mode of the data.

[0063] Preferably, in S7, the fusion feature vectors obtained from all concrete strength factor data are spliced to obtain a final feature vector for prediction, and the specific mathematical model is:

[0064]

[0065] In the formula, is a feature vector spliced by all concrete strength factor data fusion feature vectors, is a splicing operation;

[0066] The spliced feature vector is calculated by a fully connected neural network to obtain a concrete strength prediction value, and the specific mathematical model is: P = FCNN(L);

[0067] In the formula, is a final concrete strength prediction value, and FCNN(·) is a fully connected neural network.

[0068] Compared with the prior art, the present application has the following advantages: the present application converts the time domain data of the concrete strength factors into frequency domain data by discrete Fourier transform, extracts the real part, imaginary part and amplitude of each frequency component in the frequency domain data, maps the real part, imaginary part and amplitude into a high-dimensional feature vector using linear mapping, dynamically fuses the three high-dimensional feature vectors by a frequency domain information fusion module to obtain global frequency domain feature information of the frequency components, applies a multilayer perceptron to extract time domain feature information of the time series data, then uses an improved self-attention mechanism to fuse the time domain and frequency domain feature information, finally splices all the fused features and inputs them into a fully connected neural network for final concrete strength prediction, thereby improving the prediction accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0069] Figure 1 is a concrete strength prediction method step diagram.

[0070] Figure 2 is a frequency domain information fusion module diagram.

[0071] Figure 3 is a multilayer perceptron structure diagram.

[0072] Figure 4 is a time domain and frequency domain fusion module diagram.

[0073] Figure 5 is a concrete strength prediction effect diagram. DETAILED DESCRIPTION

[0074] This invention proposes a method for predicting concrete strength. This method utilizes Discrete Fourier Transform to convert time-domain data of concrete strength factors into frequency-domain data. It enhances the feature representation of frequency components in the frequency-domain data through linear mapping. A frequency-domain information fusion module is proposed to combine the real, imaginary, and amplitude information of frequency components to obtain frequency-domain feature information. A multilayer perceptron obtains the overall information of the time-series data in the time domain. Based on this, an improved self-attention mechanism is implemented to achieve deep fusion of time-domain and frequency-domain information, capturing more complex feature relationships. Finally, all fused information is concatenated and calculated using a fully connected neural network to obtain the concrete strength prediction result. The technical solutions in the embodiments of this invention will be described in detail and completely below, specifically including the following steps, such as... Figure 1 As shown.

[0075] S1. Collect data on factors affecting concrete strength and construct a concrete strength prediction dataset.

[0076] Furthermore, in S1, 102 days of concrete strength factor data are collected, including cement content, cement strength, water-cement ratio, pouring height, curing temperature, curing humidity, and curing time. The seven types of data collected are preprocessed. For missing values, the average of the data before and after the current data is used to fill in the missing values ​​to ensure the quality and consistency of the data and to construct a high-quality concrete strength prediction dataset. Then, the data is divided into training set and validation set in chronological order with a ratio of 8:2. The data of the previous 7 days of each day are used as the historical data input for model training.

[0077] S2. The factor data constitutes time series data. The time series data is converted into different frequency components through discrete Fourier transform to form frequency domain data, and a frequency domain mapping module is constructed.

[0078] Further, in step S2, a frequency domain mapping module is constructed to obtain a concrete strength prediction dataset, containing data on seven factors affecting concrete strength. The time series data of the c-th concrete strength factor is denoted as... c∈[1,7], where T is the length of the time series of the input data, T=7. Let X be the observed value of the c-th type of concrete strength factor at time step t, where t∈[1,7]. Apply Discrete Fourier Transform to transform the time series X of the c-th type of concrete strength factor data. c Mapping to the frequency domain, we obtain the different frequency components of the time series. The specific mathematical model is as follows:

[0079]

[0080] In the formula, A c k is the kth frequency component of the time series of the cth concrete strength factor data, k is the frequency component number, is in the exponential form of complex number, where e is the base of natural logarithm, i is the imaginary unit, and i 2 = -1, π is the ratio of a circle, is the real part of the kth frequency component, cos(·) is the cosine function, is the imaginary part of the kth frequency component, sin(·) is the sine function, A c is the frequency component obtained from the cth concrete strength factor data, A c = [A c，0 , A c，1 , …, A c，6 ].

[0081] S3, the amplitude is calculated according to the real part and the imaginary part of the frequency component, and the real part, the imaginary part and the amplitude are converted into a high-dimensional feature vector by linear mapping.

[0082] Further, in the S3, the kth frequency component A c，k of the frequency domain data of the cth concrete strength factor is composed of the real part Re(A c，k ) and the imaginary part Im(A c，k ), and the amplitude |A c，k | of the frequency component is calculated by the real part and the imaginary part, which helps the model to understand the significance of the frequency component. The specific mathematical model is:

[0083]

[0084] In the formula, is the amplitude of the frequency component A c，k ;

[0085] The real part, the imaginary part and the amplitude of A c，k are converted into a more abundant high-dimensional feature vector by linear mapping, so as to improve the feature expression ability of the model. The specific mathematical model is:

[0086]

[0087] In the formula, and are high-dimensional vectors with a dimension of 32 obtained by linear mapping of the real part, the imaginary part and the amplitude, respectively, W Re , W Im and are trainable weight matrices, b Re , b Im and are trainable bias parameters.

[0088] S4, a frequency domain information fusion module is proposed, and an adaptive feature adjustment factor is introduced to aggregate the real part, imaginary part and amplitude of each frequency component in the frequency domain data to obtain the frequency domain feature information of each frequency component.

[0089] Further, in the S4, the frequency domain information fusion module is proposed, as shown in Figure 2 The specific steps include the following steps: S41, the high-dimensional feature vectors of the real part, imaginary part and amplitude of all frequency components in the frequency domain data of the cth concrete strength factor are arranged into a matrix according to the frequency component number to obtain the real part matrix The imaginary part matrix And the amplitude matrix An adaptive feature adjustment factor is calculated according to the three matrices, and the specific mathematical model is:

[0090]

[0091] In the formula, The adaptive feature adjustment factors of the real part, imaginary part and amplitude are respectively, and the Frobenius norm is F

[0092] The adaptive feature adjustment factor is used to weight and sum the high-dimensional feature vectors of the real part, imaginary part and amplitude to obtain a feature vector containing overall frequency domain information, and the specific mathematical model is:

[0093]

[0094] In the formula, The feature vector matrix combining the high-dimensional feature vectors of the real part, imaginary part and amplitude is

[0095] S42, the feature vector matrix Y1 is sequentially subjected to nonlinear mapping and nonlinear mapping combined with residual connection to obtain a vector matrix sufficiently fusing different dimension information in the frequency domain feature, and the specific mathematical model is:

[0096] Y2 = ReLU(Y1W1 + b1);

[0097] Y c = ReLU((Y2 + Y1)W2 + b2);

[0098] In the formula, The new feature vector obtained by nonlinear mapping is The trainable weight matrix is The bias parameter is ReLU(·), which is the ReLU activation function, The frequency domain feature information matrix obtained by nonlinear mapping combined with residual connection is The trainable weight matrix is​ biasing parameters.

[0099] S5, by adopting a multi-layer perception, deep learning is performed on high-dimensional feature relationships in time series data of the concrete strength factor data, and time domain feature information containing global information of the time series is extracted.

[0100] Further, in the S5, as shown in Figure 3 , the time series data X c of the cth concrete strength factor data is processed by using a multi-layer perception, The multi-layer perception includes three layers of an input layer, a hidden layer and an output layer, the input layer includes a linear transformation and an activation function, and basic features in input data are captured, and a specific mathematical model is as follows:

[0101] h1=ReLU(X c W3+b3);

[0102] In the formula, is a basic feature vector of the input layer, are respectively a learnable weight matrix and a biasing parameter of the input layer, and ReLU(·) is a ReLU activation function,

[0103] The hidden layer again combines these features by using a linear transformation and an activation function to extract deeper time sequence feature vectors, and a specific mathematical model is as follows:

[0104] h2=ReLU(h1W4+b4);

[0105] In the formula, is a time sequence feature vector of the hidden layer, are respectively a learnable weight matrix and a biasing parameter of the hidden layer;

[0106] Finally, the output layer maps the learned features to a final output space to obtain a feature vector containing overall time sequence information, and a specific mathematical model is as follows:

[0107] h c =h2W out +b out ;

[0108] In the formula, is time domain feature information finally output by the output layer, representing overall time sequence information of the cth time series data, are respectively a learnable weight matrix and a biasing parameter of the output layer.

[0109] S6, improving the self-attention mechanism, using the improved self-attention mechanism to construct a time domain and frequency domain fusion module, taking the time domain feature information as a query vector, taking the frequency domain feature information as a key and value vector, and introducing an asymmetric adjustment factor in the attention score, on the basis of the time domain feature information, adaptively learning the correlation of the frequency domain feature information, combining the time domain and frequency domain feature information to generate a new fusion feature vector.

[0110] Further, in the S6, the self-attention mechanism is improved, and a time domain and frequency domain fusion module is constructed, as shown in Figure 4 , specifically comprising the following steps:

[0111] S61, taking the time domain feature information h c as a query vector, and taking the frequency domain feature vector matrix Y c as a key and value vector, Y c = [y c，0 , y c，1 , …, y c，T-1 ], wherein is the frequency domain feature information of the kth frequency component of the cth concrete strength factor data, the similarity weight of the query vector and the key vector is calculated by the dot product method, thereby obtaining the attention score, and the specific mathematical model is:

[0112]

[0113] In the formula, is the attention score of the kth frequency component of the cth concrete strength factor data, is the transpose of y c，k ;

[0114] S62, an asymmetric adjustment factor is introduced, which reflects the asymmetric relationship between the time domain feature information and the frequency domain feature information, and is added to the preliminary calculated attention score s′ c，k , and the specific mathematical model is:

[0115]

[0116] s c，k = s′ c，k + σ c，k ;

[0117] In the formula, is the asymmetric adjustment factor of the kth frequency component of the cth concrete strength factor data, β, is a learnable parameter, is the attention score of the kth frequency component of the cth concrete strength factor data after fusion of the asymmetric adjustment factor;

[0118] S63, to s c，k The normalized attention weight is obtained, and the specific mathematical model is as follows:

[0119]

[0120] In the formula, is s c，k The normalized attention weight is obtained, is the attention score of the asymmetric adjustment factor of the ith frequency component of the concrete strength factor data of the cth kind;

[0121] Based on the normalized attention weight, the corresponding value vector is weighted and summed, so as to obtain a feature vector fused with frequency domain and time domain characteristic information, and the specific mathematical model is as follows:

[0122]

[0123] In the formula, is the weighted sum of the frequency domain characteristic information of all frequency components of the concrete strength factor data of the cth kind according to the corresponding attention weight, and the time domain characteristic information h c is added, and the obtained fusion feature vector is obtained.

[0124] S7, a full connection neural network is constructed, all fusion feature vectors are spliced to obtain a prediction vector, and a concrete strength prediction value is obtained by calculating the prediction vector by using the full connection neural network.

[0125] Further, in the S7, the fusion feature vectors obtained by all concrete strength factor data are spliced according to the last dimension to obtain a feature vector for prediction, and the specific mathematical model is as follows:

[0126]

[0127] In the formula, is the feature vector obtained by splicing the fusion feature vectors of all concrete strength factor data, is a splicing operation;

[0128] Then, a concrete strength prediction value is obtained by using a full connection neural network, and the specific mathematical model is as follows: P = FCNN (L).

[0129] In the formula, is the final concrete strength prediction value, and FCNN (·) is a full connection neural network.

[0130] Further, the method is implemented based on the Python 3.8 programming language and the PyTorch framework, in order to accelerate the training process, the method runs in the CUDA 12 environment, and uses the NVIDIA RTX 3090 GPU for calculation, and during the training process, the learning rate is 1*10 -3 , the batch size is 12, the optimizer is the stochastic gradient descent (SGD), and the loss function is the mean square error (MSE).

[0131] Further, the prediction effect of the method is as shown in Figure 5 , wherein the vertical coordinate is the concrete strength (MPa), the horizontal coordinate is the time (day), the gray dotted line in the figure represents the actual strength value of the concrete, and the black solid line is the concrete strength value predicted by the method. Figure 5 It can be seen that the fitting degree between the predicted value and the actual value is high, and the change trend of the two is basically consistent with the passage of time, indicating that the method can effectively capture the change rule of the concrete strength, and verify the effectiveness and feasibility of the method in the concrete strength prediction task.

[0132] The above is only a preferred embodiment of the present application, and it should be noted that for those skilled in the art, without departing from the inventive concept, several modifications and improvements can be made, which are all within the scope of protection of the present application.

Claims

1. A method of predicting the strength of concrete, characterized by, The method comprises the following steps: S1, collecting factor data affecting the strength of concrete, and constructing a concrete strength prediction data set; S2, the factor data constitutes time series data, the time series data is converted into different frequency components by discrete Fourier transform, frequency domain data is formed, and a frequency domain mapping module is constructed; S3, the amplitude is calculated according to the real part and the imaginary part of the frequency component, and the real part, the imaginary part and the amplitude are converted into high-dimensional feature vectors by linear mapping; S4, a frequency domain information fusion module is proposed, an adaptive feature adjustment factor is introduced, the real part, the imaginary part and the amplitude of each frequency component in the frequency domain data are weighted and aggregated to obtain the frequency domain feature information of each frequency component; S5, by using a multilayer perception, the high-dimensional feature relationship in the time series data of the concrete strength factor data is deeply learned, and the time domain feature information containing the global information of the time series is extracted; S6, the self-attention mechanism is improved, the time domain and frequency domain fusion module is constructed by using the improved self-attention mechanism, the time domain feature information is used as a query vector, the frequency domain feature information is used as a key and value vector, and an asymmetric adjustment factor is introduced in the attention score, the correlation of the frequency domain feature information is adaptively learned on the basis of the time domain feature information, and a new fusion feature vector is generated by combining the time domain and frequency domain feature information; S7, a fully connected neural network is constructed, all fusion feature vectors are spliced to obtain a prediction vector, and a concrete strength prediction value is obtained by calculating the prediction vector by using the fully connected neural network.

2. The method of claim 1, wherein, In S1, the factor data of the concrete strength is collected, including cement content, cement strength, water-cement ratio, pouring height, curing temperature, curing humidity and curing time, the missing values in the data are filled with the average value of the data of each three days before and after the data, and a high-quality concrete strength prediction data set is constructed.

3. The method of claim 2, wherein, In the S2, the time series data of the concrete strength factor data in the concrete strength prediction data set is recorded as wherein is the time series length, is the data value of the concrete strength factor data of the th type collected at the th time step, , , , is the number of collected concrete strength factor data types, and the discrete Fourier transform is used to map to different frequency components, and the specific mathematical model is: ； ； ； In the formula, For the first The time series data of the concrete strength factor were obtained by discrete Fourier transform. Each frequency component Number the frequency components. It is the exponential form of the complex number, where is the base of the natural logarithm. The imaginary unit satisfies , Pi for The real part, for The imaginary part, It is a cosine function. It is a sine function. To make the first Frequency domain data obtained by arranging the frequency components of various concrete strength factor data in sequence. .

4. The method of claim 3, wherein, In S3, the first In the frequency domain data of concrete strength factors, the first Frequency components From the real part and the virtual part The amplitude of this frequency component is calculated based on its real and imaginary parts. The specific mathematical model is as follows: ； wherein is the amplitude of the frequency component of the frequency component The real part, the imaginary part and the amplitude are mapped into high-dimensional space by linear mapping to obtain high-dimensional feature vectors, and the specific mathematical model is: ； ； ； wherein, , and are high-dimensional feature vectors obtained by linear mapping of the real part, imaginary part and amplitude of the frequency component , is the dimension of the vector after linear mapping, , and are trainable weight matrices, , and are trainable bias parameters.

5. The method of claim 4, wherein, In S4, the frequency domain information fusion module comprises the following steps: S41, the first In the frequency domain data of concrete strength factors, the high-dimensional eigenvectors of the real, imaginary, and amplitude values ​​of all frequency components are used to construct a matrix according to the frequency component numbering, thus obtaining the real part matrix. Imaginary part matrix and magnitude matrix An adaptive feature adjustment factor is introduced, and the specific mathematical model is as follows: ； ； ； wherein , , are the real part, imaginary part, and adaptive characteristic adjustment factor of the amplitude, respectively, is the Frobenius norm; The real part, the imaginary part and the amplitude high-dimensional feature vectors are weighted and summed by using the adaptive feature adjustment factor to obtain a feature vector containing overall frequency domain information, and the specific mathematical model is: ； In the formula, is a feature vector matrix combining the real part, the imaginary part and the amplitude high-dimensional feature vector; S42、to obtaining new feature vectors by performing nonlinear mapping, and combining the new feature vectors with the residual connection again to obtain frequency domain feature information by performing nonlinear mapping, and the specific mathematical model is as follows: ； ； In the formula, is a new feature vector obtained through a nonlinear mapping, is a trainable weight matrix, is a bias parameter, is an activation function, is a frequency domain feature information matrix obtained through a nonlinear mapping combined with a residual connection, is a trainable weight matrix, is a bias parameter.

6. The method of concrete strength prediction according to claim 5, wherein, In S5, for the first Time series data of concrete strength factors Using a multilayer perceptron to obtain information containing The temporal features of the overall time series information; the multilayer perceptron includes an input layer, hidden layers, and an output layer. The input layer captures [data] through linear transformation and activation functions. The fundamental characteristics of [the system], and the specific mathematical model are as follows: ； wherein, is the base feature vector obtained for the input layer, is the dimension of the input layer hidden vector, , are the input layer learnable weight matrix, bias parameter respectively, is the activation function, is the activation function. The hidden layer further extracts deeper time sequence features in the basic features, and the specific mathematical model is: ； In the formula, a time sequence feature vector obtained for the hidden layer, , respectively a weight matrix and a bias parameter learnable for the hidden layer. The output layer maps the time sequence features to the final output space to obtain the time domain feature information, and the specific mathematical model is: ； In the formula, is the time series data overall information of the concrete strength factor data output by the output layer, is the time domain feature information of the time series data overall information of the concrete strength factor data output by the output layer, , are respectively the weight matrix and the bias parameter learnable by the output layer.

7. The method of concrete strength prediction according to claim 6, wherein In S6, the self-attention mechanism is improved, and a time domain and frequency domain fusion model is constructed, comprising the following steps: S61, the first Temporal characteristics of concrete strength factor data As a query vector, the frequency domain feature information of each frequency component in the frequency domain. As a key and value vector ,in No. Type of data The frequency domain feature information of the frequency components is used to calculate the attention score between the query vector and the key vector through a dot product. The specific mathematical model is as follows: ； In the formula, is the th attention score for the th frequency component of the concrete strength factor data; and S62, introduce an asymmetry adjustment factor, which is added to the attention score The specific mathematical model is: ； ； In the formula, is a first concrete strength factor data asymmetry adjustment factor of a first frequency component, , is a learnable parameter, is a first concrete strength factor data attention score of a fusion asymmetry adjustment factor of a first frequency component; S63, to The normalized attention weight is obtained, and the specific mathematical model is as follows: ； In the formula, is normalized attention weight, is the first concrete strength factor data attention score of the first frequency component fusion asymmetric adjustment factor; The corresponding value vectors are weighted and summed according to the weights to obtain a feature vector fused from the frequency domain and the time domain feature information, and the specific mathematical model is: ； In the formula, The first Concrete strength factor data will all frequency components of the frequency domain feature information according to the corresponding attention weight is weighted, get weighted sum and time domain feature information Addition, the obtained fusion feature vector.

Images