Saturation coral sand dynamic shear modulus prediction method based on explainable artificial intelligence

Abstract

The application discloses a kind of based on explainability artificial intelligence saturated coral sand dynamic shear modulus prediction method, including preparation saturated coral sand sample and carry out test, generate data set, utilize CNN-TCN-Attention combination neural network to establish saturated coral sand dynamic shear modulus prediction model and train;The data of prediction model is explained using SHAP method analysis.The application pays attention to the particularity of saturated coral sand soil quality and adopts resonance column test to obtain a large number of indoor test data, fill the data scarcity problem in the field of deep learning prediction, while SHAP method can effectively solve the black box problem existing in machine learning model, enhance the explainability of prediction model, can determine the influence law of each input variable on output variable.The application first proposes that explainability deep learning model realizes the accurate prediction of saturated coral sand dynamic shear modulus, and based on additional test verifies the reliability of model.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of coral sand performance research and digital modeling, and particularly relates to a saturated coral sand dynamic shear modulus prediction method based on explainable artificial intelligence. BACKGROUND

[0002] Coral sand is a special marine sediment mainly distributed in the tropical sea area between 30 degrees north and south latitude, accounting for about 40% of the seabed area. In recent years, with the vigorous development of offshore oil and wind energy resources and the construction of island reef engineering, many marine structures are built on coral sand sites. These projects may cause immeasurable damage to offshore infrastructure when facing dynamic load risks such as earthquakes, waves and hurricanes. As an important part of island and coral reef foundation engineering, the dynamic deformation characteristics of coral sand are crucial to the safety of island construction projects.

[0003] The dynamic shear modulus of soil is a basic characterization of the dynamic properties of soil and an important parameter required for site seismic response analysis. As an extremely important parameter in the safety evaluation and dynamic response analysis of such engineering earthquakes, the rationality of its selection is directly related to the safety and economy of the engineering structure. Therefore, the research on the performance analysis and prediction of the dynamic shear modulus of saturated coral sand is of great significance to improve the safety and stability of island engineering structures.

[0004] Resonant column test and flexural element test are two commonly used methods for testing dynamic shear modulus G. Resonant column test adjusts the excitation frequency to obtain the resonant frequency of the sample, and indirectly calculates the shear wave velocity V s of the sample based on wave theory, and finally obtains the dynamic shear modulus G of the order of 10 -6 ~ 10 -4 shear strain. The flexural element test directly tests the time of shear wave propagation through the sample, and then obtains the shear wave velocity V s of the test sample, and thus obtains the maximum dynamic shear modulus G max of the soil. These two test methods have their own advantages and disadvantages, and the data from a single test cannot accurately reflect the mechanical properties of the soil. Therefore, multiple tests are needed to reflect the characteristic parameters of the soil sample.

[0005] However, multiple tests produce a large amount of test data, and the test data may have problems, which makes the scientific research work complex and tedious. In addition, the prediction models and functions proposed by a large number of studies are different, and there are problems such as specific application object, single research condition and dependence on multiple parameters. The dynamic shear modulus in the entire strain range cannot be obtained through one experimental device, and the dynamic shear modulus corresponding to different strains needs to be obtained through empirical formula fitting. The existing empirical formula is only valid for a local range and cannot completely fit the entire range of test test range.

[0006] Currently, there are also many cases of using ML (Machine Learning) models to study the performance of coral sand, but the traditional machine learning model is difficult to accurately predict high-dimensional nonlinear problems in real time, and the interpretability of the prediction algorithm is not considered, which cannot effectively reveal the contribution of each input variable to the output variable and the interaction between variables.

[0007] Therefore, how to accurately judge the influencing factors of dynamic shear modulus, and use the correlation and influence degree between the influencing factors to predict the dynamic shear modulus is a technical problem to be solved at present. SUMMARY

[0008] In order to solve the problems of the prior art, the present application provides a saturated coral sand dynamic shear modulus prediction method based on explainable artificial intelligence, which uses resonance column test to obtain a large amount of indoor test data to solve the problem of data scarcity in the field of deep learning prediction, and uses global explainability analysis method to solve the black box problem of machine learning model, enhances the explainability of the prediction model, and can determine the influence law of each input variable on the output variable, greatly improving the reliability of the prediction model.

[0009] The embodiments of the present application provide the following solutions:

[0010] The embodiments of the present application provide a saturated coral sand dynamic shear modulus prediction method based on explainable artificial intelligence, which comprises the following steps:

[0011] Step 1, preparing saturated coral sand samples;

[0012] Step 2, using saturated samples for testing to generate a data set, the influencing factors of each sample in the data set include relative density, confining pressure, fine particle content and strain, and the influence target is dynamic shear modulus;

[0013] Step 3, using a CNN-TCN-Attention (Convolutional Neural Network-Time Domain Convolution Network-Attention Mechanism) combined neural network to establish a saturated coral sand dynamic shear modulus prediction model;

[0014] Step 4, using the data set to input the CNN-TCN-Attention prediction model for training, optimizing the parameters of the saturated coral sand dynamic shear modulus prediction model, and obtaining the optimized saturated coral sand dynamic shear modulus prediction model;

[0015] Step 5, using SHAP method to perform explainability analysis on the data of the optimized saturated coral sand dynamic shear modulus prediction model.

[0016] In an alternative embodiment, step one adopts a dry static pressure method and saturation treatment to prepare a saturated coral sand sample.

[0017] In an alternative embodiment, the test in step two adopts a resonance column test.

[0018] In an alternative embodiment, the saturated coral sand dynamic shear modulus prediction model in step three comprises a CNN (Convolutional Neural Networks) module, a TCN (Temporal Convolutional Networks) residual module and an attention mechanism module.

[0019] In an alternative embodiment, the CNN module comprises 6 convolution layers, 6 pooling layers and 1 fully connected layer, and the TCN residual module comprises a one-dimensional fully convolutional network layer, a causal convolution layer, a dilated convolution layer and a residual connection layer.

[0020] In an alternative embodiment, the processing flow of the saturated coral sand dynamic shear modulus prediction model is as follows:

[0021] S3.1. Input the influencing factors in the data set as input, and perform convolution, pooling and activation processing through the CNN module to extract deep features;

[0022] S3.2. The deep features enter the TCN residual module for time series processing to obtain time series deep features;

[0023] S3.3. The time series deep features enter the attention mechanism module to obtain a feature vector with attention weights and output as a prediction result.

[0024] In an alternative embodiment, in the data set, 70% of the samples are used as a training data set, and the remaining 30% of the samples are used as a test data set.

[0025] The present application has the beneficial effects of the technical solutions as follows:

[0026] The present application obtains the resonance frequency of the coral sand sample according to the resonance column test, so that the dynamic shear modulus G of the soil body can be calculated through one-dimensional wave theory. Then, the variation law of the dynamic shear modulus G with the material parameters and the external loading condition parameters is obtained, and the parameters with significant influence are selected as the input variables of deep learning. Then, combined with the deep feature mining advantage of convolutional neural network and the effective feature screening of time convolution network and attention mechanism, a CNN-TCN-Attention model is constructed to improve the prediction accuracy. Through a large number of data sets generated by indoor experiments, the accurate prediction of the dynamic shear modulus G of the coral sand is realized. Finally, based on the hybrid intelligent model, the SHAP (SHapley Additive exPlanation, global explanation analysis method) method is used to analyze the contribution of the input variables to the prediction of the dynamic shear modulus of the saturated coral sand, and the interaction and correlation between the input variables are compared and analyzed. Compared with the prior art, the present application focuses on the particularity of the saturated coral sand soil and uses the resonance column test to obtain a large amount of indoor test data, filling the data scarcity problem in the field of deep learning prediction. In addition, the SHAP method can effectively solve the black box problem existing in the machine learning model, enhance the explainability of the prediction model, and determine the influence law of each input variable on the output variable. The present application first proposes an explainable deep learning model to realize the accurate prediction of the dynamic shear modulus of the saturated coral sand, and verifies the reliability of the model based on additional tests. BRIEF DESCRIPTION OF DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present specification or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present specification, and other drawings can also be obtained by those skilled in the art without creative labor.

[0028] Figure 1 A flowchart of a saturated coral sand dynamic shear modulus prediction method based on explainable artificial intelligence provided by the present application.

[0029] Figure 2 A hierarchical structure diagram of a saturated coral sand dynamic shear modulus prediction model.

[0030] Figure 3 A sample space distribution Pearson correlation matrix diagram.

[0031] Figure 4 A regression distribution diagram between the predicted value and the actual value.

[0032] Figure 5 A global SHAP value swarm and feature importance diagram.

[0033] Figure 6 It is a feature interaction diagram.

[0034] Figure 7 It is a schematic diagram of the mutual influence relationship between different characteristic parameters. DETAILED DESCRIPTION

[0035] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field fall within the scope of protection of the embodiments of the present invention.

[0036] This embodiment provides a method for predicting the dynamic shear modulus of saturated coral sand based on explainable artificial intelligence. Figure 1 , the method comprises the following steps:

[0037] Step 1: Prepare saturated coral sand sample.

[0038] In this example, coral sand from an island reef in the Nansha Islands was used for testing. The coral sand particles are primarily composed of aragonite and high-magnesium calcite, with CaCO3 accounting for over 90%, making it a calcareous sand with a measured specific gravity of 2.80. Generally, coral sand particles with a diameter of less than 0.075 mm are considered fine particles, while the remaining particles are considered sand. The coral sand was placed through a 0.075 mm standard sieve and vibrated. Fine particles and sand particles of varying mass were uniformly mixed in varying proportions to obtain coral sand samples with varying fine sand contents (Fc) ranging from 0% to 30%. The coral sand samples were treated using a dry static pressure method, followed by presaturation, full saturation, and consolidation to obtain saturated coral sand samples.

[0039] Step 2: Conduct tests using saturated specimens to generate a data set. The influencing factors of each sample in the data set include relative density, confining pressure, fine particle content, and strain, and the influencing target is the dynamic shear modulus.

[0040] This example uses a resonant cylinder test. Vibration is applied to a cylindrical saturated coral sand sample, and the vibration frequency is varied to produce resonance. The corresponding natural frequency is used to determine the dynamic shear modulus G. During the experiment, the relative density, confining pressure, and fine particle content of the saturated coral sand sample are sequentially varied. The dynamic elastic modulus under different strains is recorded to form a data set.

[0041] Step 3: Use the CNN-TCN-Attention combined neural network to establish a saturated coral sand dynamic shear modulus prediction model, referring to Figure 2The saturated coral sand dynamic shear modulus prediction model comprises a CNN module, a TCN residual module and an attention mechanism module. In the embodiment, the CNN module comprises 6 convolution layers, an activation layer, 6 pooling layers and a full connection layer, the TCN residual module comprises a TCN convolution layer, a weight normalization layer, a ReLU layer, a Dropout layer and a 1*1 convolution kernel residual module layer, and the attention mechanism module comprises a permute layer, an Eense layer and a Multiply layer. The number of convolution kernels of the 6 convolution layers is 16, 32, 32, 64, 64 and 64 respectively, and the area size of the 6 pooling layers is 2*1. The Dropout layer parameter is set to 0.3, the learning rate is 0.01, and the training frequency is 100 times.

[0042] The processing flow of the saturated coral sand dynamic shear modulus prediction model is as follows:

[0043] S3.1, the influencing factors in the data set are taken as inputs, and the CNN module is used for convolution, pooling and activation processing to extract deep features;

[0044] S3.2, the deep features enter the TCN residual module for time series processing to obtain time series deep features;

[0045] S3.3, the time series deep features enter the attention mechanism module to obtain a feature vector with attention weights as a prediction result output.

[0046] Step four, the CNN-TCN-Attention prediction model is trained by using the data set as input, and the parameters of the saturated coral sand dynamic shear modulus prediction model are optimized to obtain an optimized saturated coral sand dynamic shear modulus prediction model.

[0047] In the embodiment, the saturated sample generates a data set of 660 samples through testing, and the influencing factors are as shown in the following table:

[0048]

[0049] The sample space distribution Pearson correlation matrix is as shown in the following table: Figure 3 As can be seen from the figure, the highest correlation is only 0.029, which indicates that there is no high correlation between any two variables, thereby proving the independence of the variables. This means that the database of input design combination samples is reasonable and reliable for developing an artificial intelligence model.

[0050] 70% of the samples are set as training data sets in the model training process, and the remaining 30% are set as test data sets, and the predicted prediction results are as shown in the following table: Figure 4As shown, the R2 of the training set of the CNN-TCN-Attention model is 0.985 and 0.981, which is accurate enough for optimization use. The p-value is used to test the difference between the residual distribution of the training set and the test set, indicating whether the prediction error comes from the same distribution. The p-value is less than 0.05, indicating that there is a significant difference in the residual distribution, indicating that the model performance is different. If the p-value is greater than or equal to 0.05, the hypothesis that the residual distribution is the same cannot be denied, indicating that the model performance is similar. The p-value of the training set and the test set in this paper is 0.07, indicating that the performance of the two is close, and there is no overfitting of the model, further verifying the reliability of the algorithm.

[0051] Step five, the SHAP method is used to perform explanatory analysis on the data of the optimized saturated coral sand dynamic shear modulus prediction model.

[0052] (1) Global explanatory analysis:

[0053] Referring to Figure 5 , the feature parameters are sorted from top to bottom according to the importance of the contribution to the dynamic shear modulus, and the positive and negative SHAP values on the horizontal axis represent the positive and negative influence on the output result, and the absolute value is larger, the influence is larger. The color represents the size of the feature parameter value, and the blue color represents the lower value and the red color represents the higher value. Therefore, Figure 5 not only can show the importance of the feature parameters, but also can reveal the influence trend of the change of the feature on the dynamic shear modulus. It can be seen from Figure 5 that the confining pressure is the most important parameter affecting the dynamic shear modulus, and the greater the confining pressure, the greater the positive SHAP value, and the greater the positive effect on the dynamic shear modulus. Similarly, the SHAP value of the relative density also gradually becomes positive with the increase of the parameter value, and has a positive effect on the dynamic shear modulus. Strain and fine particle content have a negative effect, and the negative effect on the dynamic shear modulus increases with the increase of the parameter value, indicating that strain and fine particle content are negatively correlated with dynamic shear modulus.

[0054] (2) Feature interaction analysis

[0055] Referring to Figure 6 , the SHAP interaction value is extended in global explanation, which decomposes the contribution into main influence and interaction influence. These values can be used to highlight and visualize the interactions in the data. It can also be a useful tool to understand how the model makes predictions, Figure 6 (a) The schematic diagram of the interaction value enables fast and accurate two-by-two interaction calculation in the model, which will return a matrix for each prediction, where the main influence is on the diagonal line and the interaction influence is outside the diagonal line. These values often reveal the interaction. In order to see how it affects the prediction of each sample, while viewing all predictions can understand the reaction of the model to different features, further drawings such as Figure 6(b) the heat map shown. The left y-axis is the importance feature ranking, and the features are sorted by influence from large to small, and the right y-axis is its visualization, where the color depth in the image represents the size of the SHAP value, that is, the value of the feature under the model, and the deeper the color, the larger the absolute value of the SHAP value, and the greater the influence on the model, and the top is the visualization of the prediction result of the model under these values. 460 data are selected from the training set for visualization, such as Figure 6 (b). Through the heat map, the feature importance and the interaction between features of the complex model can be effectively explored and understood, thereby improving the interpretability and explainability of the model. It can be seen from Figure 6 (a) that the interaction value plot shows the interaction between features. Specifically, the main influence of relative density and confining pressure is positive, while that of fine particle content and strain is negative. In the case of high relative density, the value of the dynamic shear modulus of saturated coral sand will increase with the increase of confining pressure, while the dynamic shear modulus will decrease with the increase of strain value. In the case of high confining pressure, the value of the dynamic shear modulus will decrease with the increase of fine particle content and strain. It can be seen from Figure 6 (b) that high prediction (high value in f(x) on the right) is related to high relative density and high confining pressure (red). Low prediction (low value in f(x) on the right) is related to low relative confining pressure and low strain (blue). The heat map reflects the situation as the interaction value Figure 1 .

[0056] (3) Interaction analysis between input variables:

[0057] Referring to Figure 7 is a schematic diagram of the interaction between different feature parameters. Figure 7 (a), 7(b), 7(c), 7(d) show the change of SHAP value with the change of relative density, confining pressure, fine particle content and strain, respectively. The overall law of the influence of the four input parameters on the dynamic shear modulus of saturated coral sand is that the increase of relative density and confining pressure will increase the predicted value of dynamic shear modulus, and the increase of fine particle content and strain will decrease the predicted value of dynamic shear modulus. It can be seen from Figure 7 (a) that the relative density is proportional to the dynamic shear modulus. It can be seen from Figure 7 (d) that the SHAP value of strain decreases with the increase of strain, and the decrease is rapid before 1e-4 and slows down after 1e-4. In summary, appropriately increasing the values of relative density and confining pressure and decreasing the values of fine particle content and strain can improve the dynamic shear modulus of saturated coral sand. The value of strain should be controlled above 1e-4 to better maintain the dynamic shear modulus of saturated coral sand.

[0058] To provide additional evidence for the effectiveness of the CNN-TCN-Attention prediction model proposed in this study, four models, including CNN, TCN, CNN-TCN, and TCN-Attention, were used to build prediction models to predict the dynamic shear modulus, and the results were compared with the CNN-TCN-Attention prediction results. The prediction effects of each model on the training set and the test set are shown in the following table.

[0059]

[0060] No matter in the training set or in the test set, the CNN-TCN-Attention prediction model has higher prediction accuracy than the other four traditional models. When the CNN-TCN-Attention prediction model is used for prediction, the R 2 value of the training set reaches 0.985, and the R 2 value of the test set reaches 0.981. The prediction accuracy of the CNN-TCN-Attention model is the highest, and the prediction accuracy of the CNN model is the weakest. The accuracy of the CNN-TCN model is higher than that of the TCN, because the prediction accuracy and feature extraction ability of the traditional TCN network model are poor, and through the deep feature mining advantage of the CNN, the prediction accuracy of the model is improved. It is more suitable for rolling bearing residual. In addition, compared with the algorithm results, it can be seen that the CNN-TCN-Attention prediction result is better than that of the CNN-TCN, the final value of different errors is lower, and the prediction result is more accurate. Because the CNN-TCN-Attention model can effectively extract key feature information by combining the TCN neural network and the attention mechanism, the correlation between the model input and output is increased. In the processing of the nonlinear problem of the saturated coral sand dynamic shear modulus prediction, it shows higher accuracy. In summary, the CNN-TCN-Attention model proposed in this paper has better performance in data prediction tasks, and the accuracy is significantly improved compared with the traditional neural network model.

[0061] Those skilled in the art will appreciate that embodiments of the application can be provided as methods, systems, or computer program products. Accordingly, the application can be embodied in the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the application can be embodied in the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk memory, CD-ROM, optical memory, etc.) having computer-usable program code embodied thereon.

[0062] The present invention is described with reference to flowcharts and / or block diagrams of methods, apparatus (modules, systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0063] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0064] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0065] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0066] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for predicting the dynamic shear modulus of saturated coral sand based on explainable artificial intelligence, characterized in that: The method comprises the following steps: Step 1: Prepare saturated coral sand samples using dry static pressure method and saturation treatment; Step 2: Conducting an experiment using saturated coral sand samples to generate a data set. The influencing factors of each sample in the data set include relative density, confining pressure, fine particle content, and strain, and the influencing target is the dynamic shear modulus. The experiment uses a resonant column test. Step 3: Use the CNN-TCN-Attention combined neural network to establish a saturated coral sand dynamic shear modulus prediction model; the saturated coral sand dynamic shear modulus prediction model includes a CNN module, a TCN residual module, and an attention mechanism module; the CNN module includes 6 convolutional layers, 6 pooling layers, and 1 fully connected layer, and the TCN residual module includes a one-dimensional fully convolutional network layer, a causal convolution layer, a dilated convolution layer, and a residual connection layer; Step 4: Using the data set to input the saturated coral sand dynamic shear modulus prediction model for training, optimizing the parameters of the saturated coral sand dynamic shear modulus prediction model to obtain an optimized saturated coral sand dynamic shear modulus prediction model; Step 5: Use the SHAP method to perform explanatory analysis on the data of the optimized saturated coral sand dynamic shear modulus prediction model.

2. The method for predicting the dynamic shear modulus of saturated coral sand based on explainable artificial intelligence according to claim 1, wherein: The processing flow of the saturated coral sand dynamic shear modulus prediction model is as follows: S3.

1. Take the influencing factors in the dataset as input and perform convolution, pooling, and activation processing through the CNN module to extract deep features. S3.2, the deep features enter the TCN residual module for temporal processing to obtain temporal deep features; S3.

3. The time-series deep features enter the attention mechanism module, and the feature vector with attention weight is obtained and output as the prediction result.

3. The method for predicting the dynamic shear modulus of saturated coral sand based on explainable artificial intelligence according to claim 1, wherein: In the dataset, 70% of the samples are used as training dataset, and the remaining 30% of the samples are used as testing dataset.

Images