A method, apparatus and equipment for predicting the compressive strength of cementitious materials

Abstract

This application proposes a method, apparatus, and device for predicting the compressive strength of cementitious materials, relating to the field of data processing technology. The method includes: conducting compressive strength tests and grey correlation analysis on sample cementitious materials to determine target influencing factors; creating an initial adversarial coupling model based on the target influencing factors, and inputting the original sample training set into the data augmentation sub-model of the initial adversarial coupling model; generating first intermediate data and a first loss function using SimAM attention mechanism, generative adversarial network, and conditional variational autoencoder; generating initial prediction results by combining a genetic algorithm and BP neural network embedded in the strength prediction sub-module; verifying the results with the original sample validation set to determine a second loss function; iteratively training the parameters of each module based on the obtained loss function until a preset iteration threshold is reached to obtain the target adversarial coupling model; and inputting the data of the cementitious material to be tested into the target adversarial coupling model to obtain the predicted compressive strength of the cementitious material to be tested.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus and equipment for predicting the compressive strength of cementitious materials. Background Technology

[0002] Solid potash mining generates large amounts of solid and liquid tailings. According to environmental policies, these tailings cannot be discharged, so they are often mixed and backfilled underground, with the addition of cementing materials. To improve mining recovery rates and control mine pressure, some potash companies primarily use dry tailings backfilling and wet "water-sand" backfilling to address surface tailings and brine storage issues. However, due to the low strength of the backfill, mining safety and recovery rates cannot be effectively guaranteed. Therefore, how to test the strength of the backfill has become a pressing problem. Uniaxial compressive strength (UCS) is one of the most important physical and mechanical properties of rock masses in civil and mining engineering design and is also used for rock mass classification. The main method for obtaining UCS is direct laboratory methods. However, direct laboratory methods require high-quality rock cores to obtain reliable UCS, which is extremely difficult to obtain for highly weathered rocks. Furthermore, due to limitations in sample location and lithology, the universality of empirical formulas is gradually becoming apparent. Applying the same empirical formula to different rock types can lead to underestimation or overestimation of UCS. Furthermore, in the field of mining engineering, due to cost considerations, experiments on the same material are often not conducted too many times, resulting in insufficient experimental data and thus affecting the final strength prediction results. Summary of the Invention

[0003] In view of this, this application proposes a method, apparatus and equipment for predicting the compressive strength of cementitious materials.

[0004] The technical solution of this application is implemented as follows: The first aspect of this invention provides a method for predicting the compressive strength of a cementitious structure, comprising:

[0005] Compressive strength tests and grey correlation analysis were conducted on the cementitious samples to determine the target influencing factors;

[0006] An initial adversarial coupling model is created based on the target influencing factor, and the original sample training set is input into the data augmentation sub-model of the initial adversarial coupling model. The SimAM attention mechanism, generative adversarial network and conditional variational autoencoder embedded in the data augmentation sub-model are used to generate the first intermediate data and the first loss function.

[0007] The first intermediate data is input into the intensity prediction submodule of the initial adversarial coupling model, and an initial prediction result is generated based on the genetic algorithm and BP neural network embedded in the intensity prediction submodule; the initial prediction result is verified with the original sample validation set to determine the second loss function;

[0008] Based on the first loss function and the second loss function, the parameters of each module in the data augmentation sub-model are iteratively trained until a preset iteration threshold is reached to obtain the target adversarial coupling model.

[0009] The data of the cementitious body to be tested is input into the target adversarial coupling model to obtain the predicted compressive strength of the cementitious body to be tested.

[0010] Based on the above technical solutions, preferably, the step of conducting compressive strength tests and grey correlation analysis on the sample cementitious body to determine the target influencing factors includes:

[0011] A compressive strength test was conducted on the sample cementitious body to obtain compressive strength test data; the compressive strength test data included a reference sequence composed of the uniaxial compressive strength of cementitious bodies in experiments with different ratios and a comparison sequence composed of various solidification rate control factors;

[0012] The compressive strength test data were dimensionlessly processed using the mean method to obtain a standard reference sequence and a standard comparison sequence.

[0013] Based on the principle of grey relational analysis, the grey relational coefficient between the standard reference sequence and the standard comparison sequence under each preset index is obtained, and the target influence factor is determined based on the grey relational coefficient.

[0014] Based on the above technical solutions, preferably, the data augmentation sub-model includes an encoder, a decoder, and a discriminator, and the encoder, decoder, and discriminator all incorporate a SimAM attention mechanism; the step of inputting the original sample training set into the data augmentation sub-model of the initial adversarial coupled model, and generating first intermediate data and a first loss function using the SimAM attention mechanism, generative adversarial network, and conditional variational autoencoder embedded in the data augmentation sub-model, includes:

[0015] The original sample training set is input into the encoder for reparameter sampling to obtain the first training parameters;

[0016] The first training parameters are input into the decoder for training to obtain the second training parameters;

[0017] Based on the second training parameters, the reconstruction loss function and the KL divergence loss function are determined, and the second training parameters are input into the discriminator to obtain the first intermediate data and the adversarial loss function.

[0018] The sum of the reconstruction loss function, the KL divergence loss function, and the adversarial loss function is determined as the first loss function.

[0019] Based on the above technical solutions, preferably, the step of inputting the original sample training set into the data augmentation sub-model of the initial adversarial coupling model, and generating the first intermediate data and the first loss function using the SimAM attention mechanism, generative adversarial network, and conditional variational autoencoder embedded in the data augmentation sub-model, includes:

[0020] Based on the original sample training set, the true target value and conditional features are determined, and the true target value and conditional features are input into the encoder and processed by the Adam optimizer and SimAM attention module to obtain the latent distribution parameters of the original sample training set.

[0021] The latent distribution parameters are resampled to obtain the first training parameters, which include latent variables and conditional features;

[0022] The first training parameters are input into the decoder, and after training using the ReLU activation function, Adam optimizer and SimAM attention module, the second training parameters are output.

[0023] The second training parameters and the conditional features are input into the discriminator to generate an adversarial loss function.

[0024] Based on the above technical solutions, preferably, the step of inputting the second training parameters and the conditional features into the discriminator to generate an adversarial loss function includes:

[0025] The second training parameters and the conditional features are input into the discriminator to generate the discriminator loss function;

[0026] Backpropagation is performed based on the discriminator loss function to update the weights and biases of the discriminator.

[0027] Based on the above technical solutions, preferably, the step of inputting the first intermediate data into the intensity prediction submodule of the initial adversarial coupling model, and generating the initial prediction result based on the genetic algorithm and BP neural network embedded in the intensity prediction submodule, includes:

[0028] For each individual in the first intermediate data, fitness screening is performed based on its chromosome information and a genetic algorithm to determine the appropriate network structure and hyperparameters; the chromosome information includes the hidden layer structure, activation function name, and learning rate.

[0029] Based on the above technical solutions, preferably, the step of verifying the initial prediction result with the original sample validation set to determine the second loss function includes:

[0030] The initial prediction results and the original sample validation set are processed using the ReLU activation function and the Adam optimizer to obtain the second loss function during forward propagation;

[0031] The weights and biases of the BP neural network are updated based on the second loss function.

[0032] More preferably, a second aspect of the present invention provides a device for predicting the compressive strength of a cementitious structure, comprising: an influence analysis module, a first training module, a second training module, an iterative training module, and a strength prediction module; wherein,

[0033] The impact analysis module is configured to perform compressive strength tests and grey correlation analysis on the sample cementitious body to determine the target impact factor.

[0034] The first training module is configured to create an initial adversarial coupling model based on the target influencing factor, and input the original sample training set into the data augmentation sub-model of the initial adversarial coupling model, and use the SimAM attention mechanism, generative adversarial network and conditional variational autoencoder embedded in the data augmentation sub-model to generate first intermediate data and a first loss function;

[0035] The second training module is configured to input the first intermediate data into the strength prediction submodule of the initial adversarial coupling model, generate an initial prediction result based on the genetic algorithm and BP neural network embedded in the strength prediction submodule, and verify the initial prediction result with the original sample validation set to determine the second loss function.

[0036] The iterative training module is configured to iteratively train the parameters of each module in the data augmentation sub-model based on the first loss function and the second loss function until a preset iteration threshold is reached to obtain the target adversarial coupling model.

[0037] The strength prediction module is configured to input the data of the cementitious body to be tested into the target adversarial coupling model to obtain the predicted compressive strength of the cementitious body to be tested.

[0038] More preferably, a third aspect of the present invention provides an electronic device, including a processor and a memory; the memory has a computer program stored thereon, wherein the computer program, when executed by the processor, implements the method for predicting the compressive strength of the cementitious body as described in the first aspect.

[0039] More preferably, a fourth aspect of the present invention provides a computer storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method for predicting the compressive strength of the cementitious body as described in the first aspect.

[0040] The method for predicting the compressive strength of cementitious materials proposed in this application has the following advantages over related technologies:

[0041] 1. The SimAM attention mechanism assigns different weights to features based on their importance, allowing the model to focus more on features that significantly impact prediction results during training, thus improving data quality and effectiveness. Generative Adversarial Networks (GANs) generate new data samples similar to the original data distribution through a game between the generator and discriminator, increasing data diversity. Conditional Variational Autoencoders (CVAs) generate specific types of data under given conditions, further enriching the feature combinations. Combining genetic algorithms and backpropagation neural networks (BP neural networks) enables the model to better fit the complex nonlinear relationship between the compressive strength of cementitious structures and input features, thereby improving prediction accuracy. The data augmentation sub-model expands the training set by generating high-quality augmented samples, optimizing the fitting ability of the BP neural network. Simultaneously, the strength prediction sub-module learns jointly on real and generated samples, and its prediction error is weighted and fed back into the loss function of the conditional variational autoencoder, inversely optimizing the encoder and decoder of the data augmentation sub-model, thus achieving complementarity between the two models.

[0042] 2. The target influencing factors were identified through compressive strength tests and grey correlation analysis on the cementitious samples. Compressive strength tests directly obtain data on the actual performance of the cementitious material under pressure, while grey correlation analysis filters out the factors most closely related to compressive strength from numerous potential factors. This process avoids interference from irrelevant factors, allowing subsequent model construction to focus on the key factors truly affecting compressive strength. This provides accurate and targeted input for the entire prediction model, ensuring the reliability of the prediction results from the outset and laying a solid foundation for improving prediction accuracy.

[0043] 3. After training with a small number of samples using a genetic algorithm, a superior neural network structure is applied to the BP neural network, enhancing its interpretability and significantly reducing computation time. A GAN discriminator is introduced into the Conditional Variational Autoencoder (CVA), incorporating the GAN loss function into the CVA's loss function through weighting, thus improving the quality of generated samples. The SimAM attention mechanism is used to optimize the discriminator and generator of the CVA, as well as the discriminator of the Generative Adversarial Network (GAN), strengthening the nonlinearity of the neural network. This allows the network to focus more on high-information features and suppress redundant features, thereby improving the quality of generated data. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 A schematic flowchart illustrating a method for predicting the compressive strength of a cementitious material, provided in an embodiment of this application;

[0046] Figure 2 A schematic diagram illustrating the principle framework of a method for predicting the compressive strength of a cementitious material, provided in an embodiment of this application;

[0047] Figure 3 This is a schematic diagram of the structure of a cementitious compressive strength prediction device provided in an embodiment of this application;

[0048] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0049] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0050] In some embodiments, such as Figure 1 As shown, Figure 1 A flowchart illustrating a method for predicting the compressive strength of a cementitious material, provided in an embodiment of this application; a method for conducting a compressive strength test on a sample cementitious material, provided in this application, includes:

[0051] S110, compressive strength test and grey correlation analysis were performed on the sample cementitious body to determine the target influencing factor.

[0052] In this embodiment, the sample cementitious material can be a magnesium-based tailings admixture. The target influencing factors can be determined by combining qualitative and quantitative analysis. Qualitative analysis mainly involves understanding the consolidation mechanism of magnesium-based tailings by consulting relevant literature, and identifying influencing factors from the perspective of chemical and physical reactions. Quantitative analysis mainly involves performing grey correlation analysis on the experimental data. The higher the grey correlation, the more significant the impact of the target influencing factor on compressive strength. The target influencing factors include at least one of the following: halide salt ratio, fly ash content, aggregate gradation, water-reducing agent content, and loading rate.

[0053] In some embodiments, S110, a compressive strength test and grey correlation analysis are performed on the sample cementitious material to determine the target influencing factor, including:

[0054] Compressive strength tests were conducted on the sample cementitious bodies to obtain compressive strength test data; the compressive strength test data included a reference sequence composed of the uniaxial compressive strength of cementitious bodies in experiments with different ratios and a comparison sequence composed of various solidification rate control factors;

[0055] The mean method was used to perform dimensionless processing on the compression test data to obtain the standard reference sequence and the standard comparison sequence.

[0056] Based on the principle of grey relational analysis, the grey relational coefficient between the standard reference sequence and the standard comparison sequence under each preset index is obtained, and the target influence factor is determined based on the grey relational coefficient.

[0057] In this embodiment, the uniaxial compressive strength of magnesium-based tailings cementitious aggregate is taken as an example. The corresponding sequence is the reference sequence. The corresponding sequence is a comparison sequence. The uniaxial compressive strength of magnesium-based tailings cementitious aggregates from experiments with different proportions is used as the reference sequence. Three condensation rate regulating factors were used as comparison sequences. , i represents the ratio of cementitious agent to MgCl2, the halide ratio, and the fly ash content, respectively.

[0058] The original data is dimensionless using the mean method, as shown in the following formula:

[0059] ;

[0060] When constructing the correlation discrete function and calculating the correlation coefficient, the reference sequence is first dimensionless. The resulting sequence is... This serves as the standard benchmark for system analysis. Accordingly, each comparison sequence is also dimensionless, resulting in a sequence. Used for comparison with a reference series. Based on the principle of grey relational analysis, the grey relational coefficient between the reference series and each comparison series under each index is calculated using the following formula to quantify the relative similarity.

[0061] ;

[0062] In the formula, Indicates the first The comparison sequence in the th ... Grey relational coefficients under each indicator; For the first The comparison sequence and the reference sequence at the _ ... The absolute difference under each indicator; Δmin and Δmax represent the minimum and maximum distances from the reference sequence among all indicators of the comparison sequences, respectively; This is the resolution coefficient, and its value range is 0 < <1, here we take the constant value 0.5.

[0063] Based on the grey relational coefficients calculated above, the formula for calculating grey relational degree is as follows:

[0064] .

[0065] S120: An initial adversarial coupling model is created based on the target impact factor. The original sample training set is input into the data augmentation sub-model of the initial adversarial coupling model. The SimAM attention mechanism, generative adversarial network and conditional variational autoencoder embedded in the data augmentation sub-model are used to generate the first intermediate data and the first loss function.

[0066] In this embodiment, please refer to Figure 2 , Figure 2 Figure 2 This is a schematic diagram illustrating the principle framework of a method for predicting the compressive strength of cementitious materials provided in this application embodiment. The initial adversarial coupling model mainly consists of two parts: the first part is SimAM-CVAE-GAN, namely SimAM attention mechanism-conditional variational autoencoder-generative adversarial model; the second part is GA-BP, namely genetic algorithm-backpropagation neural network. SimAM-CVAE-GAN is mainly divided into four parts: encoder, reparameter sampling, decoder, and discriminator. The encoder, decoder, and discriminator all incorporate the SimAM attention mechanism. GA-BP consists of two main parts: a genetic algorithm and a backpropagation neural network.

[0067] In some embodiments, the data augmentation sub-model includes an encoder, a decoder, and a discriminator, all of which incorporate a SimAM attention mechanism. The original sample training set is input into the data augmentation sub-model of the initial adversarial coupled model. First intermediate data and a first loss function are generated using the SimAM attention mechanism, generative adversarial network, and conditional variational autoencoder embedded in the data augmentation sub-model.

[0068] The original sample training set is input into the encoder for reparameter sampling to obtain the first training parameters;

[0069] The first training parameters are input into the decoder for training to obtain the second training parameters;

[0070] The reconstruction loss function and KL divergence loss function are determined based on the second training parameters, and the second training parameters are input into the discriminator to obtain the first intermediate data and the adversarial loss function.

[0071] The sum of the reconstruction loss function, the KL divergence loss function, and the adversarial loss function is determined as the first loss function.

[0072] In some embodiments, the original sample training set is input into the data augmentation sub-model of the initial adversarial coupling model, and the SimAM attention mechanism, generative adversarial network, and conditional variational autoencoder embedded in the data augmentation sub-model are used to generate first intermediate data and a first loss function, including:

[0073] The true target value and conditional features are determined based on the original sample training set, and the true target value and conditional features are input into the encoder and processed by the Adam optimizer and SimAM attention module to obtain the latent distribution parameters of the original sample training set.

[0074] Reparameter sampling is performed on the latent distribution parameters to obtain the first training parameters containing latent variables and conditional features;

[0075] The first training parameters are input into the decoder, and after training using the ReLU activation function, Adam optimizer and SimAM attention module, the second training parameters are output.

[0076] The second training parameters and conditional features are input into the discriminator to generate an adversarial loss function.

[0077] In this embodiment, the original sample training set is Z-score normalized and then input into the data augmentation sub-model of the initial adversarial coupling model. The SimAM energy function calculation process is as follows: Let the output of the previous layer be... The input to the current hidden layer is obtained by weighted summation based on weights and biases. The weighted summation process is shown below:

[0078] ;

[0079] In the formula, This represents the output value of the k-th neuron in the previous layer. This represents the input value of the current sample to the i-th neuron in this layer; This indicates the number of neurons in the previous layer. This indicates the number of neurons in the current layer.

[0080] The feature vector of a single sample in the current layer is:

[0081] .

[0082] Calculate the entire layer Find the mean and variance of each neuron, and calculate the energy of the i-th neuron. The calculation process is as follows:

[0083] ;

[0084] Calculate attention weights The energy is mapped to the [0,1] interval using the Sigmoid function:

[0085] ;

[0086] The neuron outputs are weighted to obtain optimized features. :

[0087] ;

[0088] Output features After undergoing nonlinear activation by the activation function, the output is passed to the next layer, i.e.:

[0089] ;

[0090] In the formula, This represents the activation function. This represents the feature value after activation.

[0091] The sample data is input into the encoder along with the true target value y and the conditional feature c. The encoder has two hidden layers, employing the ReLU activation function and the Adam optimizer. After training with a SimAM attention module, the encoder outputs the latent distribution parameters of the sample. and .

[0092] We perform reparameter sampling, sampling the first training parameter, i.e., the latent variable z, as shown in the following formula:

[0093] .

[0094] The latent variable z and conditional features c form a vector [z, c] which is input to the decoder. The decoder has two hidden layers, uses the ReLU activation function and the Adam optimizer. After training the decoder with the addition of the SimAM attention module, the output generates the second training parameters, which is the target value. .

[0095] Based on the generated target value Reconstruction losses and KL divergence This will generate the target value. The condition c is input into the discriminator, which has two hidden layers. The Leaky ReLU activation function and the Adam optimizer are used to calculate the discriminator loss. and combat losses And finally calculate the first loss function. .

[0096] The reconstruction loss function is calculated as follows:

[0097] ;

[0098] The calculation process of the KL divergence loss function is as follows:

[0099] ;

[0100] Discriminator loss function The calculation process is as follows:

[0101] ;

[0102] Combating losses The calculation process is as follows:

[0103] ;

[0104] First loss function It can be represented as:

[0105] ;

[0106] In the formula, The KL divergence loss function represents the first loss function. The weight of the content The adversarial loss function is represented by the first loss function. The weight it occupies in the middle.

[0107] In some embodiments, the second training parameters and conditional features are input into the discriminator to generate an adversarial loss function, including:

[0108] The second training parameters and conditional features are input into the discriminator to generate the discriminator loss function;

[0109] Backpropagation is performed based on the discriminator loss function to update the weights and biases of the discriminator.

[0110] In this embodiment, before updating the weights and biases, the output of SimAM is... For input Differentiation is performed to control the amplification and scaling of the gradient. The calculation process is as follows:

[0111] Gradient calculation from activation function to SimAM attention mechanism module:

[0112] ;

[0113] The gradient calculation from the SimAM attention mechanism module to the input gradient of the next layer is as follows:

[0114] .

[0115] In the formula, Indicates the first level of this layer The input values ​​of each neuron; Indicates the first level of this layer The attention weights of each neuron; Indicates the first level of this layer The energy of one neuron; Indicates the first Attention weights of each neuron Input to the i-th neuron The degree of sensitivity.

[0116] Among them, when hour:

[0117] ;

[0118] when hour:

[0119] .

[0120] The gradient between the previous layer and the linear layer is calculated as follows:

[0121] Feedback to the previous layer output The calculation process is as follows:

[0122] ;

[0123] Substituting the above formula, we get:

[0124] ;

[0125] After obtaining the derivative of the loss with respect to the weights and biases, the weights and biases of the discriminator are updated according to the Adam optimizer. The error signal is then transmitted to the previous layer to continue updating the weights and biases of the previous layer.

[0126] S130, the first intermediate data is input into the strength prediction submodule of the initial adversarial coupling model, and the initial prediction result is generated based on the genetic algorithm and BP neural network embedded in the strength prediction submodule; the initial prediction result is verified with the original sample validation set to determine the second loss function.

[0127] In some embodiments, first intermediate data is input into the intensity prediction submodule of the initial adversarial coupling model, and initial prediction results are generated based on the genetic algorithm and BP neural network embedded in the intensity prediction submodule, including:

[0128] For each individual in the first intermediate data, fitness screening is performed based on its chromosome information and a genetic algorithm to determine the appropriate network structure and hyperparameters; chromosome information includes hidden layer structure, activation function name, and learning rate.

[0129] In this embodiment, the genetic algorithm is executed only in the first forward propagation, and the process is as follows: Individuals are encoded using a genetic algorithm using chromosomes. Each chromosome contains three pieces of information: hidden layer structure, activation function name, and learning rate. The population is then initialized. Fitness is calculated using a fast evaluation method: each time, a random subset of the training set (composed of a portion of the original sample data) is selected to construct the backpropagation (BP) and optimizer for that individual, and then trained quickly for a small number of steps to obtain the individual's fitness. All individuals are evaluated and sorted according to their fitness. An elite retention mechanism is used, employing selection and mutation. Selection means selecting a certain proportion of individuals with high fitness to retain as parents, while mutation involves randomly altering the hidden layer structure, activation function name, and learning rate of the selected individuals. This process is repeated until a set number of generations is reached.

[0130] In some embodiments, validating the initial prediction results against the original sample validation set to determine the second loss function includes:

[0131] The initial prediction results and the original sample validation set are processed using the ReLU activation function and the Adam optimizer to obtain the second loss function during forward propagation;

[0132] The weights and biases of the BP neural network are updated based on the second loss function.

[0133] In this embodiment, the first intermediate data and the original samples are combined into a dataset and then divided. The first intermediate data is all used as the training set, and the original sample data is divided into the training set and the validation set. The divided dataset is input into a BP neural network, which is trained using a genetic algorithm to obtain a superior neural network. The ReLU activation function and the Adam optimizer are used, and the loss function result is obtained during forward propagation. .

[0134] BP neural network based on Update the weights and biases, and The loss function is then passed back to the CVAE, thus completing the second loss function. All calculations:

[0135] ;

[0136] In the formula, δ represents LBP in the second loss function. The weight it occupies in the middle.

[0137] The encoder and decoder of CVAE can be based on the second loss function. Update the weights and biases.

[0138] S140, the parameters of each module in the data augmentation sub-model are iteratively trained based on the first loss function and the second loss function until the preset iteration threshold is reached, thus obtaining the target adversarial coupling model.

[0139] The first and second loss functions are used to measure the difference between the model output and the expected result. By continuously adjusting the parameters, the model gradually approaches the optimal state. By comprehensively considering these two loss functions, the various modules in the data augmentation sub-model can be optimized, ensuring that the generated data not only closely approximates real data in statistical characteristics but also effectively improves the performance of downstream tasks. The iterative training process involves using the backpropagation algorithm to calculate the gradient based on the loss function and updating the model parameters in the opposite direction of the gradient, gradually reducing the value of the loss function until a preset iteration threshold is reached.

[0140] S150, input the data of the cementitious body to be tested into the target adversarial coupling model to obtain the predicted compressive strength of the cementitious body to be tested.

[0141] After iterative training of the parameters of each module of the data augmentation sub-model based on the first and second loss functions, the target adversarial coupling model has acquired the ability to effectively process input data and extract features. When the cementitious body data to be tested is input into the model, the model will use its learned data features and patterns to map the input data to the output result of compressive strength through a complex internal calculation and transformation process.

[0142] In one optional embodiment, after the target adversarial coupling model is successfully trained, the entire model can be evaluated by calculating four evaluation metrics: mean absolute error (MAE), root mean square error (RMSE), correlation coefficient (R), and mean absolute percentage error (MAPE) on the validation set, to determine whether the prediction accuracy of the trained model meets the standards. The calculation process for each evaluation metric is as follows:

[0143] ;

[0144] In one example, experiments were conducted on 36 groups of magnesium-based tailings cementitious aggregates, and the experimental data on their uniaxial compressive strength are shown in Table 1 below:

[0145]

[0146] Table 1

[0147] Based on current research on the consolidation mechanism of magnesium-based tailings, the three selected factors (mass ratio of composite binder to MgCl2, mass ratio of old brine tailings salt, and mass ratio of fly ash in the composite binder) all affect the uniaxial compressive strength of the magnesium-based tailings cementitious body. The selection of these three factors is supported by scientific evidence. Grey relational analysis was used to further analyze the correlation between these influencing factors.

[0148] The uniaxial compressive strength of magnesium-based tailings cementitious bodies obtained from experiments with different proportions was used as the reference sequence X0(k), and three factors were used as the comparison sequence. (i represents the mass ratio of composite binder to MgCl2, the mass ratio of old brine tailings salt, and the mass ratio of fly ash in the composite binder, respectively). The original data were dimensionless using the mean method, a correlation discrete function was constructed, the grey correlation coefficient was calculated, the grey correlation degree was calculated and sorted, and the results are shown in Table 2.

[0149]

[0150] Table 2

[0151] Table 2 shows that the grey relational degrees between the uniaxial compressive strength of the magnesium-based tailings cementitious body and the three factors are 0.690, 0.639, and 0.592, respectively. This indicates that all three factors are strongly correlated with the uniaxial compressive strength of the magnesium-based tailings cementitious body. Among them, the mass ratio of composite binder to MgCl2 is the key factor affecting the uniaxial compressive strength of the magnesium-based tailings cementitious body, and the mass ratio of old brine tailings salt is slightly stronger than the mass ratio of fly ash in the composite binder. Based on qualitative and quantitative analysis, the influencing factors are identified as the mass ratio of composite binder to MgCl2, the mass ratio of old brine tailings salt, and the mass ratio of fly ash in the composite binder.

[0152] Based on the experimental data, the initial adversarial coupling model was trained, and some parameters are shown in Table 3:

[0153]

[0154] Table 3

[0155] Four evaluation metrics were calculated for the validation set: Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Correlation Coefficient (R), and Mean Absolute Percentage Error (MAPE). The model was then evaluated based on these results. The training results (compared to a traditional backpropagation neural network) are shown in Tables 4 and 5.

[0156]

[0157] Table 4

[0158]

[0159] Table 5

[0160] The R-value is a coefficient that measures the correlation between two variables. The closer R is to 1, the stronger the correlation between the two variables. The table shows that the model's R-value reaches 0.9821, which is very close to 1, indicating a strong correlation between the predicted value and the original data. This indicates that the model fits the data very well and represents a 31.05% improvement over the traditional BP neural network. The table also shows that MAE=0.1029, RMSE=0.1556, and MAPE=3.76%, which are improvements of 78.58%, 71.51%, and 79.83% respectively compared to the traditional BP neural network. These three parameters all indicate that the target adversarial coupling model has a small prediction error, meaning that the coupling model has good accuracy in predicting the strength of magnesium-based tailings cementitious bodies and a significant improvement in prediction accuracy compared to the traditional BP neural network.

[0161] In some embodiments, please refer to Figure 3 , Figure 3 This is a schematic diagram of the structure of a cementitious body compressive strength prediction device provided in an embodiment of this application. This application provides a cementitious body compressive strength prediction device 300, including: an influence analysis module 310, a first training module 320, a second training module 330, an iterative training module 340, and a strength prediction module 350; wherein...

[0162] The impact analysis module 310 is configured to perform compressive strength tests and grey correlation analysis on the sample cementitious body to determine the target impact factor.

[0163] The first training module 320 is configured to create an initial adversarial coupling model based on the target influencing factor, and input the original sample training set into the data augmentation sub-model of the initial adversarial coupling model. The first intermediate data and the first loss function are generated by using the SimAM attention mechanism, generative adversarial network and conditional variational autoencoder embedded in the data augmentation sub-model.

[0164] The second training module 330 is configured to input the first intermediate data into the strength prediction submodule of the initial adversarial coupling model, generate initial prediction results based on the genetic algorithm and BP neural network embedded in the strength prediction submodule, and verify the initial prediction results with the original sample validation set to determine the second loss function.

[0165] The iterative training module 340 is configured to iteratively train the parameters of each module in the data augmentation sub-model based on the first loss function and the second loss function until a preset iteration threshold is reached, thereby obtaining the target adversarial coupling model.

[0166] The strength prediction module 350 is configured to input the data of the cementitious body to be tested into the target adversarial coupling model to obtain the predicted compressive strength of the cementitious body to be tested.

[0167] In some embodiments, the impact analysis module 310 is specifically configured as follows:

[0168] Compressive strength tests were conducted on the sample cementitious bodies to obtain compressive strength test data; the compressive strength test data included a reference sequence composed of the uniaxial compressive strength of cementitious bodies in experiments with different ratios and a comparison sequence composed of various solidification rate control factors;

[0169] The mean method was used to perform dimensionless processing on the compression test data to obtain the standard reference sequence and the standard comparison sequence.

[0170] Based on the principle of grey relational analysis, the grey relational coefficient between the standard reference sequence and the standard comparison sequence under each preset index is obtained, and the target influence factor is determined based on the grey relational coefficient.

[0171] In some embodiments, the data augmentation sub-model includes an encoder, a decoder, and a discriminator, and the SimAM attention mechanism is added to the encoder, decoder, and discriminator; the first training module 320 is specifically configured as follows:

[0172] The original sample training set is input into the encoder for reparameter sampling to obtain the first training parameters;

[0173] The first training parameters are input into the decoder for training to obtain the second training parameters;

[0174] The reconstruction loss function and KL divergence loss function are determined based on the second training parameters, and the second training parameters are input into the discriminator to obtain the first intermediate data and the adversarial loss function.

[0175] The sum of the reconstruction loss function, the KL divergence loss function, and the adversarial loss function is determined as the first loss function.

[0176] In some embodiments, the first training module 320 is specifically configured as follows:

[0177] The true target value and conditional features are determined based on the original sample training set, and the true target value and conditional features are input into the encoder and processed by the Adam optimizer and SimAM attention module to obtain the latent distribution parameters of the original sample training set.

[0178] Reparameter sampling is performed on the latent distribution parameters to obtain the first training parameters containing latent variables and conditional features;

[0179] The first training parameters are input into the decoder, and after training using the ReLU activation function, Adam optimizer and SimAM attention module, the second training parameters are output.

[0180] The second training parameters and conditional features are input into the discriminator to generate an adversarial loss function.

[0181] In some embodiments, the first training module 320 is specifically configured as follows:

[0182] The second training parameters and conditional features are input into the discriminator to generate the discriminator loss function;

[0183] Backpropagation is performed based on the discriminator loss function to update the weights and biases of the discriminator.

[0184] In some embodiments, the second training module 330 is specifically configured as follows:

[0185] For each individual in the first intermediate data, fitness screening is performed based on its chromosome information and a genetic algorithm to determine the appropriate network structure and hyperparameters; chromosome information includes hidden layer structure, activation function name, and learning rate.

[0186] In some embodiments, the second training module 330 is specifically configured as follows:

[0187] The initial prediction results and the original sample validation set are processed using the ReLU activation function and the Adam optimizer to obtain the second loss function during forward propagation;

[0188] The weights and biases of the BP neural network are updated based on the second loss function.

[0189] It should be noted that the cementitious compressive strength prediction device provided in this application embodiment and the cementitious compressive strength prediction method provided in this application embodiment are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned cementitious compressive strength prediction method, and the repeated parts will not be described again.

[0190] In some embodiments, an electronic device 400 provided in this application includes a processor 410 and a memory 420; the memory 410 stores a computer program, wherein the computer program, when executed by the processor 420, implements the above-described method for predicting the compressive strength of the cementitious body.

[0191] Specifically, processor 420 may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. Processor 420 may also include onboard memory for caching purposes. Processor 420 may be a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of this application.

[0192] Memory 410 may be any medium capable of containing, storing, transmitting, propagating, or transmitting instructions. For example, memory may include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, instruments, or propagation media. Specific examples of memory 410 include: magnetic storage devices such as magnetic tape or hard disk drives (HDDs); optical storage devices such as optical discs (CD-ROMs); and may also be random access memory (RAM) or flash memory; and / or wired / wireless communication links.

[0193] This application also provides a computer-readable medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for predicting the compressive strength of a cementitious material. This computer-readable medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into that device / apparatus / system. The aforementioned computer-readable medium carries one or more programs, which, when executed, implement the method as described in the embodiments of this application.

[0194] According to embodiments of this application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wired, optical fiber, radio frequency signals, etc., or any suitable combination thereof.

[0195] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application. Therefore, the scope of this application should not be limited to the above embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for predicting the compressive strength of a cementitious structure, characterized in that, include: Compressive strength tests and grey correlation analysis were conducted on the cementitious samples to determine the target influencing factors; An initial adversarial coupling model is created based on the target influencing factor, and the original sample training set is input into the data augmentation sub-model of the initial adversarial coupling model. The SimAM attention mechanism, generative adversarial network and conditional variational autoencoder embedded in the data augmentation sub-model are used to generate the first intermediate data and the first loss function. The first intermediate data is input into the intensity prediction submodule of the initial adversarial coupling model, and an initial prediction result is generated based on the genetic algorithm and BP neural network embedded in the intensity prediction submodule; the initial prediction result is verified with the original sample validation set to determine the second loss function; Based on the first loss function and the second loss function, the parameters of each module in the data augmentation sub-model are iteratively trained until a preset iteration threshold is reached to obtain the target adversarial coupling model. The data of the cementitious body to be tested is input into the target adversarial coupling model to obtain the predicted compressive strength of the cementitious body to be tested.

2. The method for predicting the compressive strength of cementitious bodies as described in claim 1, characterized in that, The compressive strength test and grey correlation analysis of the sample cementitious body were conducted to determine the target influencing factors, including: A compressive strength test was conducted on the sample cementitious body to obtain compressive strength test data; the compressive strength test data included a reference sequence composed of the uniaxial compressive strength of cementitious bodies in different ratio experiments and a comparison sequence composed of various solidification rate control factors; The compressive strength test data were dimensionlessly processed using the mean method to obtain a standard reference sequence and a standard comparison sequence. Based on the principle of grey relational analysis, the grey relational coefficient between the standard reference sequence and the standard comparison sequence under each preset index is obtained, and the target influence factor is determined based on the grey relational coefficient.

3. The method for predicting the compressive strength of cementitious bodies as described in claim 1, characterized in that, The data augmentation sub-model includes an encoder, a decoder, and a discriminator, all of which incorporate a SimAM attention mechanism. The process of inputting the original sample training set into the data augmentation sub-model of the initial adversarial coupled model, and generating first intermediate data and a first loss function using the SimAM attention mechanism, generative adversarial network, and conditional variational autoencoder embedded in the data augmentation sub-model, includes: The original sample training set is input into the encoder for reparameter sampling to obtain the first training parameters; The first training parameters are input into the decoder for training to obtain the second training parameters; Based on the second training parameters, the reconstruction loss function and the KL divergence loss function are determined, and the second training parameters are input into the discriminator to obtain the first intermediate data and the adversarial loss function. The sum of the reconstruction loss function, the KL divergence loss function, and the adversarial loss function is determined as the first loss function.

4. The method for predicting the compressive strength of cementitious bodies as described in claim 3, characterized in that, The step of inputting the original sample training set into the data augmentation sub-model of the initial adversarial coupled model, and generating first intermediate data and a first loss function using the SimAM attention mechanism, generative adversarial network, and conditional variational autoencoder embedded in the data augmentation sub-model, includes: The true target value and conditional features are determined based on the original sample training set, and the true target value and conditional features are input into the encoder and processed by the Adam optimizer and SimAM attention module to obtain the latent distribution parameters of the original sample training set. The latent distribution parameters are resampled to obtain the first training parameters, which include latent variables and conditional features; The first training parameters are input into the decoder, and after training using the ReLU activation function, Adam optimizer and SimAM attention module, the second training parameters are output. The second training parameters and the conditional features are input into the discriminator to generate an adversarial loss function.

5. The method for predicting the compressive strength of cementitious bodies as described in claim 4, characterized in that, The step of inputting the second training parameters and the conditional features into the discriminator to generate an adversarial loss function includes: The second training parameters and the conditional features are input into the discriminator to generate the discriminator loss function; Backpropagation is performed based on the discriminator loss function to update the weights and biases of the discriminator.

6. The method for predicting the compressive strength of a cementitious body as described in claim 1, characterized in that, The step of inputting the first intermediate data into the intensity prediction submodule of the initial adversarial coupling model, and generating initial prediction results based on the genetic algorithm and BP neural network embedded in the intensity prediction submodule, includes: For each individual in the first intermediate data, fitness screening is performed based on its chromosome information and a genetic algorithm to determine the appropriate network structure and hyperparameters; the chromosome information includes the hidden layer structure, activation function name, and learning rate.

7. The method for predicting the compressive strength of cementitious bodies as described in claim 1, characterized in that, The step of verifying the initial prediction result with the original sample validation set to determine the second loss function includes: The initial prediction results and the original sample validation set are processed using the ReLU activation function and the Adam optimizer to obtain the second loss function during forward propagation; The weights and biases of the BP neural network are updated based on the second loss function.

8. A device for predicting the compressive strength of a cementitious body, characterized in that, include: The module includes an impact analysis module, a first training module, a second training module, an iterative training module, and an intensity prediction module; among which, The impact analysis module is configured to perform compressive strength tests and grey correlation analysis on the sample cementitious body to determine the target impact factor. The first training module is configured to create an initial adversarial coupling model based on the target influencing factor, and input the original sample training set into the data augmentation sub-model of the initial adversarial coupling model, and use the SimAM attention mechanism, generative adversarial network and conditional variational autoencoder embedded in the data augmentation sub-model to generate first intermediate data and a first loss function; The second training module is configured to input the first intermediate data into the strength prediction submodule of the initial adversarial coupling model, generate an initial prediction result based on the genetic algorithm and BP neural network embedded in the strength prediction submodule, and verify the initial prediction result with the original sample validation set to determine the second loss function. The iterative training module is configured to iteratively train the parameters of each module in the data augmentation sub-model based on the first loss function and the second loss function until a preset iteration threshold is reached to obtain the target adversarial coupling model. The strength prediction module is configured to input the data of the cementitious body to be tested into the target adversarial coupling model to obtain the predicted compressive strength of the cementitious body to be tested.

9. An electronic device, characterized in that, It includes a processor and a memory; the memory has a computer program stored thereon, wherein the computer program, when executed by the processor, implements the method for predicting the compressive strength of cementitious materials as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, It stores a computer program, wherein the computer program, when executed by a processor, implements the method for predicting the compressive strength of cementitious materials as described in any one of claims 1 to 7.

Images