Method and device for predicting fluidity of alkali-activated mortar
By combining deep belief networks with the Grey Wolf algorithm, a flowability prediction model for alkali-activated mortar was constructed, which solved the problems of low efficiency and insufficient accuracy in traditional methods, and achieved high-precision flowability prediction, supporting the engineering application of alkali-activated mortar.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-26
- Publication Date
- 2026-03-13
AI Technical Summary
Traditional methods are inefficient and lack accuracy in predicting the fluidity of alkali-activated mortar, and are difficult to accurately handle complex nonlinear relationships, which affects their application in cast-in-place construction.
A flowability prediction model is constructed by optimizing hyperparameters using a deep belief network combined with the Grey Wolf algorithm. The model is trained using a training set and the flowability of the target alkali-activated mortar is predicted. By utilizing the multi-level feature extraction of the deep belief network and the optimization capability of the Grey Wolf algorithm, the problems of low efficiency and insufficient accuracy of traditional methods are solved.
It improves the prediction accuracy of the fluidity of alkali-activated mortar, provides theoretical basis and practical guidance, and offers a reliable prediction tool for mix proportion optimization and engineering application of alkali-activated mortar, while reducing experimental costs and time.
Smart Images
Figure CN121662234A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of mortar fluidity prediction, and more particularly to a method and apparatus for predicting the fluidity of alkali-activated mortar. Background Technology
[0002] Alkali-activated mortar (AAM), as an environmentally friendly building material, utilizes industrial byproducts such as fly ash, slag, and silica fume. Through an alkali activator, it generates a clinker-free cementitious material, exhibiting superior mechanical properties, durability, and a significantly reduced carbon footprint compared to traditional ordinary Portland cement concrete (OPCC). However, the rapid setting and poor flowability retention of single-component AAM limit its application in cast-in-place construction. The flowability of AAM is influenced by various factors, including the type and source of precursors, the type and chemical composition of solid activators, and the type and dosage of admixtures. The complex nonlinear relationships between these parameters make traditional experimental methods costly, inefficient, and time-consuming in flowability prediction.
[0003] In recent years, the application of artificial intelligence technology in AAM performance prediction has received widespread attention. Machine learning methods, such as support vector machines (SVM), artificial neural networks (ANN), and random forests (RF), have been used to predict the liquidity of AAM; however, these methods are greatly affected by the selection of input variables and the setting of model hyperparameters when dealing with high-dimensional, nonlinear data, resulting in limited prediction accuracy and generalization ability. Summary of the Invention
[0004] The following is an overview of the subject matter described in detail herein. This overview is not intended to limit the scope of the claims.
[0005] The main objective of this disclosure is to propose a method and apparatus for predicting the fluidity of alkali-activated mortar, which can accurately extract the complex nonlinear relationship between various parameters of alkali-activated mortar and its fluidity, thereby improving the prediction accuracy of the fluidity of alkali-activated mortar.
[0006] A first aspect of this application provides a method for predicting the fluidity of alkali-activated mortar, the method comprising: In response to the fluidity prediction command of alkali-activated mortar, determine the training set of alkali-activated mortar; A deep belief network is preset, and the hyperparameters of the deep belief network are determined; The training samples in the training set are input into the deep belief network to obtain the predicted fluidity value corresponding to the training samples output by the deep belief network. Based on the predicted fluidity value, the hyperparameters are optimized using the Grey Wolf algorithm, and the deep belief network is updated based on the optimized hyperparameters to obtain the trained deep belief network. The trained deep belief network is used as a fluidity prediction model that can predict the fluidity of alkali-activated mortar. Determine the target data for the target alkali-activated mortar, and input the target data into the fluidity prediction model to obtain the fluidity of the target alkali-activated mortar output by the fluidity prediction model.
[0007] This embodiment provides a method for predicting the fluidity of alkali-activated mortar, which has at least the following beneficial effects: This method constructs a deep belief network and combines it with the Grey Wolf algorithm to optimize hyperparameters. It then uses the training set to train the model and predict the fluidity of the target alkali-activated mortar. This solves the problems of low efficiency and insufficient prediction accuracy of traditional methods. It can accurately extract the complex nonlinear relationship between various parameters of alkali-activated mortar and its fluidity, improve the prediction accuracy of the fluidity of alkali-activated mortar, and provide theoretical basis and practical guidance for mix proportion optimization and engineering application of alkali-activated mortar.
[0008] A second aspect of this application provides a device for predicting the fluidity of alkali-activated mortar, the device comprising: The data acquisition module is used to respond to the fluidity prediction command of alkali-activated mortar and determine the training set of alkali-activated mortar. The hyperparameter acquisition module is used to preset the deep belief network and determine the hyperparameters of the deep belief network; The model training module is used to input training samples from the training set into the deep belief network, obtain the predicted fluidity value corresponding to the training sample output by the deep belief network, optimize the hyperparameters using the Grey Wolf algorithm based on the predicted fluidity value, and update the deep belief network based on the optimized hyperparameters to obtain the trained deep belief network. The trained deep belief network is used as a fluidity prediction model that can predict the fluidity of alkali-activated mortar. The model application module is used to determine the target data of the target alkali-activated mortar and input the target data into the fluidity prediction model to obtain the fluidity of the target alkali-activated mortar output by the fluidity prediction model.
[0009] A third aspect of this application provides an electronic device including at least one controller and a memory for communicatively connecting to the controller; the memory stores instructions executable by the at least one controller to cause the at least one controller to perform a fluidity prediction method for alkali-activated mortar as described above.
[0010] A fourth aspect of this application provides a computer-readable storage medium storing computer-executable instructions for causing a computer to perform a method for predicting the fluidity of alkali-activated mortar as described above.
[0011] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description
[0012] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0013] Figure 1 This is a flowchart illustrating an embodiment of the fluidity prediction method for alkali-activated mortar provided in this application; Figure 2 This is a correlation analysis matrix diagram of the AAM mortar flowability input parameters provided in the embodiments of this application; Figure 3 This is a diagram of the GWO-DBN model architecture provided in the embodiments of this application; Figure 4 This is a flowchart of GWO optimization of DBN hyperparameters provided in an embodiment of this application; Figure 5 This is a diagram illustrating the DBN hyperparameter optimization and iteration process for predicting the flowability of AAM mortar according to an embodiment of this application. Figure 6 This is a regression analysis graph of the GWO-DBN model mobility training set provided in one embodiment of this application; Figure 7 This is a regression analysis plot of the GWO-DBN model mobility test set prediction provided in another embodiment of this application; Figure 8 This is a graph showing the prediction performance error analysis of different models provided in the embodiments of this application; Figure 9 This is a structural diagram of an embodiment of an alkali-activated mortar fluidity prediction device provided in this application; Figure 10 This is a structural diagram of an embodiment of an electronic device provided in this application. Detailed Implementation
[0014] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0015] In the description of this application, the use of terms such as "first," "second," etc., is for the purpose of distinguishing technical features only and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated or the order of the technical features indicated.
[0016] In the description of this application, it should be understood that the orientation descriptions, such as up, down, etc., are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed or function in a specific orientation, and therefore should not be construed as a limitation of this application.
[0017] In recent years, the application of artificial intelligence technology in AAM performance prediction has received widespread attention. Traditional machine learning methods, such as Support Vector Machines (SVM), Artificial Neural Networks (ANN), and Random Forests (RF), have been used to predict the fluidity of AAM. However, these methods are greatly affected by the selection of input variables and the setting of model hyperparameters when dealing with high-dimensional, nonlinear data, resulting in limited prediction accuracy and generalization ability. In contrast, deep learning (DL) methods, through multi-level feature extraction and pattern recognition, can more effectively handle complex nonlinear relationships and are suitable for AAM fluidity prediction. Therefore, this application uses deep learning methods to achieve fluidity prediction of alkali-activated mortar and addresses some of the problems existing in the fluidity prediction of alkali-activated mortar using deep learning methods.
[0018] like Figure 1 One embodiment of this application provides a method for predicting the fluidity of alkali-activated mortar, the method comprising: Step S110: In response to the fluidity prediction command of alkali-activated mortar, determine the training set of alkali-activated mortar.
[0019] The training set refers to a dataset containing historical allocation data and corresponding liquidity indicators. Specifically, it can be constructed using laboratory test data or engineering field data, and is used to train subsequent models to identify the mapping relationship between input parameters and liquidity.
[0020] Step S120: Preset deep belief network and determine the hyperparameters of deep belief network.
[0021] Deep belief networks refer to neural network structures with multiple hidden layers. Specifically, they can be constructed using a stacked Restricted Boltzmann Machine approach, capturing complex nonlinear relationships through layer-by-layer feature extraction.
[0022] Step S130: Input the training samples in the training set into the deep belief network to obtain the predicted flowability value corresponding to the training samples output by the deep belief network. Based on the predicted flowability value, use the Grey Wolf algorithm to optimize the hyperparameters and update the deep belief network based on the optimized hyperparameters to obtain the trained deep belief network. Use the trained deep belief network as a flowability prediction model that can predict the flowability of alkali-activated mortar.
[0023] The Grey Wolf Algorithm Optimization refers to a swarm intelligence optimization method that simulates the hunting behavior of wolf packs. Specifically, it can adjust hyperparameters such as the number of hidden layer nodes and the learning rate through a position update mechanism to achieve adaptive optimization of the network structure.
[0024] Step S140: Determine the target data for the target alkali-activated mortar and input the target data into the fluidity prediction model to obtain the fluidity of the target alkali-activated mortar output by the fluidity prediction model.
[0025] First, deep belief networks (DPRNs) possess multi-level feature extraction capabilities, making them suitable for processing complex nonlinear data. However, DPRN performance is sensitive to hyperparameters, and traditional grid search methods are inefficient. Analysis of swarm intelligence algorithms reveals that the Grey Wolf algorithm excels in global optimization; therefore, this embodiment attempts to combine it with DPRNs to construct an intelligent optimization framework. Subsequent experimental verification determined that a training set-driven approach is used to build the prediction model, and automatic parameter tuning is achieved through iterative optimization.
[0026] Specifically, upon receiving a liquidity prediction request, the system first constructs a training set using historical data. During deep belief network initialization, the input layer dimension is set according to the number of input parameters, and the output layer uses a linear activation function to regress the predicted value. During training, each training sample undergoes feature transformation through the hidden layer, and the error between the predicted and actual values is calculated. The Grey Wolf algorithm maps hyperparameters to wolf pack positions, guiding the pack towards the optimal solution through a leader wolf, iteratively updating key parameters such as the number of hidden layer nodes and the learning rate. The optimized network parameters are fed back to the model for structural adjustments, ultimately forming a stable prediction model. In practical applications, the new allocation parameters are input into the trained model to output the predicted liquidity index.
[0027] Compared with existing technologies, this method constructs a deep belief network and combines it with the Grey Wolf algorithm to optimize hyperparameters. It then uses the training set to train the model and predict the fluidity of the target alkali-activated mortar. This solves the problems of low efficiency and insufficient prediction accuracy of traditional methods. It can accurately extract the complex nonlinear relationship between various parameters of alkali-activated mortar and its fluidity, improve the prediction accuracy of the fluidity of alkali-activated mortar, and provide theoretical basis and practical guidance for mix proportion optimization and engineering application of alkali-activated mortar.
[0028] In some embodiments of this application, the deep belief network includes an input layer, a first hidden layer, a second hidden layer, and an output layer. The number of neurons in the input layer is related to the number of input parameters in the training samples. The output layer includes one neuron and uses a linear activation function. The relationship between the number of neurons in the input layer and the number of input parameters means that the number of nodes in the input layer is dynamically adjusted according to the actual number of input variables. For example, when the training samples contain 13 parameters such as slag content and fly ash content, the number of neurons in the input layer can be set to 13 to ensure that the input data is completely transmitted to the hidden layer.
[0029] Hyperparameters include the number of neurons in the first hidden layer, the number of neurons in the second hidden layer, the number of training epochs for the deep belief network, and the learning rate. The number of training epochs controls the number of training iterations, for example, setting it to 500 to avoid underfitting or overfitting; the learning rate affects the weight update step size, for example, setting it to 0.001 to balance convergence speed and accuracy.
[0030] Specifically, the input layer receives training samples containing parameters such as slag content and water-to-solid ratio. The number of neurons in the input layer corresponds one-to-one with the input parameters to ensure data dimensionality matching. The first and second hidden layers extract high-dimensional features through layer-by-layer nonlinear transformations. For example, the first hidden layer uses the Sigmoid function to capture the interaction between input variables, and the second hidden layer uses the ReLU function to enhance the expression of sparse features. The output layer neurons directly map the weighted sum of the hidden layer outputs to the fluidity prediction value. The linear activation function avoids nonlinear compression of the regression results. During hyperparameter optimization, for example, by adjusting the number of neurons in the first hidden layer to 50-100, the second hidden layer to 20-50, the number of training epochs to 300-800, and the learning rate to 0.0005-0.005, the model's adaptability to different data distributions can be improved.
[0031] Compared with existing technologies, this embodiment utilizes a deep belief network to achieve hierarchical feature extraction through a dual hidden layer structure. For example, the first hidden layer learns the interaction between raw material ratios and activator parameters, while the second hidden layer captures complex patterns such as time dependence, significantly enhancing the modeling capability for high-dimensional nonlinear data. This embodiment optimizes the matching degree between model complexity and data features by dynamically adjusting hyperparameter combinations. It solves the problem of insufficient modeling of the multi-factor coupling effect of alkali-activated mortar fluidity in traditional models. Targeted optimization of hyperparameters further improves the model's prediction accuracy. For example, adjusting the learning rate avoids the gradient vanishing problem, and optimizing the training epochs prevents overfitting, thereby improving the model's prediction stability and accuracy in diverse alkali-activated mortar ratio scenarios.
[0032] In some embodiments of this application, training samples from the training set are input into a deep belief network to obtain a predicted mobility level corresponding to the training samples output by the deep belief network, including: Training samples from the training set are input into a deep belief network. The network is trained layer by layer using a contrastive divergence algorithm to train the first and second hidden layers, optimizes the weights based on the energy function and joint probability distribution, further optimizes the weights using a backpropagation algorithm, and adjusts the network parameters by minimizing the error between the predicted and actual mobility values corresponding to the training samples. The contrastive divergence algorithm is a fast training method approximating gradient descent, specifically implemented by updating the states of hidden layer neurons layer by layer. The weight adjustment is calculated by repeatedly sampling the difference in state distribution between the visible and hidden layers. The backpropagation algorithm is a method for adjusting network parameters based on the prediction error.
[0033] The energy function includes: (1); Energy value Visible unit state, It is a hidden unit state, and For bias terms, The weights are those between visible and hidden units; the energy function is a function used to characterize the joint distribution of neuron states in the visible and hidden layers.
[0034] The joint probability distribution includes: (2); For joint probability distribution, This is the partition function. The joint probability distribution refers to the joint probability model of the states of neurons in the visible and hidden layers.
[0035] Specifically, the training process begins with layer-by-layer pre-training of the hidden layers using a contrastive divergence algorithm. The weight parameters are updated based on the state differences between the visible and hidden layers, establishing a preliminary feature extraction capability. Subsequently, the joint probability distribution of the visible and hidden layers is calculated based on an energy function. The weights are adjusted to minimize the model's energy value, optimizing the network's representation of the input data. Further, a backpropagation algorithm is used to calculate the error gradient between the predicted and actual values, propagating the error signal backward along the network and simultaneously adjusting the weight parameters of each layer to reduce prediction bias. Through this multi-stage optimization mechanism, the network parameters gradually adapt to the data distribution characteristics of the mobility prediction task while retaining the feature extraction capability.
[0036] In some embodiments of this application, before using the Grey Wolf algorithm to optimize hyperparameters, the method further includes: Step S210: Map the hyperparameters to the position coordinates corresponding to the Gray Wolf algorithm, and construct the fitness calculation function of the Gray Wolf algorithm by minimizing the root mean square error of the training set, wherein the function for minimizing the root mean square error of the training set is expressed as: (3); in, and The first The actual value of each training sample and the corresponding predicted value of liquidity. These are the training samples for the training set.
[0037] In this context, mapping hyperparameters to location coordinates refers to converting parameters such as the number of hidden layer neurons, training epochs, and learning rate of the deep belief network into multidimensional coordinate points in the search space of the Grey Wolf algorithm. The fitness calculation function is constructed as the optimization objective of the root mean square error of the training set, which serves as the evaluation benchmark for individual fitness during the iteration process of the Grey Wolf algorithm.
[0038] Specifically, in the initialization phase of the Gray Wolf algorithm, hyperparameter combinations are mapped to coordinate points in a multi-dimensional space, with each coordinate dimension corresponding to a hyperparameter variable. The fitness calculation function is constructed based on the prediction error of the training set, ensuring that the algorithm consistently prioritizes reducing model error during the search process. During iteration, the position updates of individual gray wolves are guided by the position of the leader wolf. By calculating the distance between the individual and the leader wolf and adjusting the movement step size, the algorithm gradually approaches the optimal hyperparameter combination. This process avoids the subjectivity of manual parameter tuning and simultaneously explores the globally optimal solution through a swarm intelligence search mechanism.
[0039] This embodiment achieves automated global optimization of hyperparameters for deep belief networks, solving the problem of unstable prediction accuracy caused by traditional experimental methods relying on empirical parameter tuning. The fitness function is constructed based on the training set error, ensuring a direct correlation between the optimization process and the model's prediction performance, thereby improving the generalization ability and reliability of the alkali-activated mortar fluidity prediction model.
[0040] In some embodiments of this application, the Grey Wolf algorithm is used to optimize hyperparameters, including: Step S310: Calculate the fitness value of the wolves in the current iteration and select three leader wolves based on the fitness value; Step S320: Calculate the distance from each wolf in the population other than the three leader wolves to the three leader wolves; Step S330: Calculate the new position to move towards the three-headed leader wolf based on the distance; Step S340: Calculate the fitness value of each wolf in the population at the new location; Step S350: Update the three leader wolves based on the fitness value of each wolf in the population at the new location; Step S360, and so on, until the current iteration number reaches the preset maximum iteration number, and the position of the best wolf among the three leader wolves in the final output is taken as the optimal solution of the hyperparameters.
[0041] Specifically, the optimization process of the Gray Wolf algorithm simulates wolf pack hunting behavior to perform hyperparameter search. In each iteration, a deep belief network is first trained based on the current hyperparameter combination, and the prediction error is calculated. After the error is converted into a fitness value, three leader wolves are determined. Subsequently, the remaining individuals in the population adjust their positions according to their distance from the leader wolves. By introducing a random coefficient to control the step size, the movement is avoided from getting trapped in local optima. After the position is updated, the fitness value is re-evaluated, and the leader wolves are updated. This process is repeated until the preset number of iterations is reached. Finally, the hyperparameter combination corresponding to the position of the optimal wolf is selected as the optimization result.
[0042] Compared with existing technologies, this embodiment realizes the automated optimization of hyperparameters of deep belief networks, which solves the problems of low efficiency and easy getting trapped in local optima in traditional methods. The optimized hyperparameter combination can improve the accuracy of the model in predicting the fluidity of alkali-activated mortar, reduce the prediction deviation caused by improper parameter settings, thereby reducing the number of experimental verifications and shortening the material development cycle.
[0043] In some embodiments of this application, after obtaining the trained deep belief network, the method further includes: Step S410: Evaluate the accuracy of the updated deep confidence network output liquidity prediction value through regression analysis; the regression analysis evaluation indicators include at least one of the following: The evaluation criteria include the coefficient of determination, root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAS), mean gamma bias, mean Poisson bias, and overall performance index. Regression analysis evaluation quantifies the difference between the predicted model's output and the actual measured values using statistical methods. Specifically, the coefficient of determination (RMD) assesses the model's ability to explain data variation; its value ranges from 0 to 1, with a value closer to 1 indicating a higher model fit. The RMSE measures the average deviation between predicted and actual values, calculated as the square root of the mean squared errors; a smaller value indicates higher prediction accuracy. The MAS reflects the overall level of prediction deviation by calculating the average of relative errors and is suitable for comparing data with different dimensions. The mean gamma bias and mean Poisson bias evaluate the model's adaptability to asymmetric and count data, respectively. The overall performance index combines multiple evaluation indicators to form a comprehensive performance score.
[0044] Step S420: If the evaluation value of the regression analysis reaches the threshold, the trained deep belief network is used as a flowability prediction model that can predict the flowability of alkali-activated mortar.
[0045] Specifically, after the deep belief network is trained, the model needs to be validated using an independent test set. Samples from the test set are input into the network, and the predicted mobility levels are compared with experimental values. The model accuracy is calculated using preset evaluation metrics. For example, if the coefficient of determination is below 0.85 or the root mean square error exceeds a preset threshold, the network structure or algorithm needs to be readjusted or optimized. If all metrics meet the requirements, the model is deemed suitable for practical application. During the evaluation process, a single core metric can be selected as the criterion, or multiple metrics can be combined to construct a comprehensive evaluation system. For example, a weighted sum of the root mean square error and the mean absolute percentage error can be used to set a passing threshold.
[0046] In some embodiments of this application, the input parameters of the training samples include at least: slag content, fly ash content, fly ash type, silica fume content, silica fume type, activator content, activator modulus ratio, activator type, water-solid ratio, admixture content, admixture type, sand-binder ratio, and test time.
[0047] Through the above technical solutions, the prediction model established in this application can accurately quantify the impact of different raw material combinations on fluidity, significantly reducing prediction bias caused by neglecting key parameters. The complete multi-dimensional parameter system effectively solves the problem of insufficient generalization ability caused by the lack of features in traditional methods, providing a reliable prediction tool for fluidity control of complex-ratio alkali-activated mortars.
[0048] like Figures 2 to 8As shown in the figure, this embodiment proposes a method for predicting the flowability of alkali-activated mortar (AAM) based on Grey Wolf Optimized Deep Belief Network (GWO-DBN). This method optimizes the DBN hyperparameters using the GWO algorithm to construct a high-precision prediction model. It can accurately extract the complex nonlinear relationships between various AAM parameters and their flowability, improving prediction accuracy and reducing experimental costs. This provides a theoretical basis and practical guidance for mix proportion optimization and engineering applications of AAM. The specific steps of the method are as follows: Step S910, Data Acquisition and Processing; A training set for the fluidity of AAM mortar was collected from the experimental database, containing 203 training samples, each with 13 input parameters: The test parameters include slag content (Slag), fly ash content (FA), fly ash type (TFA), silica fume content (SF), silica fume type (TSF), activator content (AC), activator modulus ratio (MRA), activator type (TA), water-to-solid ratio (W / S), admixture content (AD), admixture type (AT), sand-to-binder ratio (S / B), test time (Time), and one output parameter: flowability (Flow).
[0049] Perform statistical analysis on the training set to calculate the maximum, minimum, mean, median, standard deviation, skewness, etc., to ensure the integrity of the data features.
[0050] Step S920: Construct a deep belief network (DBN); Input layer: Contains 13 neurons, corresponding to 13 input parameters, used to receive the mix proportion and test condition data of AAM mortar.
[0051] Hidden layers: These consist of two Restricted Boltzmann Machine (RBM) layers, using the Sigmoid activation function to extract low-level and high-level features, respectively. The first RBM layer (i.e., the first hidden layer) extracts preliminary features from the input data, while the second RBM layer (i.e., the second hidden layer) further abstracts these features into high-level feature representations.
[0052] Output layer: Contains 1 neuron, uses a linear activation function, and outputs the predicted mobility value.
[0053] DBN training is divided into two stages: Pre-training phase: The contrastive divergence algorithm is used to train the RBM layer by layer, and the weights are optimized based on the energy function and joint probability distribution. The energy function is as shown in formula (1) above, and the joint probability distribution is as shown in formula (2) above.
[0054] Fine-tuning phase: The backpropagation algorithm is used to optimize the global weights, and the network parameters are adjusted by minimizing the error between the predicted and actual values.
[0055] Step S930: Hyperparameter optimization based on GWO. The hyperparameters of DBN are optimized using the Grey Wolf Optimization (GWO) algorithm, including the number of neurons in the first hidden layer (HN1), the number of neurons in the second hidden layer (HN2), the number of training epochs, and the learning rate (LR). The optimization objective is to minimize the root mean square error (RMSE) of the training set, as shown in formula (3) above. The GWO algorithm simulates the hunting behavior of gray wolves, through... The three-tiered leadership coordinates the population search and configures its parameters, including population size and maximum number of iterations. The position update formula is as follows: Calculate the distance between the wolf and its prey: (4); in, Indicates the location of the prey. This is the wolf's current location. It is a coefficient vector, and its calculation formula is: ,in, It is a vector that is randomly generated within the interval [0,1].
[0056] Update the wolf's location: (5); in, It is another coefficient vector used to control the balance between exploration and development, and its calculation formula is: .in, The iterations decrease linearly from 2 to 0. It is a vector that is randomly generated within the interval [0,1].
[0057] based on The wolf calculates the middle position: (6); (7); (8); The final position is the average of the three: (9); Step S940, Model Training and Validation.
[0058] The DBN model was trained using optimized hyperparameters, and the prediction accuracy was evaluated by regression analysis. The metrics included the coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), mean gamma bias (MGD), mean Poisson bias (MPD), and overall performance index (OI).
[0059] The method provided in this application has at least the following beneficial effects: This method optimizes the hyperparameters of the DBN based on GWO, significantly improving the prediction accuracy of AAM mortar flowability, which is significantly better than traditional machine learning methods. Furthermore, it significantly improves optimization efficiency and reduces computational costs compared to traditional grid search, providing a scalable framework for modeling complex nonlinear problems.
[0060] The following is a set of experimental examples: (1) Acquisition and analysis of experimental data; 203 sets of AAM mortar flowability data were collected from the experimental database, including 13 input parameters: The following parameters were analyzed: slag content (Slag), fly ash content (FA), fly ash type (TFA), silica fume content (SF), silica fume type (TSF), activator content (AC), activator modulus ratio (MRA), activator type (TA), water-to-solid ratio (W / S), admixture content (AD), admixture type (AT), mortar-to-binder ratio (S / B), testing time, and output parameter flowability. Correlation analysis of each parameter is as follows: Figure 2 As shown in the table below, the statistical analysis results are as follows: the average slag content is 68.60%, the standard deviation is 10.57%, and the skewness is 2.31; the average fly ash content is 30.20%, and the skewness is -2.34; the average fluidity is 146.94 mm, the standard deviation is 25.09 mm, and the skewness is 0.28.
[0061] Table 1
[0062] (2) DBN model construction and initial configuration; Establish such as Figure 3 The DBN model shown contains 13 neurons in the input layer, two RBM hidden layers using the sigmoid activation function, and one neuron in the output layer using the linear activation function. The initial range of hyperparameters is: number of neurons in the first hidden layer [20, 80], number of neurons in the second hidden layer [20, 80], number of training epochs [1000, 3000], and learning rate [1.5, 3]. In the pre-training stage, the contrastive divergence algorithm is used to train the RBM layer by layer based on the energy function as shown in Equation (1) and the joint probability distribution as shown in Equation (2); in the fine-tuning stage, the backpropagation algorithm is used to optimize the weights.
[0063] (3) Hyperparameter optimization based on GWO; Figure 4The flowchart shown illustrates the GWO optimization process for DBN hyperparameters. The parameter configurations are: population size 20, maximum number of iterations 100, and the optimization range is shown in Table 2. The optimization objective is to minimize the RMSE. The optimization results show that the optimal hyperparameters are: HN1=26, HN2=32, Epochs=3000, LR=2.015. The optimization process is as follows: Figure 5 This shows that GWO converges quickly within 100 iterations, proving its efficiency.
[0064] Table 2
[0065] (4) Model training and performance validation; 70% of the data was used for model training, and 30% for model testing. The test set showed R² of 0.933, RMSE of 5.738 mm, MAE of 4.768 mm, MAPE of 0.033, MGD of 0.0017, MPD of 0.124, and OI of 0.935. Regression analyses for the training and test sets are shown below. Figure 6 and Figure 7 As shown, the predicted values are highly consistent with the actual values, with most data points within ±20% error range.
[0066] The method provided in this embodiment is compared with the predictive performance of Support Vector Machine (SVM), Gaussian Process Regression (GPR), Random Forest (RF), and Artificial Neural Network (ANN). The comparisons are made between SVM (R²=0.888, RMSE=8.116mm), RF (R²=0.888, RMSE=8.110mm), and ANN (R²=0.882, RMSE=8.384mm). The error analysis of each predictive model is as follows: Figure 8 As shown, GWO-DBN performs excellently, further confirming its prediction stability and low error.
[0067] like Figure 9 As shown in one embodiment of this application, an alkali-activated mortar fluidity prediction device is provided, the device comprising: The data acquisition module 1100 is used to respond to the fluidity prediction command of alkali-activated mortar and determine the training set of alkali-activated mortar; The hyperparameter acquisition module 1200 is used to preset the deep belief network and determine the hyperparameters of the deep belief network; The model training module 1300 is used to input the training samples in the training set into the deep belief network to obtain the predicted value of the fluidity of the training samples output by the deep belief network. Based on the predicted value of fluidity, the hyperparameters are optimized by the Grey Wolf algorithm, and the deep belief network is updated based on the optimized hyperparameters to obtain the trained deep belief network. The trained deep belief network is used as a fluidity prediction model that can predict the fluidity of alkali-activated mortar. The model application module 1400 is used to determine the target data of the target alkali-activated mortar and input the target data into the fluidity prediction model to obtain the fluidity of the target alkali-activated mortar output by the fluidity prediction model.
[0068] It should be noted that the alkali-activated mortar fluidity prediction device provided in this embodiment is based on the same inventive concept as the alkali-activated mortar fluidity prediction method described above. Therefore, the content of the alkali-activated mortar fluidity prediction method described above also applies to the content of the alkali-activated mortar fluidity prediction device in this embodiment, and will not be repeated here.
[0069] like Figure 10 One embodiment of this application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method for predicting the fluidity of alkali-activated mortar. The electronic device includes: At least one battery; At least one memory; At least one processor; At least one program; The program is stored in memory, and the processor executes at least one program to implement the above-described method for predicting the fluidity of alkali-activated mortar.
[0070] This electronic device can be any smart terminal, including mobile phones, tablets, personal digital assistants (PDAs), and in-vehicle computers.
[0071] The electronic devices according to embodiments of this application will now be described in detail.
[0072] The processor 1600 can be implemented using a general-purpose central processing unit (CPU), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this disclosure. The memory 1700 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 1700 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1700 and is called and executed by the processor 1600 to execute the alkali-activated mortar fluidity prediction method of the embodiments of this disclosure.
[0073] The input / output interface 1800 is used to implement information input and output. The communication interface 1900 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 2000 transmits information between various components of the device (e.g., processor 1600, memory 1700, input / output interface 1800, and communication interface 1900); The processor 1600, memory 1700, input / output interface 1800 and communication interface 1900 are connected to each other within the device via bus 2000.
[0074] This disclosure also provides a storage medium, which is a computer-readable storage medium storing computer-executable instructions for causing a computer to execute the above-described method for predicting the fluidity of alkali-activated mortar.
[0075] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
[0076] The embodiments described in this disclosure are for the purpose of more clearly illustrating the technical solutions of this disclosure and do not constitute a limitation on the technical solutions provided by this disclosure. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by this disclosure are also applicable to similar technical problems.
[0077] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this disclosure, and may include more or fewer steps than shown, or combine certain steps, or different steps.
[0078] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.
[0079] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.
[0080] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0081] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.
[0082] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.
[0083] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0084] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0085] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0086] The above is a detailed description of the preferred embodiments of this application. However, the embodiments of this application are not limited to the above-described implementation methods. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the embodiments of this application. All such equivalent modifications or substitutions are included within the scope defined by the claims of the embodiments of this application.
Claims
1. A method for predicting the fluidity of alkali-activated mortar, characterized in that, The method includes: Determine the training set for alkali-activated mortar; A deep belief network is preset, and the hyperparameters of the deep belief network are determined; The training samples in the training set are input into the deep belief network to obtain the predicted fluidity value corresponding to the training samples output by the deep belief network. Based on the predicted fluidity value, the hyperparameters are optimized using the Grey Wolf algorithm, and the deep belief network is updated based on the optimized hyperparameters to obtain the trained deep belief network. The trained deep belief network is used as a fluidity prediction model that can predict the fluidity of alkali-activated mortar. Determine the target data for the target alkali-activated mortar, and input the target data into the fluidity prediction model to obtain the fluidity of the target alkali-activated mortar output by the fluidity prediction model.
2. The method for predicting the fluidity of alkali-activated mortar according to claim 1, characterized in that, The deep belief network includes an input layer, a first hidden layer, a second hidden layer, and an output layer; the number of neurons in the input layer is related to the number of input parameters of the training samples; the output layer includes one neuron and uses a linear activation function. The hyperparameters include the number of neurons in the first hidden layer, the number of neurons in the second hidden layer, the number of training epochs of the deep belief network, and the learning rate.
3. The method for predicting the fluidity of alkali-activated mortar according to claim 2, characterized in that, The step of inputting training samples from the training set into the deep belief network to obtain the predicted liquidity level corresponding to the training samples output by the deep belief network includes: The training samples in the training set are input into the deep belief network to train the first hidden layer and the second hidden layer layer by layer based on the contrastive divergence algorithm, optimize the weights based on the energy function and joint probability distribution, optimize the weights based on the backpropagation algorithm, and adjust the network parameters of the deep belief network based on minimizing the error between the predicted value and the actual value of the mobility corresponding to the training sample. The energy function includes: ； Energy value Visible unit state, It is a hidden unit state, and For bias terms, The weights between visible and hidden units; The joint probability distribution includes: ； For joint probability distribution, This is the partition function.
4. The method for predicting the fluidity of alkali-activated mortar according to claim 3, characterized in that, Before optimizing the hyperparameters using the Grey Wolf algorithm, the method further includes: The hyperparameters are mapped to the position coordinates corresponding to the Grey Wolf algorithm, and a fitness calculation function for the Grey Wolf algorithm is constructed that minimizes the root mean square error of the training set, wherein the function that minimizes the root mean square error of the training set is expressed as: ； in, and The first The actual value of each training sample and the corresponding predicted value of liquidity. These are the training samples for the training set.
5. The method for predicting the fluidity of alkali-activated mortar according to claim 4, characterized in that, The optimization of the hyperparameters using the Grey Wolf algorithm includes: Calculate the fitness value of the wolves in the current iteration and select three leader wolves based on the fitness value; Calculate the distance from each wolf in the population other than the three leader wolves to the three leader wolves; Based on the distance, calculate the new position to move towards the three-headed leader wolf; Calculate the fitness value of each wolf in the population at the new location; Update the three leader wolves based on the fitness value of each wolf in the population at the new location; This process continues until the current iteration count reaches the preset maximum iteration count. The position of the optimal wolf among the three leader wolves in the final output is then taken as the optimal solution for the hyperparameters.
6. The method for predicting the fluidity of alkali-activated mortar according to claim 1, characterized in that, After obtaining the trained deep belief network, the method further includes: The accuracy of the updated liquidity prediction value output by the deep confidence network is evaluated using regression analysis; the indicators evaluated by the regression analysis include at least one of the following: Coefficient of determination, root mean square error, mean absolute error, mean absolute percentage error, mean gamma deviation, mean Poisson deviation, and overall performance index; If the evaluation value of the regression analysis reaches the threshold, the trained deep belief network is used as a flowability prediction model that can predict the flowability of alkali-activated mortar.
7. The method for predicting the fluidity of alkali-activated mortar according to claim 1, characterized in that, The input parameters of the training samples include at least: slag content, fly ash content, fly ash type, silica fume content, silica fume type, activator content, activator modulus ratio, activator type, water-to-solid ratio, admixture content, admixture type, sand-to-binder ratio, and test time.
8. A device for predicting the fluidity of alkali-activated mortar, characterized in that, The device includes: The data acquisition module is used to respond to the fluidity prediction command of alkali-activated mortar and determine the training set of alkali-activated mortar. The hyperparameter acquisition module is used to preset the deep belief network and determine the hyperparameters of the deep belief network; The model training module is used to input training samples from the training set into the deep belief network, obtain the predicted fluidity value corresponding to the training sample output by the deep belief network, optimize the hyperparameters using the Grey Wolf algorithm based on the predicted fluidity value, and update the deep belief network based on the optimized hyperparameters to obtain the trained deep belief network. The trained deep belief network is used as a fluidity prediction model that can predict the fluidity of alkali-activated mortar. The model application module is used to determine the target data of the target alkali-activated mortar and input the target data into the fluidity prediction model to obtain the fluidity of the target alkali-activated mortar output by the fluidity prediction model.
9. An electronic device, characterized in that, It includes at least one controller and a memory for communicatively connecting with the controller; the memory stores instructions executable by the at least one controller to cause the at least one controller to perform a method for predicting the fluidity of alkali-activated mortar as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions for causing a computer to perform a method for predicting the fluidity of alkali-activated mortar as described in any one of claims 1 to 7.