A method and device for predicting the creep coefficient of recycled aggregate concrete

Abstract

The application relates to a recycled aggregate concrete creep coefficient prediction method and device, the method constructs a creep coefficient prediction model according to a CNN-BiLSTM, then determines hyperparameters, optimizes the hyperparameters according to an enhanced sparrow search algorithm, inputs creep influencing factor data into the creep coefficient prediction model according to the optimized hyperparameters, obtains a creep coefficient prediction value, and trains the creep coefficient prediction model based on the creep coefficient prediction value and a measured value; in the application stage, target creep influencing factor data is input into the creep coefficient prediction model to obtain a target creep coefficient. Through combination of CNN-BiLSTM learning of a complex nonlinear mapping between influencing factors and creep, and optimization of model hyperparameters by using the enhanced sparrow search algorithm, the problems of strong coupling and high nonlinearity of multiple factors in RAC creep coefficient prediction are solved, and fast and accurate prediction of the RAC creep coefficient is realized.

Description

Technical Field

[0001] This application relates to the field of creep coefficient prediction technology for recycled aggregate concrete, and in particular to a method and apparatus for predicting the creep coefficient of recycled aggregate concrete. Background Technology

[0002] Recycled aggregate concrete (RAC) is a new type of concrete made by crushing, washing, and grading waste concrete to partially or completely replace natural aggregates. It is significant for promoting the resource utilization of construction waste and reducing carbon emissions. However, recycled aggregates often contain old mortar, resulting in high porosity, high water absorption, and low elastic modulus. This makes the long-term deformation performance of recycled aggregate concrete, especially creep, significantly higher than that of natural aggregate concrete. Creep is the deformation of concrete under continuous load over time. Excessive creep can lead to prestress loss and increased structural deflection, affecting the long-term safety and serviceability of the structure.

[0003] Accurate prediction of the creep behavior of recycled aggregate concrete is crucial for its widespread application in important structures. Currently, methods for predicting concrete creep mainly rely on empirical models or standard formulas derived from regression analysis of extensive experimental data, such as MC2010, ACI 209, GL2000, and the B3 model. These models are primarily based on databases of natural aggregate concrete, and often exhibit significant prediction biases when applied to recycled aggregate concrete, which has a more complex composition and microstructure. Furthermore, the creep of recycled aggregate concrete is complexly influenced by numerous factors, including the replacement rate of coarse recycled aggregate, aggregate water absorption, water-cement ratio, loading age, and environmental humidity, exhibiting high nonlinearity and coupling, further limiting the applicability of traditional models. 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 creep coefficient of recycled aggregate concrete. This method can solve the problem of strong coupling of multiple factors and high nonlinearity in the prediction of creep of recycled aggregate concrete, and achieve rapid and accurate prediction of the creep coefficient of RAC. It has the advantages of improving prediction accuracy and enhancing engineering applicability.

[0006] A first aspect of this application provides a method for predicting the creep coefficient of recycled aggregate concrete, the method comprising:

[0007] Determine the data on the factors affecting the creep of recycled aggregate concrete and the corresponding measured values ​​of the creep coefficient; A creep coefficient prediction model is constructed based on a CNN-BiLSTM network, and the hyperparameters of the creep coefficient prediction model are determined. The hyperparameters are then optimized using an enhanced sparrow search algorithm to obtain the optimized hyperparameters. Based on the optimized hyperparameters, the creep influencing factor data are input into the creep coefficient prediction model to obtain the creep coefficient prediction value output by the creep coefficient prediction model. Backpropagation is performed based on the measured value and the predicted value of the creep coefficient until the trained creep coefficient prediction model is obtained. In response to the creep coefficient prediction command, acquire data on the target creep influencing factors of the target recycled aggregate concrete; The target creep influencing factor data are input into the trained creep coefficient prediction model to obtain the target creep coefficient output by the trained creep coefficient prediction model.

[0008] The method for predicting the creep coefficient of recycled aggregate concrete provided in this embodiment has at least the following beneficial effects: This method effectively solves the problem of strong coupling of multiple factors and high nonlinearity in the prediction of creep of recycled aggregate concrete by combining CNN-BiLSTM network to fully learn the complex nonlinear mapping between influencing factors and creep, and by using enhanced sparrow search algorithm to optimize model hyperparameters. It achieves fast and accurate prediction of RAC creep coefficient, and has the advantages of improving prediction accuracy and enhancing engineering applicability.

[0009] A second aspect of this application provides a device for predicting the creep coefficient of recycled aggregate concrete, the device comprising: The data acquisition module is used to determine the data of creep influencing factors of recycled aggregate concrete and the corresponding measured values ​​of creep coefficient; The hyperparameter optimization module is used to construct a creep coefficient prediction model based on the CNN-BiLSTM network, determine the hyperparameters of the creep coefficient prediction model, and optimize the hyperparameters according to the enhanced sparrow search algorithm to obtain the optimized hyperparameters. The model prediction module is used to obtain the predicted creep coefficient output by the creep coefficient prediction model by inputting the creep influencing factor data into the creep coefficient prediction model according to the optimized hyperparameters. The model optimization module is used to perform backpropagation based on the measured value of the creep coefficient and the predicted value of the creep coefficient until the trained creep coefficient prediction model is obtained. The instruction response module is used to respond to creep coefficient prediction instructions and obtain data on the target creep influencing factors of the target recycled aggregate concrete. The creep coefficient determination module is used to input the target creep influencing factor data into the trained creep coefficient prediction model to obtain the target creep coefficient output by the trained creep coefficient prediction model.

[0010] 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 method for predicting the creep coefficient of recycled aggregate concrete as described above.

[0011] 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 creep coefficient of recycled aggregate concrete as described above.

[0012] 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

[0013] 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.

[0014] Figure 1 This is a flowchart illustrating a method for predicting the creep coefficient of recycled aggregate concrete provided in an embodiment of this application. Figure 2 This is a heatmap of the Pearson correlation coefficient matrix of the input parameters provided in the embodiments of this application; Figure 3 This is a schematic flowchart illustrating the CNN-BiLSTM hybrid model architecture provided in the embodiments of this application; Figure 4 This is a flowchart of the hyperparameter optimization process for the enhanced sparrow search algorithm provided in the embodiments of this application; Figure 5(a) is a schematic diagram of the L2 regularization factor and the number of iterations provided in the embodiments of this application; Figure 5(b) is a schematic diagram of the initial learning rate and number of iterations provided in the embodiments of this application; Figure 5(c) is a schematic diagram of the number of first-layer units and the number of iterations provided in the embodiments of this application; Figure 5(d) is a schematic diagram of the number of second-layer units and the number of iterations provided in the embodiments of this application; Figure 5(e) is a schematic diagram of the number of third-layer units and the number of iterations provided in the embodiments of this application; Figure 5(f) is a schematic diagram of the discard factor and iteration number provided in the embodiments of this application; Figure 6(a) is a schematic diagram of the training set predicted values ​​and the training set measured values ​​provided in the embodiments of this application; Figure 6(b) is a schematic diagram of the predicted values ​​and measured values ​​of the test set provided in the embodiments of this application; Figure 7 This is a Taylor chart comparing the performance of different prediction models and design specifications provided in the embodiments of this application; Figure 8 These are box plots of prediction errors for different models and design specifications provided in the embodiments of this application; Figure 9 This is a global importance ranking graph of input parameters based on SHAP values ​​provided in the embodiments of this application; Figure 10 This is the SHAP diagram of the model interpretability analysis provided in the embodiments of this application; Figure 11 This is a schematic diagram of the structure of a device for predicting the creep coefficient of recycled aggregate concrete provided in an embodiment of this application; Figure 12 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0015] 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.

[0016] 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.

[0017] 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.

[0018] In predicting the creep coefficient of recycled aggregate concrete, existing empirical models are mainly based on databases of natural aggregate concrete. When applied to recycled aggregate concrete, due to differences in characteristics such as high porosity and high water absorption caused by the adhesion of old mortar to recycled aggregate, the model prediction bias exists. In particular, the creep of recycled aggregate concrete is affected by the nonlinear coupling of multiple factors such as the replacement rate of coarse recycled aggregate, aggregate water absorption rate, water-cement ratio, loading age, and environmental humidity. Traditional models are difficult to handle accurately, resulting in inconsistencies between the predicted results and the actual creep behavior, which affects the reliability of structural design.

[0019] For example, in the roof structure design of large public buildings, recycled aggregate concrete is used as the main load-bearing material. Furthermore, when predicting creep behavior, the MC2010 model is used for calculation. However, due to the differences in material properties of recycled aggregate concrete, the creep coefficient output by the model deviates from the measured value, resulting in an incorrect estimate of prestress loss. Under continuous load, the deflection of the structure increases, affecting the normal use function of the building.

[0020] If the above problems are not addressed, inaccurate creep prediction in the engineering application of recycled aggregate concrete will lead to insufficient structural design margins. This may result in excessive prestress loss, structural deformation exceeding the allowable range, and consequently affect the long-term safety and serviceability of the structure, increasing the risks of structural maintenance and repair.

[0021] like Figure 1 As shown in one embodiment of this application, a method for predicting the creep coefficient of recycled aggregate concrete is provided, the method comprising: Step S110: Determine the data of creep influencing factors and the corresponding measured values ​​of creep coefficient for recycled aggregate concrete.

[0022] In this embodiment, the creep influencing factors data may include the physicochemical properties of the concrete constituent materials, environmental conditions, and loading parameters, etc., which are mainly to provide input information reflecting the actual behavior of recycled aggregate concrete.

[0023] Step S120: Construct a creep coefficient prediction model based on the CNN-BiLSTM network, determine the hyperparameters of the creep coefficient prediction model, and optimize the hyperparameters using the enhanced sparrow search algorithm to obtain the optimized hyperparameters.

[0024] The CNN-BiLSTM network refers to a deep learning architecture that integrates convolutional neural networks and bidirectional long short-term memory networks. Its primary purpose is to extract local spatial features from input data and model time-series dependencies. The hyperparameter optimization process can employ intelligent optimization methods such as particle swarm optimization or genetic algorithms, thereby ensuring the efficiency and global convergence of the hyperparameter search.

[0025] Step S130: Based on the optimized hyperparameters, the creep influencing factor data are input into the creep coefficient prediction model to obtain the creep coefficient prediction value output by the creep coefficient prediction model.

[0026] Step S140: Backpropagation is performed based on the measured and predicted creep coefficient values ​​until a trained creep coefficient prediction model is obtained.

[0027] Backpropagation is performed based on the measured and predicted values ​​of the creep coefficient. Backpropagation refers to a training mechanism that adjusts the weights within the model through error gradients. For example, stochastic gradient descent or adaptive moment estimator optimizers can be used to achieve this. Its main purpose is to make the model gradually approximate the true creep law of recycled aggregate concrete.

[0028] Step S150: In response to the creep coefficient prediction command, obtain the target creep influencing factor data of the target recycled aggregate concrete.

[0029] Step S160: Input the target creep influencing factor data into the trained creep coefficient prediction model to obtain the target creep coefficient output by the trained creep coefficient prediction model.

[0030] This embodiment trains a deep learning prediction model based on measured data of recycled aggregate concrete and combines it with the Enhanced Sparrow Search (ESSA) algorithm to adjust the hyperparameters of the deep learning prediction model. This avoids the bias caused by existing empirical models relying on natural aggregate concrete databases, thereby solving the problems of large prediction deviations due to differences in material properties and the difficulty in accurately handling the nonlinear coupling effects of multiple factors in the prediction of creep coefficient of recycled aggregate concrete. Therefore, in response to creep coefficient prediction commands, the model acquires data on the target creep influencing factors and inputs this data into the trained model, enabling it to output a target creep coefficient that meets the actual engineering requirements.

[0031] In the implementation of this method, firstly, by collecting data on creep influencing factors of recycled aggregate concrete and their corresponding measured creep coefficient values ​​in actual engineering projects, a training dataset specifically for the characteristics of recycled materials is constructed, thereby avoiding the systematic bias caused by the reliance of traditional empirical models on natural aggregate concrete databases. The creep influencing factor data directly reflects the unique physical characteristics of recycled aggregates, such as high porosity and high water absorption, ensuring that both input and output are faithful to the real behavior of recycled materials. Then, a creep coefficient prediction model is constructed based on a CNN-BiLSTM network that integrates convolutional neural networks and bidirectional long short-term memory networks. This model extracts the spatial feature interaction relationships of multiple factors through convolutional layers, while using bidirectional LSTM layers to capture the temporal evolution dependence during the creep process, effectively handling the highly nonlinear coupling effects between multiple factors such as coarse recycled aggregate replacement rate and water-cement ratio. Then, an enhanced sparrow search algorithm is used to globally optimize the model's hyperparameters. This algorithm dynamically adjusts the search step size through an adaptive convergence factor and adaptively adjusts the number of scouts based on the iteration process, thereby efficiently exploring the high-dimensional parameter space and avoiding the pitfalls of traditional optimization methods that easily get trapped in local optima. This ensures that the hyperparameters accurately match the complex mechanism of creep in recycled aggregate concrete. Finally, the optimized hyperparameter configuration is applied to the model training stage. After inputting creep influencing factor data, a predicted creep coefficient is generated. Backpropagation is performed through the error feedback between the measured and predicted creep coefficient values ​​to dynamically adjust the model's internal weight parameters, allowing the model to gradually learn the true laws of creep in recycled aggregate concrete. When a creep coefficient prediction command is received, the creep influencing factor data of the target recycled aggregate concrete is obtained and input into the trained model, and a high-precision target creep coefficient is output, realizing end-to-end application from engineering parameters to prediction results.

[0032] The creep coefficient prediction method for recycled aggregate concrete provided in this embodiment effectively solves the problem of strong coupling and high nonlinearity of multiple factors in the creep prediction of recycled aggregate concrete by combining CNN-BiLSTM network to capture the nonlinear characteristics of creep influencing factors and using enhanced sparrow search algorithm to optimize model hyperparameters. It has the advantages of improving prediction accuracy and enhancing engineering applicability.

[0033] In some embodiments of this application, hyperparameters include: At least one of the following: L2 regularization factor, initial learning rate, number of units in each BiLSTM layer of the CNN-BiLSTM network, and dropout factor.

[0034] Among them, L2 regularization factor refers to the hyperparameter used to control the complexity of the model; initial learning rate is the learning step size control parameter in the initial stage of neural network training; the number of units in each BiLSTM layer in the CNN-BiLSTM network refers to the number of neurons in the hidden layer of the bidirectional long short-term memory network, which can be flexibly set according to the dimension and complexity of the input data, with the aim of reasonably allocating the model's expressive power, effectively capturing the nonlinear coupling relationship between creep influencing factors, and avoiding redundant computational resources; dropout factor refers to the probability parameter of randomly masking neuron connections during training.

[0035] This embodiment, by clearly defining the specific components of hyperparameters, enables the Enhanced Sparrow Search algorithm to precisely focus on the core parameters affecting model performance. The L2 regularization factor constrains the weight update magnitude during training, effectively addressing the high noise characteristics of recycled aggregate concrete data; the initial learning rate dynamically adjusts the learning step size based on the training loss, ensuring a smooth and efficient convergence process; the number of units in the BiLSTM layer is rationally configured to optimize network capacity, fully modeling the complex interactions between creep-influencing factors; and the dropout factor introduces randomness during iteration, enhancing the model's adaptability to data uncertainty. These collaborative optimization mechanisms of hyperparameters enable the Enhanced Sparrow Search algorithm to efficiently select the optimal parameter combination, significantly improving the adaptability and reliability of the creep coefficient prediction model.

[0036] This embodiment enables the hyperparameter optimization process to precisely focus on the core parameters affecting model performance, avoiding interference from redundant parameters and improving the adaptability and reliability of the creep coefficient prediction model. Specifically, it effectively controls model complexity to prevent overfitting, dynamically adjusts the learning process to ensure convergence stability, rationally configures network capacity to capture the complex coupling relationships between creep influencing factors, and enhances the model's robustness to data anomalies, thereby improving the accuracy and training efficiency of creep coefficient prediction for recycled aggregate concrete.

[0037] In some embodiments of this application, the fitness function in the enhanced sparrow search algorithm is: ; in, For fitness value, This represents the sample size in the data on creep influencing factors. For the first Measured values ​​of creep coefficient for each sample. For the first Predicted creep coefficient values ​​for each sample.

[0038] The fitness function is a mathematical index used to quantitatively evaluate the overall fitting quality of a creep coefficient prediction model. Its purpose is to transform prediction errors into comparable numerical benchmarks, providing an objective quantitative basis for hyperparameter optimization and avoiding biases caused by subjective experience. Specifically, this embodiment defines the fitness function explicitly as a method for calculating the statistical error between measured and predicted values ​​based on all training samples. This allows the enhanced sparrow search algorithm to objectively evaluate the performance of different hyperparameter combinations in each iteration based on the fitness value. The enhanced sparrow search algorithm dynamically adjusts the position update strategy of individual sparrows according to the fitness value, prioritizing hyperparameter combinations with lower fitness values ​​and ensuring coverage of all training data during iteration. This effectively avoids local optima traps caused by data subset selection, while simultaneously enhancing the model's generalization ability to the creep characteristics of recycled aggregate concrete, enabling the hyperparameter optimization process to converge efficiently to the global optimum.

[0039] This embodiment achieves objectification and efficiency in the hyperparameter optimization process, enabling the creep coefficient prediction model to more accurately capture the complex nonlinear laws of creep behavior in recycled aggregate concrete, significantly improving prediction accuracy and reliability, and effectively solving the problem of blind optimization caused by the lack of evaluation standards in the prediction of creep coefficient of recycled aggregate concrete.

[0040] In some embodiments of this application, the formula for calculating the adaptive convergence factor in the enhanced sparrow search algorithm includes: ; in, As the initial weighting factor, A random number that is uniformly distributed in the range [0,1]. This represents the current iteration number; The formula for updating the producer's position during the iteration process in the enhanced sparrow search algorithm is adjusted as follows: ; in, For the first Only sparrows in the first The position of the dimension It is a random number. The maximum number of iterations, For random numbers that follow a normal distribution, It is a matrix of all ones. This is a warning value. This is the safety threshold.

[0041] In this embodiment, the adaptive convergence factor refers to a key parameter that dynamically adjusts the search range of the algorithm according to the iteration process. The purpose is to promote global exploration capability through a larger search range in the early stage of iteration and improve local development accuracy through fine shrinkage in the later stage of iteration. The initial weight factor can be understood as a basic constant that controls the initial search intensity of the algorithm, and its value range can be dynamically set according to the dimension of the optimization problem. Among them, the random number that follows a normal distribution in the producer position update formula refers to the random variable that introduces environmental disturbances. The synergistic effect of the warning value and the safety threshold refers to the dynamic response mechanism based on environmental risk assessment.

[0042] This embodiment achieves dynamic algorithm optimization through the collaborative design of an adaptive convergence factor and a producer position update formula: the adaptive convergence factor dynamically adjusts its decay rate based on the current iteration number, and combines a random perturbation term to break the fixed convergence pattern, maintaining a relatively large value in the early stages of iteration. The value is used to expand the solution space exploration range, and gradually decreases as iterations progress. The system focuses on potentially optimal regions, thus systematically balancing global exploration and local development capabilities. Simultaneously, the producer location update formula incorporates normally distributed random numbers and an early warning mechanism; when environmental monitoring values ​​exceed a safety threshold, the system triggers an alert. Enhance the strength of random perturbations and utilize an all-one matrix. Ensuring consistency across multiple dimensions allows location updates to both respond to environmental risks and utilize the maximum number of iterations. The algorithm guides the search direction and avoids premature convergence caused by a fixed update strategy. This embodiment uses the above mechanisms to enable the algorithm to adaptively adjust its search behavior according to the iteration state during the hyperparameter optimization of the creep coefficient prediction model, effectively addressing the nonlinear coupling characteristics of creep influencing factors.

[0043] This embodiment effectively solves the problem of exploration-development imbalance caused by fixed convergence factors, avoids premature convergence caused by the lack of dynamic adaptability of the producer position update mechanism, and significantly improves the convergence efficiency and global search capability of the enhanced sparrow search algorithm in the hyperparameter optimization process of the creep coefficient prediction model of recycled aggregate concrete. Thus, it provides a reliable adaptive optimization mechanism for dealing with the high nonlinearity and coupling of creep influencing factors.

[0044] In some embodiments of this application, the adaptive adjustment of the number of scouts in the enhanced sparrow search algorithm during the iteration process includes: ; in, and These represent the lower and upper limits for the number of scouts, respectively. This is for the integer operation.

[0045] Among them, the number of scouts refers to the number of individuals responsible for environmental scouting and area exploration in the enhanced sparrow search algorithm; the lower limit is the minimum allowed value of the number of scouts; the upper limit is the maximum allowed value of the number of scouts; the current iteration number refers to the number of iterations currently being executed by the algorithm, which can be recorded by an internal counter, and its purpose is to reflect the stage characteristics of the optimization process; the maximum number of iterations refers to the upper limit of the total number of iterations preset by the algorithm; and the rounding operation refers to the mathematical operation of converting continuous numerical values ​​into integers.

[0046] Specifically, in each iteration, the enhanced sparrow search algorithm first obtains the current iteration number and the preset maximum iteration number, then substitutes them into the formula to calculate the number of scouts, and rounds the result to obtain an integer to determine the number of individuals acting as scouts in this iteration. In the early stages of the iteration, Approaching 1 brings the number of scouts close to the upper limit, promoting global exploration. As the number of iterations increases, this value decreases, and the number of scouts smoothly declines to the lower limit, strengthening local exploration. The setting of the lower and upper limits constrains the range of the number, ensuring the stability of the algorithm at different iteration stages. The rounding operation converts continuous calculation results into integers, adapting to the algorithm's execution requirements. Overall, this mechanism dynamically adjusts the number of scouts through iteration progress information, enabling the algorithm to enhance global coverage in the early stages of optimization to avoid local optima, and focus on high-potential regions in the convergence phase to improve optimization accuracy, thus forming a complete adaptive optimization process.

[0047] This embodiment effectively improves the global exploration capability and local development efficiency of hyperparameter optimization, reduces the risk of getting trapped in local optima, and accelerates the convergence speed, thereby improving the accuracy and robustness of the creep coefficient prediction model for recycled aggregate concrete.

[0048] In some embodiments of this application, creep influencing factors include at least one of the following: coarse recycled aggregate replacement rate, water absorption rate, water-cement ratio, compressive strength, test age, loading age, theoretical thickness, relative humidity, stress-intensity ratio at loading age, and elastic modulus at loading age.

[0049] Specifically, this embodiment constructs a comprehensive dataset of creep influencing factors by systematically integrating material composition parameters, mechanical performance indicators, time-dimensional variables, and environmental conditions. The replacement rate of coarse recycled aggregate and water absorption rate jointly characterize the microstructural properties of recycled aggregate; the water-cement ratio and compressive strength reflect the comprehensive impact of mix design on concrete density and strength; test age and loading age define time frames to distinguish creep development stages; theoretical thickness and relative humidity capture the mechanism by which structural geometry and the environment affect humidity migration; and the stress-intensity ratio and elastic modulus at loading age quantify the correlation between load effects and material stiffness. These factors work synergistically from different dimensions, enabling the prediction model to comprehensively consider the contribution of the unique microstructural characteristics of recycled aggregate concrete, such as high porosity and high water absorption, to creep, thereby effectively improving the characterization ability of complex nonlinear creep behavior.

[0050] This embodiment ensures that the data on creep influencing factors comprehensively cover the key influencing mechanisms of creep in recycled aggregate concrete, effectively improving the prediction model's ability to characterize complex nonlinear creep behavior and reducing prediction bias caused by missing or improperly selected input variables.

[0051] In some embodiments of this application, after obtaining the predicted creep coefficient value output by the creep coefficient prediction model by inputting creep influencing factor data into the creep coefficient prediction model according to the optimized hyperparameters, the method further includes: Step S170: Quantitatively analyze the contribution of each input creep influencing factor data to the predicted creep coefficient value based on SHAP. Step S180: Generate an importance ranking based on contribution and draw a summary diagram.

[0052] In practical applications, SHAP quantitative analysis refers to a game theory-based model interpretation method, which can be implemented using KernelSHAP or TreeSHAP algorithms. Its purpose is to accurately quantify the marginal contribution of each creep influencing factor to the predicted creep coefficient. The contribution can be understood as the specific impact of each input feature on the prediction result. Specifically, importance ranking refers to globally ranking creep influencing factors based on the absolute magnitude of their contribution. This can be achieved by calculating and ranking the average |SHAP| value, highlighting core influencing factors and allowing users to prioritize key parameters. Furthermore, the summary diagram can be understood as a visualization tool, implemented using functions from the SHAP library. Its purpose is to simplify complex numerical relationships into easily understandable graphical representations, helping users quickly grasp the model's decision-making logic.

[0053] Specifically, in this embodiment, after the creep coefficient prediction model outputs the predicted value, it first performs a SHAP quantitative analysis step. Based on the Shapley value principle, it calculates the contribution of each creep influencing factor data. This process considers the nonlinear coupling relationship between features and avoids ignoring the limitations of interaction. Subsequently, it generates an importance ranking based on the contribution, highlighting the core influencing factors through a global ranking mechanism. Finally, the analysis results are transformed into a summary diagram for visualization, intuitively presenting the distribution of feature importance and influence trends. These steps are executed sequentially and closely linked, forming a complete interpretability enhancement process, transforming the decision logic of the prediction model from a black box state into a transparent and traceable mechanism.

[0054] This embodiment makes the decision-making process of the creep coefficient prediction model for recycled aggregate concrete transparent, enabling a clear understanding of the specific contribution of each creep influencing factor to the prediction results. This effectively verifies the reliability of the model's decision-making and allows for targeted optimization of concrete mix proportions and construction parameters, significantly improving the interpretability and application value of the prediction method in practical engineering.

[0055] like Figures 2 to 10 As shown, for ease of understanding, one embodiment of this application provides a method for predicting the creep coefficient of recycled aggregate concrete, including the following steps: Step S910, Data preparation and preprocessing; Data from RAC creep tests were collected to construct a database containing 247 valid samples. Each sample includes measured values ​​of 10 creep influencing factors: coarse recycled aggregate replacement rate (RCA, %), water absorption rate (WA, %), water-cement ratio (w / c), compressive strength (CS, MPa), test age (t, days), loading age (ts, days), theoretical thickness (NS, mm), relative humidity (RH, %), loading age stress intensity ratio (SSR), loading age modulus of elasticity (EM, GPa), and the corresponding measured creep coefficient (CC).

[0056] Before model training, the following preprocessing steps were performed on all input data: Normalization: Z-score standardization is used to transform each input parameter so that its mean is 0 and its standard deviation is 1. The transformation formula is: ; in, Representing the Normalized data of the model input parameters. The original data representing this parameter. This represents the arithmetic mean of all data for this parameter. This represents the standard deviation of all data for this parameter.

[0057] After normalization, the 247 sets of data were randomly shuffled and divided into a training set (173 sets of samples) and a test set (74 sets of samples) in a 7:3 ratio.

[0058] The training set was used for model building and hyperparameter optimization, while the test set was used to evaluate the model's generalization performance. Before modeling, the Pearson correlation coefficient matrix among all 10 input parameters was calculated. Analysis confirmed that the absolute values ​​of the correlation coefficients between all parameter pairs were below 0.7, indicating that there was no serious multicollinearity problem among the selected features, making them suitable as independent predictor variables input into the model. This ensures the robustness and interpretability of the model from the outset.

[0059] Step S920: Construct a creep coefficient prediction model based on CNN-BiLSTM; Design a cascaded deep learning model architecture. The model takes a 10-dimensional vector of influencing factors as input. First, it performs deep feature extraction via a CNN module, which contains two one-dimensional convolutional layers and one max-pooling layer. Then, the extracted feature sequence is input into a BiLSTM module, which contains three bidirectional LSTM layers to capture the temporal context information before and after creep development. Each BiLSTM layer is followed by a dropout layer to prevent overfitting. Finally, a fully connected layer outputs the predicted value of the creep coefficient. Details include: (1) Input layer: Receives a vector with a dimension of 10×1 (corresponding to data of 10 creep influencing factors), corresponding to 10 normalized input parameters; (2) Convolutional Neural Network Module: First convolutional layer, second convolutional layer, and max pooling layer, wherein: First convolutional layer: One-dimensional convolution is used, with 10 convolutional kernels of size 2×1. This layer is used to initially extract local patterns from the input features. The second convolutional layer also uses one-dimensional convolution, with 10 convolutional kernels of size 1×1. The role of 1×1 convolution is to perform cross-channel feature combination and dimensionality reduction, and further refine the features. Max pooling layer: used to downsample the feature map output by the second convolutional layer, retaining the most salient features while reducing the number of parameters, and enhancing the translation invariance of the model; (3) Bidirectional Long Short-Term Memory (BiLSTM) Network Module: This module consists of three bidirectional LSTM layers stacked sequentially. Each BiLSTM layer can learn contextual information in the feature sequence from front to back and from back to front simultaneously, thereby capturing the temporal dependencies in the creeping development process more comprehensively. A dropout layer is immediately connected after each BiLSTM layer, randomly dropping a certain proportion of neurons during training to prevent overfitting. The dropout rate is one of the hyperparameters to be optimized. The number of units in the three BiLSTM layers are key hyperparameters, determined by subsequent ESSA optimization. (4) Fully connected output layer: The feature sequence output from the last BiLSTM layer is integrated and mapped to the final predicted value through a fully connected layer. The tanh function is used as the activation function in this layer.

[0060] Step S930: Hyperparameter optimization based on the enhanced sparrow search algorithm; The enhanced sparrow search algorithm of this embodiment is used to automatically and globally optimize the core hyperparameters of the hybrid model constructed in step S920. The hyperparameters to be optimized include: L2 regularization factor, initial learning rate, number of units in the three BiLSTM layers, and dropout factor. The enhanced sparrow search algorithm effectively improves search efficiency and the ability to escape local optima by introducing an adaptive convergence factor and a dynamic scout number mechanism. The optimization process uses the root mean square error between the model's predicted values ​​and the measured values ​​as the fitness function.

[0061] (1) Enhance and improve the configuration of the sparrow search algorithm; An enhanced sparrow search algorithm is used to globally optimize the hyperparameters of the aforementioned hybrid model. The enhanced sparrow search algorithm makes two key improvements over the standard SSA: the introduction of an adaptive convergence factor. Its calculation formula is , A random number that is uniformly distributed in the range [0,1]. This represents the current iteration number. This factor is used to improve the producer's position update formula. An adaptive update strategy using the number of scouts (VSN) is employed to dynamically adjust the number of scouts as the iteration progresses, balancing global exploration with local development.

[0062] The algorithm parameters are set as follows: Population size: 30; Maximum number of iterations: 100; Safety threshold: 0.8; Producer ratio: 20%; Scout number range: 10% / 20%.

[0063] (2) Optimize the problem definition; This embodiment optimizes six hyperparameters, including the L2 regularization factor, initial learning rate, number of units in the three BiLSTM layers, and dropout factor. Fitness function: based on fitness values... Minimize as the objective The root mean square error (RMSE) is defined as the difference between the predicted creep coefficient and the measured value under 5-fold cross-validation. Cross-validation is used to more robustly evaluate the generalization performance of the hyperparameter configuration.

[0064] (3) Optimization process and results; The optimization process is as follows: Figure 4 As shown. (Attached) Figures 5(a) to 5(f) The iterative curves of each hyperparameter during the optimization process of the enhanced sparrow search algorithm are shown. It can be seen that all hyperparameters tend to stabilize after approximately 40 iterations, demonstrating the fast and stable convergence performance of the enhanced sparrow search algorithm. The optimal combination of hyperparameters obtained after optimization is shown below: Table 1

[0065] Step S940, Model Training and Application; The optimal hyperparameter combination, obtained by optimizing the enhanced sparrow search algorithm, is used to fully train the CNN-BiLSTM model on the training set, resulting in a final, stable model suitable for practical prediction. For new RAC mix proportions or engineering conditions, only the corresponding 10-dimensional parameters need to be input to quickly obtain the predicted creep coefficient value using this model.

[0066] Training environment: The model was implemented in MATLAB R2023a, utilizing its deep learning toolbox. The hardware platform was a laptop equipped with an Intel i7-1265U processor and 16GB of RAM, with GPU acceleration for the training process.

[0067] Training configuration: The CNN-BiLSTM model was trained on the training set (173 samples) using optimized hyperparameters.

[0068] Prediction: After training, the model weights are fixed. For any new RAC mix ratio or working condition, simply input the corresponding 10 parameters into the model after the same normalization process to instantly obtain the predicted value of its creep coefficient.

[0069] Step S950, Model Validation and Interpretability Analysis: The performance of the trained model was comprehensively evaluated on the test set. Multi-metric comparisons were conducted with various classic machine learning models such as Gaussian Process Regression (GPR), Ensemble Decision Tree (EDT), Support Vector Machine (SVM), and Artificial Neural Network (ANN), as well as current design specifications such as MC2010 and JTG3362, to verify the superiority of this application. Furthermore, the SHAP interpretable artificial intelligence framework was used to interpret the model, quantitatively analyze the contribution of each input feature to the prediction results, generate an importance ranking, and draw a summary diagram to reveal its nonlinear influence, transforming the model's "black box" predictions into understandable engineering knowledge. Specifically: (1) Basic performance; The trained model was evaluated on a test set (74 samples). See attached. Figures 5(a) to 5(f) The scatter plot showing predicted and measured values ​​indicates that, in both the training and test sets, the data points are closely distributed. On both sides of the diagonal. Quantitative evaluation shows that the model's coefficient of determination on the training set... The mean square error (MSE) reached 0.9960, and 0.9728 on the test set. At the same time, the MSE and mean absolute error remained at extremely low levels, demonstrating the model's excellent prediction accuracy and generalization ability.

[0070] (2) Comparative analysis; To highlight the superiority of this embodiment, a comprehensive comparison was made with four classic machine learning models (such as Gaussian Process Regression (GPR), Ensemble Decision Tree (EDT), Support Vector Machine (SVM), and Artificial Neural Network (ANN)) and four mainstream international design standards (MC2010, JTG3362, AS3600, GL2000). Taylor diagram analysis (as attached) Figure 7 The diagram illustrates the degree of similarity between the predicted results of each model and the measured values ​​across three dimensions: standard deviation, correlation coefficient, and centered root mean square error. It is evident that the model proposed in this embodiment is closest to the "Real" point, representing perfect prediction, and its overall performance is significantly superior to all comparative models and standards. Error distribution analysis (see attached diagram) Figure 8 The distribution of prediction errors for each method was further compared. The model error bin in this embodiment is the shortest and closest to the zero line, with a median of almost zero and very few outliers. This indicates that its prediction results not only have high average accuracy but also excellent stability and reliability.

[0071] Step S960: Model interpretability analysis and application guidance; The SHAP framework is used to interpret the trained optimal model in order to reveal its decision-making logic and provide guidance for engineering applications.

[0072] (1) Global importance analysis; Appendix Figure 9The global importance ranking of input parameters based on the average absolute value of SHAP is presented. The analysis results clearly show that the test age (t) is the absolute dominant factor affecting the prediction of the creep coefficient. The key factors following closely behind are: water-cement ratio (w / c), stress intensity ratio at loading age (SSR), and modulus of elasticity (EM). The recycled coarse aggregate replacement rate (RCA) also shows significant importance, confirming the essential influence of recycled aggregate on creep behavior.

[0073] (2) In-depth analysis of the influencing patterns; Appendix Figure 10 The SHAP plot provides a deeper analysis: Water-cement ratio (w / c): exhibits a complex nonlinear relationship. The plot shows that a lower water-cement ratio tends to produce a higher predicted creep coefficient (points on the right). This counterintuitive phenomenon may stem from complex mechanisms such as the performance of the interfacial transition zone between aggregate and cement paste in low-water-cement-ratio, high-strength concrete, and the relative changes in the aggregate's constraint effect on paste creep, highlighting the model's ability to capture complex nonlinear relationships. Recycled coarse aggregate replacement rate (RCA): High replacement rates are mainly concentrated on the right, clearly indicating that higher RCA content leads to a larger creep coefficient, providing a key control target for RAC creep.

[0074] This method has the following significant advantages: (1) Significantly improved prediction accuracy: The CNN-BiLSTM hybrid model constructed in this embodiment can automatically and efficiently learn the deep nonlinear and temporal dependencies between complex influencing factors of RAC and creep coefficient. Combined with the enhanced sparrow search algorithm to accurately optimize the model hyperparameters, the final model achieves prediction accuracy close to the measured value on the independent test set, and its performance is significantly better than other comparative machine learning models and existing design specifications.

[0075] (2) Input feature independence assurance and model robustness enhancement: Before model construction, the Pearson correlation coefficient matrix among all input parameters was calculated to confirm that the correlation between each variable was less than 0.7, effectively avoiding multicollinearity and ensuring the independence and effectiveness of input information, thereby improving the model's interpretability and generalization potential from the source. Robustness analysis results show that as the amount of training data increases, the model's predictive performance and stability are enhanced simultaneously.

[0076] (3) Verification of prediction consistency and reliability: as attached Figure 8 As shown in the box plot, compared with traditional design specifications and benchmark machine learning models, the hybrid model proposed in this embodiment exhibits a more concentrated distribution of prediction errors, a median closer to zero, and fewer outliers. This indicates that the prediction results of this embodiment not only have high average accuracy but also excellent stability and consistency, making them more reliable in practical engineering applications.

[0077] (4) Mechanism insight and design guidance capability: This embodiment goes beyond the traditional "black box" prediction. Through SHAP interpretability analysis, it can clearly reveal the specific influence law and importance ranking of each input parameter. This links the model's predictive ability with the physical mechanism, providing an intuitive and quantitative control direction and key parameter range for the design of low creep RAC.

[0078] like Figure 11 As shown in one embodiment of this application, a device for predicting the creep coefficient of recycled aggregate concrete is provided. The device includes: The data acquisition module 1001 is used to determine the data of creep influencing factors of recycled aggregate concrete and the corresponding measured values ​​of creep coefficient.

[0079] The hyperparameter optimization module 1002 is used to construct a creep coefficient prediction model based on the CNN-BiLSTM network, determine the hyperparameters of the creep coefficient prediction model, and optimize the hyperparameters according to the enhanced sparrow search algorithm to obtain the optimized hyperparameters.

[0080] The model prediction module 1003 is used to obtain the predicted creep coefficient output by the creep coefficient prediction model by inputting the creep influencing factor data into the creep coefficient prediction model based on the optimized hyperparameters.

[0081] The model optimization module 1004 is used to perform backpropagation based on the measured and predicted values ​​of the creep coefficient until a trained creep coefficient prediction model is obtained.

[0082] The instruction response module 1005 is used to respond to the creep coefficient prediction instruction and obtain the target creep influencing factor data of the target recycled aggregate concrete.

[0083] The creep coefficient determination module 1006 is used to input the target creep influencing factor data into the trained creep coefficient prediction model to obtain the target creep coefficient output by the trained creep coefficient prediction model.

[0084] It should be noted that the recycled aggregate concrete creep coefficient prediction device provided in this embodiment and the recycled aggregate concrete creep coefficient prediction method described above are based on the same inventive concept. Therefore, the content of the recycled aggregate concrete creep coefficient prediction method described above is also applicable to the content of the recycled aggregate concrete creep coefficient prediction device in this embodiment, and will not be repeated here.

[0085] like Figure 12 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 creep coefficient of recycled aggregate concrete. 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 creep coefficient of recycled aggregate concrete.

[0086] This electronic device can be any smart terminal, including mobile phones, tablets, personal digital assistants (PDAs), and in-vehicle computers.

[0087] The electronic devices according to embodiments of this application will now be described in detail.

[0088] 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 called and executed by the processor 1600 to execute a method for predicting the creep coefficient of recycled aggregate concrete according to an embodiment of this disclosure.

[0089] 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.

[0090] 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 creep coefficient of recycled aggregate concrete.

[0091] 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.

[0092] 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.

[0093] 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.

[0094] 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.

[0095] 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.

[0096] 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.

[0097] 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.

[0098] 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.

[0099] 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.

[0100] 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.

[0101] 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.

[0102] 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 creep coefficient of recycled aggregate concrete, characterized in that, The method includes: Determine the data on the factors affecting the creep of recycled aggregate concrete and the corresponding measured values ​​of the creep coefficient; A creep coefficient prediction model is constructed based on a CNN-BiLSTM network, and the hyperparameters of the creep coefficient prediction model are determined. The hyperparameters are then optimized using an enhanced sparrow search algorithm to obtain the optimized hyperparameters. Based on the optimized hyperparameters, the creep influencing factor data are input into the creep coefficient prediction model to obtain the creep coefficient prediction value output by the creep coefficient prediction model. Backpropagation is performed based on the measured value and the predicted value of the creep coefficient until the trained creep coefficient prediction model is obtained. In response to the creep coefficient prediction command, acquire data on the target creep influencing factors of the target recycled aggregate concrete; The target creep influencing factor data are input into the trained creep coefficient prediction model to obtain the target creep coefficient output by the trained creep coefficient prediction model.

2. The method for predicting the creep coefficient of recycled aggregate concrete according to claim 1, characterized in that, The hyperparameters include at least one of the following: L2 regularization factor, initial learning rate, number of units in each BiLSTM layer of the CNN-BiLSTM network, and dropout factor.

3. The method for predicting the creep coefficient of recycled aggregate concrete according to claim 2, characterized in that, The fitness function in the enhanced sparrow search algorithm is: ； in, For fitness value, This represents the sample size in the data on creep influencing factors. For the first Measured values ​​of creep coefficient for each sample. For the first Predicted creep coefficient values ​​for each sample.

4. The method for predicting the creep coefficient of recycled aggregate concrete according to claim 3, characterized in that, The formula for calculating the adaptive convergence factor in the enhanced sparrow search algorithm includes: ； in, As the initial weighting factor, A random number that is uniformly distributed in the range [0,1]. This represents the current iteration number; The formula for updating the producer's position during the iteration process in the enhanced sparrow search algorithm is adjusted as follows: ； in, For the first Only sparrows in the first The position of the dimension It is a random number. The maximum number of iterations, For random numbers that follow a normal distribution, It is a matrix of all ones. This is a warning value. This is the safety threshold.

5. The method for predicting the creep coefficient of recycled aggregate concrete according to claim 4, characterized in that, The adaptive adjustment process of the number of scouts in the enhanced sparrow search algorithm during iteration includes: ； in, and These represent the lower and upper limits for the number of scouts, respectively. This is for the integer operation.

6. The method for predicting the creep coefficient of recycled aggregate concrete according to claim 1, characterized in that, The creep influencing factors data include at least one of the following: coarse recycled aggregate replacement rate, water absorption rate, water-cement ratio, compressive strength, test age, loading age, theoretical thickness, relative humidity, stress-intensity ratio at loading age, and elastic modulus at loading age.

7. The method for predicting the creep coefficient of recycled aggregate concrete according to claim 1, characterized in that, After obtaining the predicted creep coefficient value output by the creep coefficient prediction model by inputting the creep influencing factor data into the creep coefficient prediction model based on the optimized hyperparameters, the method further includes: The contribution of each input creep influencing factor data to the predicted creep coefficient value was quantitatively analyzed based on SHAP. Based on the contribution level, an importance ranking is generated, and a summary diagram is drawn.

8. A device for predicting the creep coefficient of recycled aggregate concrete, characterized in that, The device includes: The data acquisition module is used to determine the data of creep influencing factors of recycled aggregate concrete and the corresponding measured values ​​of creep coefficient; The hyperparameter optimization module is used to construct a creep coefficient prediction model based on the CNN-BiLSTM network, determine the hyperparameters of the creep coefficient prediction model, and optimize the hyperparameters according to the enhanced sparrow search algorithm to obtain the optimized hyperparameters. The model prediction module is used to obtain the predicted creep coefficient output by the creep coefficient prediction model by inputting the creep influencing factor data into the creep coefficient prediction model according to the optimized hyperparameters. The model optimization module is used to perform backpropagation based on the measured value of the creep coefficient and the predicted value of the creep coefficient until the trained creep coefficient prediction model is obtained. The instruction response module is used to respond to creep coefficient prediction instructions and obtain data on the target creep influencing factors of the target recycled aggregate concrete. The creep coefficient determination module is used to input the target creep influencing factor data into the trained creep coefficient prediction model to obtain the target creep coefficient output by the trained creep coefficient 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 creep coefficient of recycled aggregate concrete 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 creep coefficient of recycled aggregate concrete as described in any one of claims 1 to 7.

Images