A method, device and equipment for proportioning design of LC3 concrete and a storage medium

Abstract

The application discloses a kind of LC3 concrete's proportioning design method, device, equipment and storage medium, it is related to machine learning technical field, comprising: generating the current proportioning population of LC3 concrete;Predict the compressive strength of each candidate proportioning in current proportioning population, and determine the performance evaluation index of each candidate proportioning;Compressive strength and performance evaluation index are filtered target candidate proportioning based on, and the current optimized proportioning population is obtained by optimizing target candidate proportioning;Current optimized proportioning population is determined as current proportioning population, and jump to the step of predicting the compressive strength of each candidate proportioning in current proportioning population, and determine the performance evaluation index of each candidate proportioning, until meet preset optimization termination condition, current optimized proportioning population is determined as target proportioning population, and the proportioning scheme of LC3 concrete is determined based on target proportioning population.The application can realize the LC3 concrete proportioning design of compressive strength prediction and multi-objective collaborative optimization.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to a method, apparatus, equipment and storage medium for mix design of LC3 concrete. Background Technology

[0002] LC3 concrete (LC3, Limestone Calcined Clay Cement) is a novel low-carbon and environmentally friendly cementitious material, prepared by replacing some of the cement clinker with calcined clay and limestone. Compared to traditional Portland cement, LC3 concrete exhibits significant advantages in reducing carbon dioxide emissions, saving energy, and improving concrete durability and crack resistance. In materials engineering practice, the compressive strength of the LC3 concrete system is mainly affected by a variety of factors, including the ratio of clinker to mineral admixtures, the activity of calcined clay, limestone filling and nucleation effects, water-cement ratio, sand-cement ratio, curing age, and curing temperature. These factors exhibit significant nonlinear coupling relationships, resulting in a multi-scale and multi-mechanism characteristic in the strength evolution process of LC3 concrete.

[0003] In terms of LC3 concrete mix design, most existing studies are still at the stage of compressive strength prediction. The results of compressive strength prediction are mainly used to evaluate the performance of existing mix proportions. There is a lack of a design mechanism that combines the compressive strength prediction model with mix proportion optimization. It is difficult to comprehensively consider the mutual constraints between multiple objectives such as compressive strength, low carbon emissions, material cost and physical constraints, and thus cannot achieve intelligent reverse design and optimal solution recommendation for LC3 concrete mix proportions.

[0004] In summary, how to establish an intelligent reverse design that takes into account both compressive strength prediction and multi-objective collaborative optimization in the mix design process of LC3 concrete is a technical problem that urgently needs to be solved. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method, apparatus, equipment, and storage medium for the mix design of LC3 concrete. This method enables the establishment of an intelligent reverse design that balances compressive strength prediction with multi-objective collaborative optimization during the LC3 concrete mix design process, thereby comprehensively balancing multiple interdependent engineering requirements such as material mechanical properties, carbon emissions, economic costs, and physical rationality. The specific solution is as follows: In a first aspect, this application provides a mix design method for LC3 concrete, including: Obtain the design variables of each mix proportion of LC3 concrete and the preset constraint range corresponding to each mix proportion design variable, and generate a current mix proportion population containing several candidate mix proportions based on each preset constraint range. Using a pre-constructed target compressive strength prediction model, the predicted compressive strength values ​​corresponding to each candidate mix proportion in the current mix proportion population are predicted, and several performance evaluation indicators corresponding to each candidate mix proportion are determined. The construction process of the target compressive strength prediction model is as follows: Original sample data of LC3 concrete is obtained, and the original sample data is preprocessed to obtain target sample data. The sample data includes input data and compressive strength; the input data includes material composition parameters, mix proportion parameters, and curing parameters of LC3 concrete. The predicted physical compressive strength values ​​corresponding to each group of input data in the target sample data are determined, and an initial physical-information neural network is trained based on the target sample data and the predicted physical compressive strength values ​​to obtain a target physical-information neural network. Several initial teacher models are determined based on several machine learning models, and each initial teacher model is trained using the target sample data to obtain a target teacher model. The fusion weights corresponding to each target teacher model and the target physical-information neural network are determined based on a gating mechanism, and the target teacher model and the target physical-information neural network are fused based on the fusion weights to obtain the target compressive strength prediction model. Based on the predicted compressive strength values ​​of each candidate ratio and several performance evaluation indicators, several target candidate ratios that meet the preset performance conditions are screened in the current ratio population, and crossover and mutation operations are performed on several target candidate ratios to obtain the current optimized ratio population. The optimized mix proportion population is determined as the new current mix proportion population, and the process jumps to the step of using the pre-constructed target compressive strength prediction model to predict the compressive strength prediction value corresponding to each candidate mix proportion in the current mix proportion population, until the preset optimization termination condition is met, the optimized mix proportion population is determined as the target mix proportion population, and the mix proportion scheme of LC3 concrete is determined based on each target mix proportion in the target mix proportion population.

[0006] Optionally, generating a current ratio population containing several candidate ratios based on each of the preset constraint ranges includes: Based on the preset constraint range corresponding to each of the proportion design variables, the upper limit and lower limit values ​​of each of the proportion design variables are determined; Based on the upper and lower limits of each of the aforementioned proportion design variables, random sampling is performed within the preset constraint range to generate several initial candidate proportions; Determine a number of candidate ratios that satisfy the preset constraint range from the initial candidate ratios, and generate the current ratio population based on the number of candidate ratios.

[0007] Optionally, the data preprocessing includes missing value handling, outlier identification, duplicate sample removal, variable standardization and normalization, data encoding, sample partitioning, and data expansion.

[0008] Optionally, the step of using a pre-constructed target compressive strength prediction model to predict the compressive strength prediction value corresponding to each of the candidate ratios in the current ratio population includes: Input any of the candidate ratios in the current population ratio into the target compressive strength prediction model; The first compressive strength prediction result corresponding to the candidate ratio is generated using the target physical knowledge neural network in the target compressive strength prediction model, and the second compressive strength prediction result corresponding to the candidate ratio is generated using each of the target teacher models in the target compressive strength prediction model. The first compressive strength prediction result and each of the second compressive strength prediction results are fused based on the fusion weight to obtain the compressive strength prediction value corresponding to the candidate ratio.

[0009] Optionally, the performance evaluation indicators include carbon emission indicators, material cost indicators, and the prediction uncertainty of compressive strength. Accordingly, determining the performance evaluation indicators corresponding to each of the candidate ratios includes: For any of the candidate formulations, the carbon emission index corresponding to the candidate formulation is calculated based on the amount of each component material in the candidate formulation and the carbon emission factor of each component material. Calculate the material cost index corresponding to the candidate formulation based on the amount of each component material used and the unit price of each component material in the candidate formulation; The target compressive strength prediction model is used to predict the compressive strength of the candidate mix ratio several times, and the prediction uncertainty index corresponding to the candidate mix ratio is determined according to the degree of dispersion of each predicted compressive strength value.

[0010] Optionally, the step of screening several target candidate ratios that meet preset performance conditions in the current ratio population based on the predicted compressive strength values ​​of each of the candidate ratios and several performance evaluation indicators includes: Based on the predicted compressive strength values ​​of each candidate ratio and several performance evaluation indicators, the candidate ratios in the current ratio population are non-dominated and sorted to determine the non-dominated level corresponding to each candidate ratio. Calculate the crowding distance corresponding to each of the candidate allocations located in the same non-dominated hierarchy; The current population ratio is screened based on the non-dominated level and the crowding distance corresponding to each candidate ratio, so as to determine the candidate ratio with a higher non-dominated level and / or a larger crowding distance as the target candidate ratio.

[0011] Secondly, this application provides a mix design device for LC3 concrete, comprising: The mix proportion population generation module is used to obtain each mix proportion design variable of LC3 concrete and the preset constraint range corresponding to each mix proportion design variable, and generate a current mix proportion population containing several candidate mix proportions based on each preset constraint range. The performance evaluation index determination module is used to predict the compressive strength of each candidate mix proportion in the current mix proportion population using a pre-constructed target compressive strength prediction model, and to determine several performance evaluation indices corresponding to each candidate mix proportion. The construction process of the target compressive strength prediction model involves: acquiring raw sample data of LC3 concrete and performing data preprocessing on the raw sample data to obtain target sample data; wherein the sample data includes input data and compressive strength, and the input data includes material composition parameters, mix proportion parameters, and curing parameters of LC3 concrete; and determining each group of input data in the target sample data. The corresponding predicted physical compressive strength value is used, and an initial physical intelligence neural network is trained based on the target sample data and the predicted physical compressive strength value to obtain a target physical intelligence neural network; several initial teacher models are determined based on several machine learning models, and each initial teacher model is trained using the target sample data to obtain a target teacher model; the fusion weights corresponding to each target teacher model and the target physical intelligence neural network are determined based on a gating mechanism, and each target teacher model and the target physical intelligence neural network are fused based on the fusion weights to obtain the target compressive strength prediction model; The ratio population optimization module is used to screen several target candidate ratios that meet preset performance conditions in the current ratio population based on the predicted compressive strength values ​​of each candidate ratio and several performance evaluation indicators, and to perform crossover and mutation operations on several target candidate ratios to obtain the current optimized ratio population. The mix proportioning scheme determination module is used to determine the current optimized mix proportion population as the new current mix proportion population, and jump to the step of using the pre-constructed target compressive strength prediction model to predict the compressive strength prediction value corresponding to each candidate mix proportion in the current mix proportion population, until the preset optimization termination condition is met, the current optimized mix proportion population is determined as the target mix proportion population, and the mix proportioning scheme of LC3 concrete is determined based on each target mix proportion in the target mix proportion population.

[0012] Optionally, the matching population generation module includes: The upper limit value determination unit is used to determine the upper limit value and lower limit value of each of the proportion design variables based on the preset constraint range corresponding to each of the proportion design variables; An initial candidate ratio generation unit is used to randomly sample within the preset constraint range based on the upper limit and lower limit values ​​of each of the ratio design variables to generate a number of initial candidate ratios; A matching population generation unit is used to determine a number of candidate matching ratios that satisfy the preset constraint range in the initial candidate matching ratios, and to generate a current matching population based on the number of candidate matching ratios.

[0013] Thirdly, this application provides an electronic device, comprising: Memory, used to store computer programs; A processor is used to execute the computer program to implement the aforementioned LC3 concrete mix design method.

[0014] Fourthly, this application provides a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned LC3 concrete mix design method.

[0015] In this application, the following steps are taken: First, the design variables for each mix proportion of LC3 concrete and the preset constraint ranges corresponding to each mix proportion design variable are obtained. Then, a current mix proportion population containing several candidate mix proportions is generated based on each preset constraint range. Next, a pre-constructed target compressive strength prediction model is used to predict the compressive strength prediction value corresponding to each candidate mix proportion in the current mix proportion population, and several performance evaluation indicators corresponding to each candidate mix proportion are determined. Then, based on the predicted compressive strength value of each candidate mix proportion and several performance evaluation indicators, several target candidate mix proportions in the current mix proportion population that meet the preset performance conditions are selected. Crossover and mutation operations are performed on several target candidate mix proportions to obtain the current optimized mix proportion population. Subsequently, the current optimized mix proportion population is determined as the new current mix proportion population, and the process jumps to the step of using the pre-constructed target compressive strength prediction model to predict the compressive strength prediction value corresponding to each candidate mix proportion in the current mix proportion population, until the preset optimization termination condition is met. The current optimized mix proportion population is then determined as the target mix proportion population, and the mix proportion scheme of LC3 concrete is determined based on each target mix proportion in the target mix proportion population. As can be seen from the above, this application generates an initial population containing several candidate mix proportions and iteratively optimizes it. It uses a compressive strength prediction model to predict the compressive strength of each candidate mix proportion and calculates multiple performance evaluation indicators, providing a quantitative basis for multi-objective comprehensive evaluation. By screening target candidate mix proportions that meet preset performance conditions based on predicted values ​​and evaluation indicators and performing crossover and mutation operations, the mix proportion scheme achieves the co-evolution of multiple mutually restrictive objectives. Through iterative optimization until the preset termination conditions are met, a multi-objective mix proportion scheme that takes into account mechanical performance, environmental benefits, and economic costs is finally obtained, thus realizing the intelligent reverse design of LC3 concrete mix proportion that takes into account compressive strength prediction and multi-objective co-optimization. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1 A flowchart of a mix design method for LC3 concrete provided in this application; Figure 2 This application provides an architecture diagram of an LC3 concrete compressive strength prediction system. Figure 3 A specific mix design flowchart for LC3 concrete provided for this application; Figure 4 This application provides a flowchart of a specific LC3 concrete mix design based on a target compressive strength prediction model; Figure 5 A schematic diagram of a mix design device for LC3 concrete provided in this application; Figure 6 This application provides a structural diagram of an electronic device. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] LC3 concrete is a novel low-carbon and environmentally friendly cementitious material, prepared by replacing some of the cement clinker with calcined clay and limestone. Compared to traditional Portland cement, LC3 concrete exhibits significant advantages in reducing carbon dioxide emissions, saving energy, and improving concrete durability and crack resistance. In materials engineering practice, the compressive strength of the LC3 concrete system is mainly affected by a variety of factors, including the ratio of clinker to mineral admixtures, the activity of calcined clay, limestone filling and nucleation effects, water-cement ratio, sand-cement ratio, curing age, and curing temperature. These factors exhibit significant nonlinear coupling relationships, resulting in a multi-scale and multi-mechanism characteristic in the strength evolution process of LC3 concrete.

[0020] In the design of LC3 concrete mix proportions, existing research mostly focuses on compressive strength prediction. The predicted compressive strength results are primarily used to evaluate the performance of existing mix proportions, lacking a design mechanism that combines compressive strength prediction models with mix proportion optimization. This makes it difficult to comprehensively consider the interrelationships between multiple objectives such as compressive strength, low carbon emissions, material costs, and physical constraints, thus hindering intelligent reverse design and optimal scheme recommendation for LC3 concrete mix proportions. Therefore, this application provides a mix proportion design scheme for LC3 concrete that can establish an intelligent reverse design that balances compressive strength prediction with multi-objective collaborative optimization during the LC3 concrete mix proportion design process.

[0021] See Figure 1 As shown in the figure, an embodiment of the present invention discloses a mix design method for LC3 concrete, which may include: Step S11: Obtain the design variables of each mix proportion of LC3 concrete and the preset constraint range corresponding to each mix proportion design variable, and generate a current mix proportion population containing several candidate mix proportions based on each preset constraint range.

[0022] In this embodiment, the mix design variables for multi-objective optimization are first determined based on the LC3 concrete mix design requirements and engineering constraints. These mix design variables may include parameters such as the cement-cement ratio (OPC / B), calcined clay-cement ratio (CC / B), limestone-cement ratio (LS / B), gypsum-cement ratio (GY / B), water-cement ratio (W / B), sand-cement ratio (B / S), high-efficiency water-reducing agent dosage (SP), calcination temperature (HT), and curing temperature (CT).

[0023] The candidate mix design variables can be expressed as: X=[OPC / B,CC / B,LS / B,GY / B,W / B,B / S,SP,HT,CT]; Where X represents a candidate LC3 concrete mix design.

[0024] At the same time, reasonable upper and lower limits need to be set for each mix design variable. For example, OPC / B, CC / B, LS / B and GY / B should meet the constraints of cementitious material composition, W / B should be within a reasonable water-cement ratio range, SP should meet the engineering workability requirements, and HT and CT should be within the range allowed by actual process or curing conditions.

[0025] In this embodiment, the upper and lower limits of each proportion design variable can be determined based on the preset constraint range corresponding to each proportion design variable. That is, each proportion design variable should satisfy: ; in, This represents the i-th proportion design variable. and These represent the lower and upper limits of the mix design variables, respectively. In this way, by setting the variable range, this embodiment can restrict the optimization search to a mix design space with engineering feasibility, avoiding the generation of invalid solutions that clearly do not conform to the material composition rules or construction conditions.

[0026] In one embodiment, the proportion of cementitious materials can satisfy the following constraints: OPC / B+CC / B+LS / B+GY / B=1.

[0027] Subsequently, based on the upper and lower limits of the design variables for each ratio, random sampling is performed within the preset constraint range to generate several initial candidate ratios; then, several candidate ratios that satisfy the preset constraint range are determined from the initial candidate ratios, and the current ratio population is generated based on these candidate ratios.

[0028] Specifically, after determining the range of design variables for the mix proportions, an initial LC3 concrete mix proportion population is randomly generated based on a preset population size. Each individual represents a set of candidate mix proportion schemes, and each individual consists of a set of design variables. The current mix proportion population can be represented as: P0 = {X1, X2, X3, ..., XN}; Where P0 represents the initial mix proportion population, N represents the population size, and XN represents the Nth candidate LC3 concrete mix proportion scheme.

[0029] It should be noted that when generating the initial mix proportion population, constraint checks need to be performed on candidate individuals. For individuals that do not meet the constraints of cementitious material ratio, water-cement ratio range, material dosage range, or process parameter range, resampling, normalization correction, or constraint penalties can be used for processing. Through the above methods, it can be ensured that the candidate mix proportions in the initial mix proportion population are all within a reasonable design space, providing an effective starting point for subsequent multi-objective optimization.

[0030] Step S12: Using a pre-constructed target compressive strength prediction model, predict the compressive strength of each candidate mix proportion in the current mix proportion population, and determine several performance evaluation indicators corresponding to each candidate mix proportion; wherein, the construction process of the target compressive strength prediction model is as follows: obtain the original sample data of LC3 concrete, and perform data preprocessing on the original sample data to obtain target sample data; wherein, the sample data includes input data and compressive strength, the input data including the material composition parameters, mix proportion parameters, and curing parameters of LC3 concrete; determine the material composition parameters, mix proportion parameters, and curing parameters corresponding to each group of input data in the target sample data. The system calculates the predicted compressive strength value and trains an initial physical intelligence neural network based on the target sample data and the predicted physical compressive strength value to obtain a target physical intelligence neural network. It then determines several initial teacher models based on several machine learning models and trains each initial teacher model using the target sample data to obtain a target teacher model. Finally, it determines the fusion weights corresponding to each target teacher model and the target physical intelligence neural network based on a gating mechanism and fuses each target teacher model and the target physical intelligence neural network based on the fusion weights to obtain the target compressive strength prediction model.

[0031] To train the target compressive strength prediction model, namely the PIKAN model (Physics-Informed Kolmogorov-Arnold Network), the original sample data of LC3 concrete is first obtained, and the original sample data is preprocessed to obtain the target sample data. The sample data includes input data and compressive strength. The input data includes the material composition parameters, mix proportion parameters, and curing parameters of LC3 concrete.

[0032] Specifically, the original sample data can come from experimental data, public databases, engineering testing data, laboratory accumulated data, or other information sources that can reflect the composition and performance of LC3 concrete; the data format can be spreadsheets, text files, databases, or other data formats that can store structured data.

[0033] It should be noted that the original sample data should include at least the relevant variables that can reflect the material composition, mix proportions, curing conditions, environmental parameters, and target performance of LC3 concrete. The input information may include the composition of cementitious materials, water-cement ratio, curing age, curing environment, temperature conditions, admixture dosage, and other parameters that affect the formation of compressive strength. The output information may be the compressive strength at different ages or other evaluation indicators that can reflect the mechanical properties of the material.

[0034] Subsequently, the original sample data is organized, classified, or reconstructed according to the data source. Several data subsets can be formed according to different ages, different experimental conditions, or other classification methods to meet the needs of subsequent model training and prediction.

[0035] Subsequently, the aforementioned data subset undergoes preprocessing. This preprocessing may include missing value handling, outlier identification, duplicate sample removal, variable standardization, variable normalization, data encoding, sample partitioning, and other data processing methods that can improve model training quality. Specifically, for missing information, median imputation, mean imputation, mode imputation, interpolation methods, model prediction, neighborhood estimation, or other methods suitable for data completion can be used. To preserve the characteristics of missing information, corresponding missing information labels can also be constructed, enabling subsequent prediction models to identify the degree of data completeness.

[0036] In one specific implementation, to further improve the model training performance under small sample conditions, the preprocessed target sample data can be augmented or enhanced. The data augmentation can employ Kriging, Gaussian process regression, random sampling, Latin hypercube sampling, generative models, resampling, or other methods that increase the number of effective training samples, thereby forming a dataset for subsequent model training.

[0037] As can be seen, this embodiment is not limited to a single data source, data format, or preprocessing method. Instead, it establishes a unified data processing workflow, enabling data from different sources and with different structures to be converted into a data format suitable for subsequent prediction models. Simultaneously, through missing value handling, sample reorganization, and data expansion, sample utilization and data quality are improved, providing a reliable data foundation for subsequent physical modeling, intelligent prediction, and optimization design. This also enhances model training stability and prediction reliability under small sample conditions.

[0038] Subsequently, in this embodiment, a kinetic physical layer is established based on the strength formation mechanism of LC3 concrete to describe the influence of material composition, curing conditions, and reaction processes on the development of compressive strength. The kinetic physical layer can comprehensively consider factors such as the activity of calcined clay, the contribution of limestone, the reaction of cement clinker, hydration age, curing temperature, and the synergistic effect of components, and establish corresponding dynamic parameters or physical relationships so that the material activity under different working conditions can be dynamically adjusted with changes in environmental conditions.

[0039] In one specific implementation, a temperature-related activity parameter is established: Temperature parameter = Short-time temperature activity. Hydration activity over age. Further, a comprehensive activity parameter is constructed by combining the reaction process over age to reflect the changes in the material's reactivity under the combined effects of calcination and hydration. Specifically, in this embodiment, the activity of calcined clay is modified from a single temperature effect to the combined effect of calcination temperature activity and hydration activity over age, and both rapid and slow hydration processes are used to characterize the hydration activity over age. It should be noted that hydration activity over age can also be expressed using kinetic methods such as single exponential growth functions, double exponential growth functions, multi-exponential growth functions, Avrami-type hydration kinetic functions, maturity functions, logarithmic age functions, power functions, piecewise age functions, or equivalent age functions; calcination temperature activity can be described using Gaussian, piecewise linear, parabolic, Sigmoid, or empirical lookup table functions.

[0040] Furthermore, a calculation formula, Beff, is established to describe the effective cementing capacity of materials. Specifically, in this embodiment, the effective cementing material quantity characterizes the contribution of clinker, calcined clay, limestone, and the synergistic effect of calcined clay and limestone to strength formation. It should be noted that the effective cementing material quantity can be further increased by adding gypsum control terms, silica fume contribution terms, fly ash contribution terms, slag contribution terms, nucleation effect terms, filling effect terms, aluminate formation terms, or activity index terms. Each contribution coefficient can be obtained using empirical constants, trainable parameters, lookup table parameters, or predictions from machine learning models.

[0041] Based on this, a formula for calculating physical strength is constructed by combining material composition, water-cement ratio, environmental conditions, and relevant dynamic parameters. Furthermore, by calculating the physical strength relationship, the predicted physical compressive strength value corresponding to each group of input data in the target sample data is determined, providing physical prior information for the subsequent physical informed neural network.

[0042] Specifically, in this embodiment, the physical strength calculation formula comprehensively considers the contribution of effective cementitious materials, the water-cement ratio, and the influence of curing temperature on compressive strength. It should be noted that the physical strength formula can employ Abrams-type water-cement ratio formulas, Bolomey-type formulas, Ferret-type formulas, maturity strength formulas, exponential growth formulas, power function formulas, logarithmic age formulas, temperature correction formulas, or other semi-empirical strength development formulas. For example, the age term can be replaced by an equivalent age function, the temperature term can be replaced by a maturity function, and the water-cement ratio term can be replaced by the glue-to-water ratio, porosity, or effective water-cement ratio.

[0043] It should be noted that the kinetic physical layer can be established using exponential functions, power functions, piecewise functions, maturity models, empirical models, mechanistic models, or combinations thereof, or it can be implemented using other mathematical models that can reflect the dynamic characteristics of the LC3 concrete system. In this way, by constructing the kinetic physical layer, the model can incorporate material reaction mechanism information during the prediction process, improving the consistency between the prediction results and the actual evolution of the material, while also enhancing the model's adaptability under different mix proportions and curing conditions.

[0044] After obtaining the prior predicted physical compressive strength value through the kinetic physics layer, an initial physical information neural network is further constructed to establish the mapping relationship between LC3 concrete mix proportions, curing conditions, physical information, and compressive strength. In this embodiment, the initial physical information neural network can be trained based on target sample data and the predicted physical compressive strength value to obtain the corresponding target physical information neural network.

[0045] The physical-informed neural network takes material composition information, curing conditions, and the output of the kinetic physics layer as input, and outputs the predicted compressive strength. The model can simultaneously utilize the nonlinear characteristics in the experimental data and the physical constraint information provided by the kinetic physics layer, so that the prediction process can balance data-driven capability and physical consistency.

[0046] In one specific implementation, the physical awareness neural network can employ a neural network structure to express the complex nonlinear relationship between input variables and target variables. The network structure used can be an MLP (Multilayer Perceptron), radial basis function neural network, Transformer network, graph neural network, or other neural network structures capable of approximating nonlinear functions. Preferably, a network structure with strong nonlinear expressive power is used to improve the model's learning ability under small sample conditions.

[0047] Furthermore, during the training of the physical-informed neural network, a joint optimization training mechanism is established, considering both the data error between the predicted results and the actual compressive strength, as well as the consistency between the predicted results and the dynamic physical layer. This ensures that the model maintains a reasonable adherence to the physical laws of materials while learning the patterns in experimental data. Additionally, training methods combining single-stage training, multi-stage training, and pre-training with fine-tuning can be employed. Training strategies such as transfer learning, self-supervised learning, semi-supervised learning, incremental learning, online learning, or federated learning can also be used.

[0048] It should be noted that in this embodiment, regularization constraints, physical constraints, boundary constraints, monotonicity constraints, age constraints, uncertainty constraints, or other constraint information can be introduced into the loss function according to training requirements to optimize the model training process. The optimization algorithm can adopt gradient optimization algorithm, swarm intelligence optimization algorithm, or other parameter optimization methods to further improve the stability and generalization ability of the model.

[0049] In this way, this embodiment integrates physical information with the neural network prediction process, enabling the prediction model to not only learn the complex nonlinear relationships of the LC3 concrete system, but also fully utilize the prior knowledge provided by the dynamic physics layer, thereby improving the model's convergence performance, prediction accuracy, and generalization ability under small sample conditions. Furthermore, this embodiment is not limited to a single neural network structure, but is applicable to various prediction models with nonlinear modeling capabilities, thus having a wider range of applicability.

[0050] Furthermore, this embodiment establishes a dual-teacher gating adaptive fusion module to integrate the advantages of different prediction models and improve the stability and reliability of the LC3 concrete compressive strength prediction results. The fusion module includes at least two teacher prediction models and a gating decision module. The teacher models establish the prediction relationship between input variables and compressive strength, and provide auxiliary prediction information to the final prediction result. The teacher models can employ gradient boosting trees, ensemble learning models, support vector machines, random forests, neural networks, or other machine learning models with predictive capabilities.

[0051] In one implementation, two initial teacher models employ different types of prediction algorithms, enabling them to learn the variation patterns of LC3 concrete compressive strength from different perspectives, thereby generating complementary prediction results. Each initial teacher model is trained using target sample data to obtain a target teacher model.

[0052] Next, a gating module is established. This module automatically analyzes the applicability of different teacher models to the current sample based on the feature information of the input sample, and dynamically generates corresponding fusion weights. The gating module can employ neural networks, probabilistic models, attention mechanisms, hybrid expert models, or other methods capable of dynamic weight allocation.

[0053] Furthermore, based on a gating mechanism, the fusion weights corresponding to each target teacher model and the target physical awareness neural network are determined, and the target teacher model and the target physical awareness neural network are fused based on the fusion weights to obtain the target stress resistance prediction model. In the target stress resistance prediction model, the prediction results of the target teacher model and the prediction results of the physical awareness neural network are fused to form the final prediction result. The fusion method can employ fixed weights, dynamic weights, adaptive weighting, attention fusion, hybrid expert fusion, or other methods that can integrate multiple prediction results.

[0054] It should be noted that a joint training mechanism can also be established in this embodiment to optimize the target compressive strength prediction model, enabling the teacher model, gating module, and physical awareness neural network to be optimized collaboratively, thereby further improving the fusion prediction effect. After training is completed, the final compressive strength prediction value and the corresponding fusion weight information are output.

[0055] In this way, by constructing a dual-teacher gating adaptive fusion mechanism, the contribution ratio of different prediction models can be dynamically adjusted according to the characteristics of different LC3 concrete samples, without relying on a fixed-weight fusion method. Therefore, it can effectively improve the prediction stability and generalization ability of the model under complex working conditions, small sample conditions, and boundary samples. At the same time, this embodiment does not limit the teacher model and fusion method, and has strong scalability.

[0056] In this embodiment, the predictive performance of the target compressive strength prediction model can be comprehensively evaluated. The evaluation process can employ one or more evaluation indicators based on actual application requirements to analyze the accuracy, stability, generalization ability, and overall performance of the prediction model. Evaluation indicators may include the coefficient of determination, mean square error, mean absolute error, root mean square error, mean absolute percentage error, or other evaluation indicators suitable for regression prediction tasks.

[0057] It should be noted that this embodiment can also output comparisons of prediction results between different models, changes in fusion weights, distribution of prediction errors, residual analysis results, feature contribution information, and other information that can reflect model performance.

[0058] Furthermore, in this embodiment, the prediction results can be displayed in tables, graphs, curves or other visualization methods to provide a reference for model analysis and engineering applications.

[0059] In this embodiment, an LC3 concrete compressive strength prediction system was designed and implemented based on the target compressive strength prediction model. See the system architecture diagram below. Figure 2 As shown, the system includes a data input module, a data preprocessing module, an LC3 concrete dynamics and physical modeling module, a physical-informed neural network prediction module, a dual-teacher model training module, a gating weight calculation module, an adaptive fusion prediction module, a model evaluation module, and a result output module. The modules communicate with each other through data interfaces. Data can be transferred sequentially between modules or interactively called according to system configuration. Information output from one module can be used as input to the next module or multiple related modules, enabling data processing, physical constraint modeling, prediction result fusion, model performance evaluation, and prediction result output, thus forming a complete LC3 concrete compressive strength prediction system.

[0060] In this embodiment, for the initial breeding population and each candidate breeding scheme generated during subsequent iterations, it is necessary to call the target compressive strength prediction model to predict the compressive strength over 28 days. The specific process may include: first, inputting any candidate breeding scheme from the current breeding population into the target compressive strength prediction model; using the target physical awareness neural network in the target compressive strength prediction model to generate the first compressive strength prediction result corresponding to the candidate breeding scheme, and using each target teacher model in the target compressive strength prediction model to generate each second compressive strength prediction result corresponding to the candidate breeding scheme; then, fusing the first compressive strength prediction result and each second compressive strength prediction result based on the fusion weight to obtain the compressive strength prediction value corresponding to the candidate breeding scheme.

[0061] In this embodiment, the target compressive strength prediction model is not used alone for strength prediction of a given sample, but is further used as a surrogate evaluation model in the multi-objective optimization module. That is to say, when searching for candidate ratios, the multi-objective optimization algorithm does not need to conduct real experiments on each candidate scheme, but quickly evaluates the compressive strength by calling the target compressive strength prediction model, thereby significantly improving the ratio optimization efficiency.

[0062] After obtaining the predicted 28-day compressive strength values ​​for the candidate mix proportions, several performance evaluation indicators corresponding to each candidate mix proportion are further calculated; among them, the performance evaluation indicators include carbon emission indicators, material cost indicators, and the prediction uncertainty of compressive strength indicators.

[0063] For any candidate formulation, first calculate the corresponding carbon emission index based on the amount of each component and its carbon emission factor. This is expressed as: CO2(X) = Σmi·ei; Where CO2(X) represents the carbon emission per unit volume corresponding to candidate ratio X, mi represents the amount of material i, and ei represents the carbon emission factor corresponding to material i.

[0064] Next, based on the quantity of each component material and its unit price in the candidate formulation, the material cost index corresponding to the candidate formulation is calculated. This is expressed as: Cost(X) = Σmi·ci; Where Cost(X) represents the unit volume material cost corresponding to candidate ratio X, and ci represents the unit price of the i-th material.

[0065] Next, the target compressive strength prediction model is used to predict the compressive strength of the candidate mix ratios several times, and the prediction uncertainty index corresponding to each predicted compressive strength value is determined based on the dispersion of the predicted values. Specifically, prediction uncertainty is used to measure the reliability of the target compressive strength prediction model's prediction results for the candidate mix ratio strength. Prediction uncertainty can be obtained through model ensemble, bootstrap sampling, MC Dropout, multiple forward propagation, or a measure of distance from the support domain of the training data.

[0066] In one specific implementation, three independently initialized compressive strength prediction models can be trained to predict the strength of the same candidate mix ratio. The mean of the prediction results from the three models is used as the final predicted strength, and the variance between the prediction results is used as the uncertainty index, represented by the standard deviation of multiple prediction results. UQ(X)=std(f1(X),f2(X),...,fk(X)); Where UQ(X) represents the prediction uncertainty corresponding to candidate ratio X, and f1(X) to fk(X) represent the intensity prediction values ​​obtained by the model making k predictions for the same candidate ratio.

[0067] By calculating compressive strength, carbon emissions, cost, and predicting uncertainty, this embodiment can perform multidimensional evaluation of each candidate formulation, providing a foundation for the subsequent construction of a multi-objective function.

[0068] Step S13: Based on the predicted compressive strength values ​​of each candidate ratio and several performance evaluation indicators, select several target candidate ratios in the current ratio population that meet the preset performance conditions, and perform crossover and mutation operations on several target candidate ratios to obtain the current optimized ratio population.

[0069] In this embodiment, a multi-objective function is constructed based on the predicted compressive strength, carbon emissions, cost, and prediction uncertainty of the candidate mix proportions. This multi-objective function transforms the design requirements of "high strength, low carbon emissions, low cost, and low uncertainty" into mathematical objectives that can be calculated and compared by the optimization algorithm.

[0070] In one specific implementation, the multi-objective function can be expressed as: F(X)=[f1(X),f2(X),f3(X),f4(X)]; Where f1(X) = -Strength28 days(X); f2(X) = CO2(X); f3(X) = Cost(X); f4(X) = UQ(X).

[0071] Since multi-objective optimization algorithms are typically based on minimizing objectives, the inverse of the 28-day compressive strength, which needs to be maximized, can be used as the optimization objective, i.e., minimizing -Strength28day(X). For carbon emissions, costs, and prediction uncertainties, optimization is performed directly in the form of minimization.

[0072] Through the aforementioned multi-objective function, this embodiment can balance multiple conflicting objectives. For example, increasing intensity may lead to higher costs or carbon emissions, while reducing carbon emissions may lead to decreased intensity or increased uncertainty, and reducing uncertainty may limit the scope of the optimization search. The role of the multi-objective function is precisely to unify these conflicting objectives into the optimization framework, enabling the algorithm to search for candidate solutions under different trade-offs.

[0073] In this embodiment, in order to screen several target candidate ratios that meet preset performance conditions in the current ratio population based on the predicted compressive strength values ​​of each candidate ratio and several performance evaluation indicators, the candidate ratios in the current ratio population are first sorted non-dominated based on the predicted compressive strength values ​​of each candidate ratio and several performance evaluation indicators to determine the non-dominated level corresponding to each candidate ratio.

[0074] Specifically, for two candidate solutions Xa and Xb, if Xa is not inferior to Xb on all objectives and is superior to Xb on at least one objective, then Xa is considered to dominate Xb. If there is no dominance relationship between the two solutions, it means that each has its own advantages on different objectives, and it cannot be simply judged that one is absolutely superior to the other.

[0075] By using non-dominated sorting, the current population can be divided into multiple Pareto fronts of different levels: F1, F2, F3, ..., Fm; where F1 represents the first non-dominated layer, i.e., the set of optimal Pareto candidate schemes in the current population; F2 represents the second non-dominated layer, and so on. Candidate schemes located in higher non-dominated layers have better overall objective performance and therefore have a higher retention priority in subsequent selection processes.

[0076] As can be seen, the introduction of non-dominated ranking means that this embodiment no longer relies on a single weighted score to judge the merits of the ratio, but can retain multiple candidate solutions with different engineering preferences, such as high-intensity solutions, low-carbon solutions, low-cost solutions and low-uncertainty solutions.

[0077] After completing the non-dominated sorting, this embodiment further calculates the crowding distance corresponding to each candidate allocation at the same non-dominated level. The crowding distance is used to measure the sparsity of a candidate solution in the target space. If there are few other solutions distributed around a candidate solution, the crowding distance is large, indicating that the solution is highly representative in the target space. If there are already a large number of similar solutions near a candidate solution, the crowding distance is small.

[0078] The purpose of crowding distance is to maintain the diversity of the Pareto solution set and prevent optimization results from concentrating in a narrow region. For example, if individuals are selected solely based on non-dominance level, the algorithm may tend to retain similar solutions of a certain type while ignoring other ratios with different engineering significance. By introducing crowding distance, it is possible to simultaneously retain different types of candidate solutions, such as high intensity, low carbon emissions, low cost, low uncertainty, and comprehensive compromises.

[0079] After completing the non-dominated ranking and crowding distance calculation, a selection operation is performed on the current population ratio based on the non-dominated level and crowding distance of each candidate pairing to determine the candidate pairing with higher non-dominated levels and / or larger crowding distances as the target candidate pairings. That is, during the selection process, individuals with higher non-dominated levels are prioritized; when non-dominated levels are the same, individuals with larger crowding distances are prioritized. Individuals with better fitness and more even distribution are selected as the target candidate pairings.

[0080] In this embodiment, the target candidate matching is used as the parent generation. Subsequently, crossover and mutation operations are performed on the parent individuals to generate new candidate matching.

[0081] Cross operations are used to combine some variables of two candidate ratios to form new candidate solutions. For example, if one candidate solution has lower carbon emissions and another has higher intensity, cross operations may generate a new solution that combines both higher intensity and lower carbon emissions.

[0082] The mutation operation is used to make small random perturbations to some variables in the candidate ratios, such as adjusting variables like W / B, CC / B, LS / B, GY / B, or SP, in order to expand the search space and prevent the algorithm from getting trapped in local optima too early.

[0083] The new individuals resulting from crossover and mutation still need to meet the constraints of the design variable range and material composition. For individuals that do not meet the constraints, methods such as variable correction, resampling, or constraint penalties can be used to ensure that the generated new candidate solutions are engineering feasible.

[0084] Step S14: Determine the current optimized mix proportion population as the new current mix proportion population, and jump to the step of using the pre-constructed target compressive strength prediction model to predict the compressive strength prediction value corresponding to each candidate mix proportion in the current mix proportion population, until the preset optimization termination condition is met, determine the current optimized mix proportion population as the target mix proportion population, and determine the mix proportion scheme of LC3 concrete based on each target mix proportion in the target mix proportion population.

[0085] A new generation of candidate LC3 concrete mix proportions was generated through selection, crossover, and mutation operations. For each candidate mix proportion in the new generation, the target compressive strength prediction model was invoked again to predict the 28-day compressive strength, and the corresponding carbon emissions, costs, and prediction uncertainties were calculated.

[0086] This process can be represented as: Pt→Selection→Crossover→Mutation→Pt+1; Where Pt represents the t-th generation population and Pt+1 represents the t+1-th generation population.

[0087] As the iteration process continues, the candidate configurations in the population will gradually move closer to the better Pareto front. In other words, in this embodiment, schemes that perform poorly in terms of intensity, carbon emissions, cost, and uncertainty will be continuously eliminated, while those that have a good trade-off between multiple objectives will be retained.

[0088] After each iteration, determine whether the preset optimization termination condition has been met. The optimization termination condition may include reaching the maximum number of iterations, the change in the Pareto front being less than a preset threshold, no significant improvement in the optimal solution for several consecutive generations, or reaching the preset limit of computing resources.

[0089] In one specific implementation, the maximum number of iterations is used as the optimization termination condition. When the number of iterations has not reached the preset value, non-dominated sorting, crowding distance calculation, selection, crossover and mutation operations are continued to generate the next generation of candidate matching ratios. When the number of iterations reaches the preset value, the optimization stops and the final Pareto optimal candidate solution set is output.

[0090] By setting optimized termination conditions, this embodiment can achieve a balance between optimization accuracy and computational cost, avoiding the waste of computational resources caused by infinite iteration.

[0091] Once the optimization process reaches the termination condition, the current optimized population is designated as the target population, and the target proportions in the target population are output as the final Pareto optimal proportion set. It should be noted that the target proportions in the Pareto optimal proportion set are independent of each other among the multiple optimization objectives; that is, no single scheme can simultaneously outperform another scheme in terms of intensity, carbon emissions, cost, and uncertainty.

[0092] The Pareto optimal ratio scheme set can be represented as: ; in, This represents the final set of Pareto optimal proportions. This represents the q-th Pareto optimal ratio scheme.

[0093] It should be noted that the Pareto optimal ratio scheme does not represent a unique optimal ratio, but rather a set of ratio schemes with different trade-offs. For example, one scheme may have the highest intensity but higher cost; another scheme may have the lowest carbon emissions but slightly lower intensity; and yet another scheme may achieve a good overall balance between intensity, carbon emissions, cost, and uncertainty.

[0094] Therefore, the output of this embodiment is a set of formulation schemes that can be used for subsequent experimental verification and engineering screening, rather than directly replacing the actual experiment to determine the final formulation.

[0095] To facilitate engineering applications and result interpretation, representative proportioning schemes were further selected from the Pareto optimal proportioning scheme set. Representative schemes may include schemes with maximum compressive strength, minimum carbon emissions, minimum cost, minimum prediction uncertainty, and Knee point comprehensive compromise schemes.

[0096] Among them, the maximum strength scheme represents the ratio with the highest 28-day compressive strength in the Pareto solution set; the minimum carbon emission scheme represents the ratio with the lowest carbon emissions; the minimum cost scheme represents the ratio with the lowest material cost; the minimum UQ (Uncertainty Quantification) scheme represents the ratio with the lowest prediction uncertainty and the most reliable model prediction; and the Knee point scheme represents the ratio that achieves a good overall balance among multiple objectives.

[0097] The above representative formulation can be expressed as: X_strength=argmaxStrength28 days(X); X_carbon = argminCO2(X); X_cost = argminCost(X); X_UQ=argminUQ(X); X_knee=Knee(P ); Where X_strength represents the maximum strength scheme, X_carbon represents the minimum carbon emission scheme, X_cost represents the minimum cost scheme, X_UQ represents the scheme with the lowest prediction uncertainty, and X_knee represents the Knee-point comprehensive compromise scheme. The Knee-point scheme usually represents a compromise that can significantly improve other objectives while sacrificing less of one objective, and therefore can be given priority as a candidate ratio recommended to engineers or experimentalists for further verification.

[0098] By selecting the above representative mix design schemes, the complex Pareto solution set can be transformed into a mix design scheme that is easier to understand and apply, providing decision support for the low-cost, low-carbon-emission, low-uncertainty, and high-strength mix design of LC3 concrete.

[0099] In one specific implementation, the process for designing the LC3 concrete mix proportions can be found in [reference needed]. Figure 3 As shown. Specifically, using the range of LC3 concrete mix design variables as input, an initial candidate mix population is randomly generated, and a pre-trained target compressive strength prediction model is used to predict the 28-day compressive strength. Simultaneously, the corresponding carbon emissions, cost, and prediction uncertainty are calculated. Subsequently, a multi-objective function incorporating high strength, low carbon emissions, low cost, and low uncertainty is constructed, and a new generation of candidate mixes is continuously generated through non-dominated sorting, crowding distance calculation, and selection, crossover, and mutation operations. After reaching a preset number of iterations, a Pareto optimal candidate scheme set is output, and representative schemes with maximum strength, lowest carbon emissions, lowest cost, lowest UQ, and Knee point are further selected.

[0100] In this way, the target compressive strength prediction model in this embodiment is further extended to the material design stage. It can not only predict the compressive strength of LC3 concrete, but also optimize the material mix ratio based on multiple performance targets, realizing the functional expansion from performance prediction to mix design. Since the optimization targets, constraints, and optimization algorithms can all be adjusted according to actual needs, this embodiment has strong adaptability and scalability, and can meet the design requirements of low-carbon, high-performance LC3 concrete in different engineering scenarios.

[0101] It should be noted that, see Figure 4 As shown, the specific process for LC3 concrete mix design based on the target compressive strength prediction model can be as follows: 1. Data processing stage: First, input data such as ratio and maintenance conditions are obtained, and preprocessing such as missing value handling, outlier screening and normalization is performed. Finally, the data is divided into training set and test set to provide a high-quality data foundation for model training.

[0102] 2. Kinetic physical layer correction stage: Temperature parameter = short-time temperature activity Hydration activity at different ages, constructing the formulas for calculating effective gelling capacity (Beff) and physical strength. .

[0103] 3. Physically informed neural network prediction stage: Construct a physically informed neural network, use the neural network to express nonlinear relationships, train it by combining data loss and physical consistency loss, and output the predicted value of compressive strength.

[0104] 4. Dual-Teacher Gated Adaptive Fusion Stage: Two teacher models are trained to obtain the predicted values ​​of the teacher models and the physical awareness model. The gating network calculates the weights based on the sample features and outputs the gated fusion compressive strength prediction results.

[0105] 5. Results Output and Evaluation Stage: The model is evaluated using performance metrics such as R², RMSE (Root Mean Squared Error), MAE (Mean Absolute Error), and MSE (Mean Squared Error), and the predicted values, gating weights, and model comparison results are output.

[0106] 6. Multi-objective ratio optimization design stage: The trained prediction model is used as a surrogate model to establish a multi-objective optimization model with the objectives of maximizing intensity, minimizing carbon emissions, minimizing cost, and minimizing physical consistency error. The corresponding algorithm is used to perform ratio optimization search, and finally outputs the Pareto optimal solution set and recommended ratio scheme.

[0107] As can be seen from the above, in this embodiment, the design variables of each mix proportion of LC3 concrete and the preset constraint range corresponding to each mix proportion design variable are first obtained, and a current mix proportion population containing several candidate mix proportions is generated based on each preset constraint range. Then, the pre-constructed target compressive strength prediction model is used to predict the compressive strength prediction value corresponding to each candidate mix proportion in the current mix proportion population, and several performance evaluation indicators corresponding to each candidate mix proportion are determined. Next, based on the compressive strength prediction value of each candidate mix proportion and several performance evaluation indicators, several target candidate mix proportions that meet the preset performance conditions in the current mix proportion population are screened, and crossover and mutation operations are performed on several target candidate mix proportions to obtain the current optimized mix proportion population. Subsequently, the current optimized mix proportion population is determined as the new current mix proportion population, and the process jumps to the step of using the pre-constructed target compressive strength prediction model to predict the compressive strength prediction value corresponding to each candidate mix proportion in the current mix proportion population, until the preset optimization termination condition is met, the current optimized mix proportion population is determined as the target mix proportion population, and the mix proportion scheme of LC3 concrete is determined based on each target mix proportion in the target mix proportion population. As can be seen from the above, this embodiment generates an initial population containing several candidate mix proportions and iteratively optimizes it. It uses a compressive strength prediction model to predict the compressive strength of each candidate mix proportion and calculates multiple performance evaluation indicators, providing a quantitative basis for multi-objective comprehensive evaluation. By screening target candidate mix proportions that meet preset performance conditions based on predicted values ​​and evaluation indicators and performing crossover and mutation operations, the mix proportion scheme achieves the co-evolution of multiple mutually restrictive objectives. Through iterative optimization until the preset termination condition is met, a multi-objective mix proportion scheme that takes into account mechanical performance, environmental benefits, and economic costs is finally obtained, thereby realizing the intelligent reverse design of LC3 concrete mix proportion that takes into account compressive strength prediction and multi-objective co-optimization.

[0108] Accordingly, see Figure 5 As shown in the embodiment of this application, a mix design device for LC3 concrete is also provided, which may include: The mix proportion population generation module 11 is used to obtain each mix proportion design variable of LC3 concrete and the preset constraint range corresponding to each mix proportion design variable, and generate a current mix proportion population containing several candidate mix proportions based on each preset constraint range. The performance evaluation index determination module 12 is used to predict the compressive strength of each candidate mix proportion in the current mix proportion population using a pre-constructed target compressive strength prediction model, and to determine several performance evaluation indices corresponding to each candidate mix proportion. The construction process of the target compressive strength prediction model is as follows: acquiring raw sample data of LC3 concrete and performing data preprocessing on the raw sample data to obtain target sample data; wherein the sample data includes input data and compressive strength, and the input data includes material composition parameters, mix proportion parameters, and curing parameters of LC3 concrete; determining the input data for each group in the target sample data. Based on the corresponding predicted physical compressive strength value, and using the target sample data and the predicted physical compressive strength value, an initial physical information neural network is trained to obtain a target physical information neural network; several initial teacher models are determined based on several machine learning models, and each initial teacher model is trained using the target sample data to obtain a target teacher model; based on a gating mechanism, the fusion weights corresponding to each target teacher model and the target physical information neural network are determined, and the target teacher model and the target physical information neural network are fused based on the fusion weights to obtain the target compressive strength prediction model; The ratio population optimization module 13 is used to screen several target candidate ratios that meet preset performance conditions in the current ratio population based on the predicted compressive strength values ​​of each candidate ratio and several performance evaluation indicators, and to perform crossover and mutation operations on several target candidate ratios to obtain the current optimized ratio population. The mix proportion determination module 14 is used to determine the current optimized mix proportion population as the new current mix proportion population, and jump to the step of using the pre-constructed target compressive strength prediction model to predict the compressive strength prediction value corresponding to each candidate mix proportion in the current mix proportion population, until the preset optimization termination condition is met, the current optimized mix proportion population is determined as the target mix proportion population, and the mix proportion scheme of LC3 concrete is determined based on each target mix proportion in the target mix proportion population.

[0109] In some specific embodiments, the matching population generation module 11 may include: The upper limit value determination unit is used to determine the upper limit value and lower limit value of each of the proportion design variables based on the preset constraint range corresponding to each of the proportion design variables; An initial candidate ratio generation unit is used to randomly sample within the preset constraint range based on the upper limit and lower limit values ​​of each of the ratio design variables to generate a number of initial candidate ratios; A matching population generation unit is used to determine a number of candidate matching ratios that satisfy the preset constraint range in the initial candidate matching ratios, and to generate a current matching population based on the number of candidate matching ratios.

[0110] In some specific implementations, the data preprocessing includes missing value handling, outlier identification, duplicate sample removal, variable standardization and normalization, data encoding, sample partitioning, and data expansion.

[0111] In some specific embodiments, the performance evaluation index determination module 12 may include: A candidate ratio input unit is used to input any of the candidate ratios in the current ratio population into the target compressive strength prediction model; The prediction result generation unit is used to generate a first compressive strength prediction result corresponding to the candidate ratio using the target physical knowledge neural network in the target compressive strength prediction model, and to generate a second compressive strength prediction result corresponding to the candidate ratio using each of the target teacher models in the target compressive strength prediction model. The prediction result fusion unit is used to fuse the first compressive strength prediction result and each of the second compressive strength prediction results based on the fusion weight, so as to obtain the compressive strength prediction value corresponding to the candidate ratio.

[0112] In some specific embodiments, the performance evaluation indicators include carbon emission indicators, material cost indicators, and the prediction uncertainty indicator of compressive strength; Accordingly, the performance evaluation index determination module 12 may include: A carbon emission index calculation unit is used to calculate the carbon emission index corresponding to any of the candidate formulations based on the amount of each component material in the candidate formulation and the carbon emission factor of each component material. The material cost index calculation unit is used to calculate the material cost index corresponding to the candidate ratio based on the amount of each component material used in the candidate ratio and the unit price of each component material. The prediction uncertainty index determination unit is used to perform several compressive strength predictions on the candidate mix ratio using the target compressive strength prediction model, and determine the prediction uncertainty index corresponding to the candidate mix ratio based on the degree of dispersion of each predicted compressive strength value.

[0113] In some specific embodiments, the population optimization module 13 may include: The non-dominated level determination unit is used to perform non-dominated sorting of each candidate ratio in the current ratio population based on the predicted compressive strength value of each candidate ratio and several performance evaluation indicators, so as to determine the non-dominated level corresponding to each candidate ratio. A congestion distance calculation unit is used to calculate the congestion distance corresponding to each of the candidate ratios located in the same non-dominated level; The matching population screening unit is used to screen the current matching population according to the non-dominated level and the crowding distance corresponding to each of the candidate matching ratios, so as to determine the candidate matching ratios with higher non-dominated levels and / or larger crowding distances as the target candidate matching ratios.

[0114] Furthermore, embodiments of this application also disclose an electronic device, Figure 6 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the LC3 concrete mix design method disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0115] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0116] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0117] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the LC3 concrete mix design method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0118] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned LC3 concrete mix design method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0119] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0120] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0121] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0122] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0123] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A mix design method for LC3 concrete, characterized in that, include: Obtain the design variables of each mix proportion of LC3 concrete and the preset constraint range corresponding to each mix proportion design variable, and generate a current mix proportion population containing several candidate mix proportions based on each preset constraint range. Using a pre-constructed target compressive strength prediction model, the predicted compressive strength values ​​corresponding to each candidate mix proportion in the current mix proportion population are predicted, and several performance evaluation indicators corresponding to each candidate mix proportion are determined. The construction process of the target compressive strength prediction model is as follows: Original sample data of LC3 concrete is obtained, and the original sample data is preprocessed to obtain target sample data. The sample data includes input data and compressive strength; the input data includes material composition parameters, mix proportion parameters, and curing parameters of LC3 concrete. The predicted physical compressive strength values ​​corresponding to each group of input data in the target sample data are determined, and an initial physical-information neural network is trained based on the target sample data and the predicted physical compressive strength values ​​to obtain a target physical-information neural network. Several initial teacher models are determined based on several machine learning models, and each initial teacher model is trained using the target sample data to obtain a target teacher model. The fusion weights corresponding to each target teacher model and the target physical-information neural network are determined based on a gating mechanism, and the target teacher model and the target physical-information neural network are fused based on the fusion weights to obtain the target compressive strength prediction model. Based on the predicted compressive strength values ​​of each candidate ratio and several performance evaluation indicators, several target candidate ratios that meet the preset performance conditions are screened in the current ratio population, and crossover and mutation operations are performed on several target candidate ratios to obtain the current optimized ratio population. The optimized mix proportion population is determined as the new current mix proportion population, and the process jumps to the step of using the pre-constructed target compressive strength prediction model to predict the compressive strength prediction value corresponding to each candidate mix proportion in the current mix proportion population, until the preset optimization termination condition is met, the optimized mix proportion population is determined as the target mix proportion population, and the mix proportion scheme of LC3 concrete is determined based on each target mix proportion in the target mix proportion population.

2. The mix design method for LC3 concrete according to claim 1, characterized in that, The generation of a current matching population containing several candidate matching ratios based on each of the preset constraint ranges includes: Based on the preset constraint range corresponding to each of the proportion design variables, the upper limit and lower limit values ​​of each of the proportion design variables are determined; Based on the upper and lower limits of each of the aforementioned proportion design variables, random sampling is performed within the preset constraint range to generate several initial candidate proportions; Determine a number of candidate ratios that satisfy the preset constraint range from the initial candidate ratios, and generate the current ratio population based on the number of candidate ratios.

3. The mix design method for LC3 concrete according to claim 1, characterized in that, The data preprocessing includes missing value handling, outlier identification, duplicate sample removal, variable standardization and normalization, data encoding, sample partitioning, and data expansion.

4. The mix design method for LC3 concrete according to claim 1, characterized in that, The step of using a pre-constructed target compressive strength prediction model to predict the compressive strength prediction value corresponding to each of the candidate ratios in the current ratio population includes: Input any of the candidate ratios in the current population ratio into the target compressive strength prediction model; The first compressive strength prediction result corresponding to the candidate ratio is generated using the target physical knowledge neural network in the target compressive strength prediction model, and the second compressive strength prediction result corresponding to the candidate ratio is generated using each of the target teacher models in the target compressive strength prediction model. The first compressive strength prediction result and each of the second compressive strength prediction results are fused based on the fusion weight to obtain the compressive strength prediction value corresponding to the candidate ratio.

5. The mix design method for LC3 concrete according to claim 1, characterized in that, The performance evaluation indicators include carbon emission indicators, material cost indicators, and the prediction uncertainty indicator of compressive strength; Accordingly, determining the performance evaluation indicators corresponding to each of the candidate ratios includes: For any of the candidate formulations, the carbon emission index corresponding to the candidate formulation is calculated based on the amount of each component material in the candidate formulation and the carbon emission factor of each component material. Calculate the material cost index corresponding to the candidate formulation based on the amount of each component material used and the unit price of each component material in the candidate formulation; The target compressive strength prediction model is used to predict the compressive strength of the candidate mix ratio several times, and the prediction uncertainty index corresponding to the candidate mix ratio is determined according to the degree of dispersion of each predicted compressive strength value.

6. The mix design method for LC3 concrete according to any one of claims 1 to 5, characterized in that, The process of selecting several target candidate ratios that meet preset performance conditions from the current ratio population based on the predicted compressive strength values ​​of each candidate ratio and several performance evaluation indicators includes: Based on the predicted compressive strength values ​​of each candidate ratio and several performance evaluation indicators, the candidate ratios in the current ratio population are non-dominated and sorted to determine the non-dominated level corresponding to each candidate ratio. Calculate the crowding distance corresponding to each of the candidate allocations located in the same non-dominated hierarchy; The current population ratio is screened based on the non-dominated level and the crowding distance corresponding to each candidate ratio, so as to determine the candidate ratio with a higher non-dominated level and / or a larger crowding distance as the target candidate ratio.

7. A mix design device for LC3 concrete, characterized in that, include: The mix proportion population generation module is used to obtain the mix proportion design variables of LC3 concrete and the preset constraint range corresponding to each mix proportion design variable, and generate a current mix proportion population containing several candidate mix proportions based on each preset constraint range. The performance evaluation index determination module is used to predict the compressive strength of each candidate mix proportion in the current mix proportion population using a pre-constructed target compressive strength prediction model, and to determine several performance evaluation indices corresponding to each candidate mix proportion. The construction process of the target compressive strength prediction model involves: acquiring raw sample data of LC3 concrete and performing data preprocessing on the raw sample data to obtain target sample data; wherein the sample data includes input data and compressive strength, and the input data includes material composition parameters, mix proportion parameters, and curing parameters of LC3 concrete; and determining each group of input data in the target sample data. The corresponding predicted physical compressive strength value is used, and an initial physical intelligence neural network is trained based on the target sample data and the predicted physical compressive strength value to obtain a target physical intelligence neural network; several initial teacher models are determined based on several machine learning models, and each initial teacher model is trained using the target sample data to obtain a target teacher model; the fusion weights corresponding to each target teacher model and the target physical intelligence neural network are determined based on a gating mechanism, and each target teacher model and the target physical intelligence neural network are fused based on the fusion weights to obtain the target compressive strength prediction model; The ratio population optimization module is used to screen several target candidate ratios that meet preset performance conditions in the current ratio population based on the predicted compressive strength values ​​of each candidate ratio and several performance evaluation indicators, and to perform crossover and mutation operations on several target candidate ratios to obtain the current optimized ratio population. The mix proportioning scheme determination module is used to determine the current optimized mix proportion population as the new current mix proportion population, and jump to the step of using the pre-constructed target compressive strength prediction model to predict the compressive strength prediction value corresponding to each candidate mix proportion in the current mix proportion population, until the preset optimization termination condition is met, the current optimized mix proportion population is determined as the target mix proportion population, and the mix proportioning scheme of LC3 concrete is determined based on each target mix proportion in the target mix proportion population.

8. The mix design device for LC3 concrete according to claim 7, characterized in that, The matching population generation module includes: The upper limit value determination unit is used to determine the upper limit value and lower limit value of each of the proportion design variables based on the preset constraint range corresponding to each of the proportion design variables; An initial candidate ratio generation unit is used to randomly sample within the preset constraint range based on the upper limit and lower limit values ​​of each of the ratio design variables to generate a number of initial candidate ratios; A matching population generation unit is used to determine a number of candidate matching ratios that satisfy the preset constraint range in the initial candidate matching ratios, and to generate a current matching population based on the number of candidate matching ratios.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory; wherein the memory is used to store a computer program, which is loaded and executed by the processor to implement the mix design method for LC3 concrete as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, Used to store a computer program, which, when executed by a processor, implements the mix design method for LC3 concrete as described in any one of claims 1 to 6.

Images