A concrete optimization design method based on text information enhancement in a chlorine salt environment
By using a deep learning model enhanced with textual information and an improved non-dominated sorting genetic algorithm, the problems of long experimental cycles and high costs in traditional low-carbon concrete design methods are solved. This enables rapid optimization of concrete mix proportions under complex working conditions, reducing carbon emissions and costs, and improving design efficiency and low-carbon effects.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XI'AN UNIVERSITY OF ARCHITECTURE AND TECHNOLOGY
- Filing Date
- 2026-03-04
- Publication Date
- 2026-06-05
AI Technical Summary
Traditional low-carbon concrete design methods rely on laboratory trial-and-error experiments, resulting in long testing cycles, high costs, and limited parameter combinations. This makes it difficult to quickly obtain the optimal mix proportion that balances mechanical properties, durability, and low-carbon characteristics under complex working conditions, thus limiting its application and promotion in practical engineering.
By employing a text-based information enhancement method, combined with a deep learning model and an improved non-dominated sorting genetic algorithm, and through multi-objective optimization techniques, the optimal raw material mix ratio for concrete is determined to meet the optimization objectives of minimizing carbon emissions and costs, while also considering constraints on compressive strength and chloride ion diffusion coefficient.
It achieves a significant reduction in carbon emissions and material costs while meeting mechanical performance and durability requirements, improves the efficiency and low-carbonization of concrete design, and breaks through the limitations of traditional design.
Smart Images
Figure CN122154460A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of concrete structure durability technology, and in particular to a method for optimizing the design of concrete in a chloride salt environment based on text information enhancement. Background Technology
[0002] Concrete is a widely used man-made building material, extensively used in road and bridge engineering, industrial and civil construction, port and dock construction, and water conservancy projects. It is estimated that global annual concrete production exceeds 10 billion tons, accounting for approximately 8% of global carbon emissions. Therefore, developing low-carbon concrete technology and promoting the green transformation of the concrete industry are of significant strategic importance for achieving global carbon neutrality. In the concrete production process, using low-carbon raw materials is one of the key technological pathways to reduce carbon emissions. Examples include the use of auxiliary cementitious materials (such as fly ash and slag), recycled concrete aggregates, inert fillers (such as limestone powder), and alternative fibers.
[0003] However, traditional low-carbon concrete design methods mainly rely on laboratory trial-and-error experiments. Such methods often have problems such as long test cycles, high costs, and limited parameter combinations. They are difficult to quickly obtain the optimal mix proportion that takes into account mechanical properties, durability, and low-carbon characteristics under complex working conditions, which limits their application and promotion in engineering practice. Summary of the Invention
[0004] Therefore, it is necessary to provide a method for optimizing the design of concrete in a chloride-salt environment based on text information enhancement to address the aforementioned technical problems.
[0005] The following technical solution is adopted in this specification: This specification provides a text-enhanced method for optimizing the design of concrete in chloride-rich environments, including: The project aims to obtain a concrete compressive strength prediction model and a text-enhanced deep learning model. The compressive strength prediction model is used to predict the compressive strength of concrete based on the raw material mix proportions. The text-enhanced deep learning model is used to predict the chloride ion diffusion coefficient based on the raw material mix proportions and text information. The text information includes the raw material characteristics, environmental conditions, and microstructural features of the concrete. With the optimization objectives of minimizing concrete carbon emissions and cost, and the raw material mix proportions as optimization variables, an improved non-dominated sorting genetic algorithm is used for multi-objective optimization under the condition that the raw material mix proportions meet conventional and performance constraints. This algorithm determines the Pareto optimal front, i.e., the feasible solution set of the raw material mix proportions. Conventional constraints include the range, proportion, and absolute volume constraints of the raw materials; performance constraints include compressive strength constraints and chloride ion diffusion coefficient constraints. The improved non-dominated sorting genetic algorithm uses a linear annealing mechanism to dynamically adjust the mutation rate during the population mutation process and employs a reflection method to handle parameter out-of-bounds issues. The optimal raw material mix proportion for concrete is determined from the feasible solution set of raw material mix proportions by using the approximation ideal solution sorting method.
[0006] Optionally, the raw material proportions must meet performance constraints, including: The raw material mix proportions and curing age are input into the compressive strength prediction model to obtain the compressive strength of the concrete; when the compressive strength of the concrete is greater than or equal to the compressive strength required by the structural design, it is determined that the raw material mix proportions meet the compressive strength constraints. The raw material mix proportions, exposure period, test depth, and text information are input into a text-enhanced deep learning model to obtain the chloride ion concentration in the concrete. Based on the chloride ion concentration in concrete, the chloride ion diffusion coefficient is derived using Fick's second law. When the chloride ion diffusion coefficient is less than or equal to the preset maximum chloride ion diffusion coefficient, the raw material mix proportion is determined to meet the chloride ion diffusion coefficient constraint.
[0007] Optionally, the stress resistance prediction model includes three hidden layers connected in sequence, with the number of neurons in the hidden layers being 128, 64, and 32, respectively; each hidden layer includes a fully connected layer, a normalization layer, and a ReLU activation function connected in sequence; during the training of the stress resistance prediction model, the fully connected layer is initialized with Kaiming, the weights of the batch normalization layer are initialized to 1, the biases are initialized to 0, and a dropout layer with a dropout rate of 0.2 is added at the end of the last hidden layer.
[0008] Optionally, the text-enhanced deep learning model includes normalization units, natural language processing units, splicing units, and deep neural networks; the raw material mix proportions, exposure cycles, test depths, and text information are input into the text-enhanced deep learning model to obtain the chloride ion concentration of the concrete, including: The raw material mix ratio, exposure period, test depth and text information are input into the text enhancement deep learning model. The raw material mix ratio, exposure period and test depth are Z-score standardized by the standardized unit to obtain normalized numerical features. The text information is preprocessed, word embedded, and feature extracted sequentially by the natural language processing unit to obtain the text feature vector. The text feature vector and the normalized numerical features are concatenated by a concatenation unit, and the concatenated features are then input into a deep neural network to obtain the chloride ion concentration of the concrete.
[0009] Optionally, carbon emissions The calculation formula is: in, This refers to the quantity of raw materials used in concrete. Indicates the first in concrete The mass content of each raw material; Indicates the first in concrete Carbon emission factors of each raw material; Indicates the number of stages in the concrete production process; Indicates the first Energy consumption at each stage of production; Indicates the first Carbon emission factors of energy types used in each production stage; cost The calculation formula is: in, Indicates the first The actual market price per unit weight of raw materials; Indicates the first The unit price of the type of energy used in each stage of production.
[0010] Optionally, under the condition that the raw material mix proportions meet conventional constraints and performance constraints, an improved non-dominated sorting genetic algorithm is used for multi-objective optimization to determine the Pareto optimal front, including: Within a given range of materials, obtain a parent population that satisfies both general and performance constraints; each individual in the population represents a raw material mix proportion for concrete; the population size is N. Calculate the carbon emissions and raw material costs for each individual in the parent population; Based on the carbon emissions and raw material costs of each individual in the parent population, binary tournament selection, binary crossover, and polynomial mutation operations are performed on the individuals in sequence. If an individual after mutation exceeds the upper and lower bounds of the corresponding raw material, the corresponding offspring individual is reflected back into the upper and lower bounds of the raw material using the reflection method, so as to obtain an offspring population that satisfies the conventional constraints and performance constraints. Merge the parent and offspring populations, perform non-dominated sorting and crowding calculation, and select the N individuals with the highest crowding from the parent and offspring populations as the new parent population. Continue performing binary tournament selection, binary crossover, and polynomial mutation operations on the newly generated parent population until the maximum number of iterations is reached. Use all individuals in the parent population with a non-dominated level of 1 as the Pareto optimal frontier.
[0011] This specification provides a text-enhanced concrete optimization design device for chloride-salt environments, comprising: The acquisition module is used to acquire the concrete compressive strength prediction model and the text-enhanced deep learning model. The compressive strength prediction model is used to predict the compressive strength of concrete based on the raw material mix proportions of the concrete. The text-enhanced deep learning model is used to predict the chloride ion diffusion coefficient based on the raw material mix proportions of the concrete and text information. The text information includes the raw material characteristics, environmental conditions, and microstructural features of the concrete. The optimization module aims to minimize the carbon emissions and raw material costs of concrete, using the concrete mix proportions as optimization variables. Under the condition that the raw material mix proportions satisfy conventional and performance constraints, an improved non-dominated sorting genetic algorithm is employed for multi-objective optimization to determine the Pareto optimal front, i.e., the feasible solution set of the raw material mix proportions. Conventional constraints include range constraints, proportion constraints, and volume constraints of raw materials; performance constraints include compressive strength constraints and chloride ion diffusion coefficient constraints. The improved non-dominated sorting genetic algorithm dynamically adjusts the mutation rate during the population mutation process using a linear annealing mechanism and employs a reflection method to handle parameter out-of-bounds issues. The determination module is used to determine the optimal raw material mix proportion of concrete from the feasible solution set of raw material mix proportions by using the approximation ideal solution sorting method.
[0012] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for optimizing concrete design in a chloride-salt environment based on text-enhanced information.
[0013] This specification provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method for optimizing concrete design in a chloride-salt environment based on text-enhanced information.
[0014] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: The text-enhanced concrete optimization design method for chloride-enhanced environments provided in this specification employs an improved non-dominated sorting genetic algorithm to perform multi-objective optimization of mix proportions, with the dual optimization objectives of minimizing carbon emissions and minimizing costs. The algorithm dynamically adjusts the mutation rate through a linear annealing mechanism, enhancing the balance between global search and local convergence; it also combines reflection to handle boundary overflows, ensuring the effectiveness of the solution. Under the conditions of satisfying conventional constraints, compressive strength constraints, and chloride ion diffusion coefficient constraints, this method automatically finds the Pareto optimal solution set that minimizes both carbon emissions and costs, achieving synergistic optimization of mechanical properties, durability, and low-carbon characteristics. By using an approximation-ideal-solution sorting method, the optimal mix proportion is selected from the Pareto optimal frontier, achieving the best overall state in terms of low carbon emissions, low cost, and performance balance. Therefore, the concrete designed by this method can significantly reduce carbon emissions and material costs while meeting mechanical performance and durability requirements. Attached Figure Description
[0015] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:
[0016] Figure 1 This specification provides a flowchart illustrating a method for optimizing the design of concrete in a chloride-salt environment based on text-enhanced information. Figure 2 This document provides a flowchart for the implementation of a text enhancement deep learning model. Figure 3 A flowchart of an improved non-dominated sorting genetic algorithm provided in this specification; Figure 4 A schematic diagram of another text-enhanced concrete optimization design method for chloride salt environments provided in this specification. Figure 5 This specification provides a schematic diagram of the evolution of the Pareto front in the low-carbon optimization design of concrete raw materials. Figure 6 This specification provides a schematic diagram of the evolution of the Pareto front during the optimized design of the concrete mix proportion for the piers of the Hong Kong-Zhuhai-Macau Bridge. Figure 7 This specification provides a schematic diagram of a computer device for implementing a text-enhanced concrete optimization design method in a chloride-salt environment. Detailed Implementation
[0017] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.
[0018] This invention proposes a text-enhanced concrete optimization design method for chloride-salt environments. This method integrates a text-enhanced deep learning model with improved multi-objective intelligent optimization technology. While meeting the mechanical performance and durability requirements of concrete structures, it minimizes carbon emissions and costs, breaking through the limitations of traditional design that relies on experiments. It has significant advantages in improving concrete design efficiency and promoting concrete decarbonization.
[0019] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.
[0020] Figure 1 This is a flowchart illustrating a text-enhanced concrete optimization design method for chloride-salt environments, as described in this specification. The method includes the following steps: S101, Obtain the concrete compressive strength prediction model and the text-enhanced deep learning model; the compressive strength prediction model is used to predict the compressive strength of concrete based on the concrete raw material mix proportions; the text-enhanced deep learning model is used to predict the chloride ion diffusion coefficient based on the concrete raw material mix proportions and text information; the text information includes the raw material characteristics of concrete, environmental conditions, and microstructural features.
[0021] In this invention, a deep neural network (DNN) model is used as the compressive strength prediction model to predict the compressive strength of concrete, and the compressive strength is used as a constraint index to meet mechanical performance requirements. Considering that the transport behavior of chloride ions in concrete is affected by a variety of factors, and these factors are usually described in text form, a text-enhanced deep learning model is developed to predict the chloride ion distribution, aiming to derive the chloride ion diffusion coefficient for evaluating the durability of concrete.
[0022] In one embodiment, a database containing data on the compressive strength and chloride ion concentration of concrete specimens is established, and textual information on complex factors affecting the diffusion behavior of chloride ions in concrete, including raw material characteristics, environmental conditions, and microstructural features, is collected and processed as input information for the database.
[0023] The DNN-based compressive strength prediction model was trained using 1338 sets of concrete experimental data. It can be divided into nine input features. The first eight input features are the raw material mix design parameters (such as cement content, water-cement ratio, coarse aggregate content, fine aggregate content, silica fume content, fly ash content, blast furnace slag content, and water-reducing agent content). The ninth feature is the curing age.
[0024] The text-enhanced deep learning model described in this embodiment is trained using a dataset containing 2919 experimental samples. The model input features a total of 10 numerical variables, with the first 8 features being raw material mix design parameters, consistent with the compressive strength prediction model. The other two features are chloride ion exposure period and test depth, respectively. Simultaneously, the dataset also incorporates 4 textual information items, including the raw material characteristics, environmental conditions, and microstructural features of concrete. Specifically, these include raw material characteristics, experimental methods, chloride erosion mechanisms, and commentary. The commentary section primarily draws from conclusive descriptions and empirical observations obtained from existing experimental research, summarizing common patterns in chloride ion transport behavior and its influencing factors across different studies. Specific examples are shown in Table 1.
[0025] Table 1. Specific content of the text information S102 aims to minimize the carbon emissions and cost of concrete, using the raw material mix proportions as optimization variables. Under the condition that the raw material mix proportions satisfy conventional and performance constraints, an improved non-dominated sorting genetic algorithm is used for multi-objective optimization to determine the Pareto optimal front, i.e., the feasible solution set of raw material mix proportions. The conventional constraints include the range constraints, proportion constraints, and absolute volume constraints of raw materials; the performance constraints include compressive strength constraints and chloride ion diffusion coefficient constraints. The improved non-dominated sorting genetic algorithm uses a linear annealing mechanism to dynamically adjust the mutation rate during the population mutation process and uses the reflection method to handle parameter out-of-bounds problems.
[0026] In one embodiment, conventional constraints can be divided into three categories: First, range constraints, which set reasonable ranges based on engineering experience or relevant standards to control the amount of raw materials used, such as the allowable content ranges of cement, aggregates, water, and admixtures; second, proportion constraints, such as water-cement ratio and sand-aggregate ratio, used to ensure the workability of concrete, with specific values to be determined based on actual project requirements; and finally, absolute volume constraints, which require the total volume of all aggregates to be controlled. V tol It equals 1 m³.
[0027] In one embodiment, ensuring that the raw material mix proportions meet performance constraints includes: inputting the raw material mix proportions and curing age into a compressive strength prediction model to obtain the compressive strength of the concrete; determining that the raw material mix proportions meet the compressive strength constraints when the compressive strength of the concrete is greater than or equal to the compressive strength required by the structural design; inputting the raw material mix proportions, exposure period, test depth, and text information into a text-enhanced deep learning model to obtain the chloride ion concentration of the concrete; deriving the chloride ion diffusion coefficient based on Fick's second law according to the chloride ion concentration of the concrete; determining that the raw material mix proportions meet the chloride ion diffusion coefficient constraints when the chloride ion diffusion coefficient is less than or equal to the preset maximum chloride ion diffusion coefficient.
[0028] Specifically, the compressive strength constraint is expressed as: (1) in, Indicates the compressive strength required for the structural design; This indicates the raw material mix proportions and curing age of the concrete; and it is used in the compressive strength prediction model. Make predictions; Indicates compressive strength prediction model The predicted compressive strength of concrete is mainly determined by the mix proportions of concrete raw materials and the curing age.
[0029] Chloride ion diffusion coefficient constraint representation: (2) in, It is the maximum chloride ion diffusion coefficient set according to the environmental exposure level and design service life, and is usually specified in relevant specifications and standards; The chloride ion diffusion coefficient is determined by the inherent properties of concrete and its exposed chloride environment. It is derived based on Fick's second law after predicting the spatiotemporal distribution (chloride ion concentration) of chloride ions in concrete using a text-enhanced deep learning model.
[0030] The chloride ion transport behavior in concrete is described by Fick's second diffusion law, namely: (3) (4) in, Indicates the concentration of free chloride ions; Indicates the surface chloride ion concentration; Indicates the initial chloride ion concentration; Indicates the apparent chloride ion diffusion coefficient; This indicates the depth from which the chloride ion concentration is measured at the exposed surface of the concrete; Indicates exposure time; This represents the error function.
[0031] In one embodiment, the stress resistance prediction model includes three hidden layers connected in sequence, with the number of neurons in the hidden layers being 128, 64, and 32, respectively. Each hidden layer includes a fully connected layer, a normalized layer, and a ReLU activation function connected in sequence. During the training of the stress resistance prediction model, the fully connected layer is initialized with Kaiming, the weights of the batch normalized layer are initialized to 1, the biases are initialized to 0, and a dropout layer with a dropout rate of 0.2 is added at the end of the last hidden layer.
[0032] In one embodiment, such as Figure 2 As shown, the text-enhanced deep learning model includes a standardization unit, a natural language processing unit, a concatenation unit, and a deep neural network. The model inputs raw material mix proportions, exposure period, test depth, and text information to obtain the chloride ion concentration of the concrete. This process includes: inputting the raw material mix proportions, exposure period, test depth, and text information into the text-enhanced deep learning model; using the standardization unit to perform Z-score standardization on the raw material mix proportions, exposure period, and test depth to obtain normalized numerical features; using the natural language processing unit to perform text preprocessing, word embedding, and feature extraction operations on the text information to obtain a text feature vector; and using the concatenation unit to concatenate the text feature vector and the normalized numerical features, and then inputting the concatenated features into the deep neural network to obtain the chloride ion concentration of the concrete.
[0033] carbon emissions The calculation formula is: (5) in, This refers to the quantity of raw materials used in concrete. Indicates the first in concrete The mass content of each raw material; Indicates the first in concrete Carbon emission factors of each raw material; Indicates the number of stages in the concrete production process; Indicates the first Energy consumption at each stage of production (such as mixing, vibration, transportation, etc.); Indicates the first Carbon emission factors of the energy type used in each production stage.
[0034] cost The calculation formula is: (6) in, Indicates the first The actual market price per unit weight of raw materials; Indicates the first The unit price of the type of energy used in each stage of production.
[0035] It should be noted that the main source of carbon emissions from concrete is the raw material stage, including the mining, processing, and transportation of raw materials. Carbon emissions during the production stage are relatively small, mainly affected by production equipment and process conditions, and can be ignored in the calculations. Ignoring these carbon emissions will not affect the final optimization results. The cost of concrete consists of raw material costs and production process costs. Production costs mainly depend on production scale and equipment utilization rate, and have no significant correlation with the concrete mix proportion. Therefore, production costs can be ignored in mix proportion optimization design, and only raw material costs are considered as an economic evaluation indicator.
[0036] In one embodiment, under the condition that the candidate raw material proportions meet conventional constraints and performance constraints, an improved Non-dominated Sorting Genetic Algorithm (NSGA-II) is used for multi-objective optimization to determine the Pareto optimal front, such as... Figure 3 As shown, it includes the following steps: S201, within a given range of materials, obtain the parent population that satisfies the general constraints and performance constraints; each individual in the population represents a raw material mix proportion of concrete; the population size is N.
[0037] S202, calculate the carbon emissions and raw material costs for each individual in the parent population.
[0038] S203. Based on the carbon emissions and raw material costs of each individual in the parent population, binary tournament selection, binary crossover, and polynomial mutation operations are performed on the individuals in sequence. If the individual after mutation exceeds the upper and lower bounds of the corresponding raw materials, the corresponding offspring individuals are reflected into the upper and lower bounds of the raw materials using the reflection method to obtain an offspring population that satisfies the conventional constraints and performance constraints.
[0039] The mutation probability of the polynomial mutation is dynamically adjusted based on the linear annealing strategy, and its calculation formula is as follows: , The mutation probability; The maximum mutation probability; The minimum mutation probability; For the current algebra; The maximum algebra is usually the upper limit of the number of iterations set in the algorithm.
[0040] S204: Merge the parent and offspring populations, perform non-dominated sorting and crowding calculation, and select the N individuals with the highest crowding from the parent and offspring populations as the new parent population.
[0041] Specifically, an elite strategy is applied by merging parent and offspring populations. Then, a fast non-dominated sort is performed, assigning a non-dominated rank to each individual. Within each rank, crowding distance is calculated, and individuals with larger crowding distances are preferentially selected to form a new population.
[0042] S205, continue to perform binary tournament selection, binary crossover and polynomial mutation operations on the newly generated parent population until the maximum number of iterations is reached, and take all individuals in the parent population with a non-dominated level of 1 as the Pareto optimal frontier.
[0043] Specifically, based on the new parent population, S202-S204 are repeated until the maximum number of generations is reached or other termination conditions are met, and then the Pareto optimal frontier is output.
[0044] During the generation of the offspring population, all individuals must meet the constraints. Any infeasible individual will be discarded until a qualified offspring population of the same size as the parent population is formed.
[0045] In this embodiment, 1) during the initial population generation stage, a random sampling method is used to generate feasible solutions that satisfy all constraints. This improvement reduces invalid evaluations caused by infeasible individuals, thereby improving the efficiency of initial optimization. 2) A linear annealing mechanism is introduced to dynamically adjust the mutation rate. A higher mutation rate is used in the early stages of evolution to enhance population diversity, and the mutation rate is gradually reduced as iterations progress to improve convergence. 3) During the optimization process, the principle of "feasibility first" is strictly followed, eliminating infeasible individuals that violate constraints. Crossover and mutation operations are only performed on feasible solutions to ensure the feasibility of the solution from the initial stage. 4) A reflection method is used to handle parameter boundary violations, effectively preventing solutions from clustering near variable boundaries, further enhancing the algorithm's exploration capability and improving population diversity within the solution domain.
[0046] S103, by using the approximation ideal solution sorting method to determine the optimal raw material mix proportion of concrete from the feasible solution set of raw material mix proportions.
[0047] Since the solution set generated by the constrained NSGA-II consists of a set of non-dominated solutions satisfying Pareto optimality, there are often trade-offs between the objectives. Therefore, a comprehensive evaluation of cost and carbon emissions is still needed to determine the solution most suitable for engineering applications. To this end, this study introduces the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) to further rank and filter the Pareto front solutions, thereby determining the final optimal solution. The calculation process of TOPSIS is as follows:
[0048] (7) (8) (9) in, Solution h Distance from the ideal solution; Solution h Distance from the negative ideal solution This represents the quantity of the objective function (carbon emissions and cost); and These represent the first and second objectives in single-objective optimization. The optimal and worst values of the objective function. This represents the relative proximity of each solution to the ideal solution and the negative ideal solution, and its value ranges from 0 to 1.
[0049] According to formulas (7)-(9), calculate the feasible solution for each raw material mix proportion in the feasible solution set. value, The solution with the highest value is considered the optimal solution, which is the optimal raw material mix ratio for concrete.
[0050] In one embodiment, this embodiment also provides a method for optimizing the design of concrete in a chloride-salt environment based on text-enhanced information, such as... Figure 4 As shown, this embodiment includes: obtaining candidate mix proportions that meet conventional constraints; obtaining feasible mix proportions that meet performance constraints from the candidate mix proportions; then optimizing the feasible mix proportions with the goal of minimizing carbon emissions and material costs; determining the Pareto optimal frontier; and then outputting the optimal solution, i.e. the optimal raw material mix proportion, through TOPSIS.
[0051] In summary, the method provided by this invention has the following main technical advantages compared with traditional low-carbon concrete design methods: 1. Development of a high-precision prediction model: Considering that the transport behavior of chloride ions in concrete is affected by a variety of factors, and these factors are usually described in text form, a deep learning model based on text information enhancement was constructed to predict the chloride ion distribution, thereby significantly improving the prediction accuracy.
[0052] 2. Construction of an innovative optimization framework: A multi-objective optimization framework for concrete design was established, aiming to meet the mechanical performance and durability requirements of concrete structures while minimizing carbon emissions and costs.
[0053] 3. Improvement of the optimization algorithm: This invention uses an improved NSGA-II algorithm to carry out multi-objective optimization design, and improves the traditional NSGA-II algorithm to meet the strict constraints introduced in the research, thereby effectively improving the convergence performance and global search capability of the algorithm.
[0054] 4. Significant engineering application value: This invention combines deep learning models with multi-objective optimization technology, breaking through the limitations of traditional concrete design that relies on a large number of experimental verifications. It can significantly improve design efficiency and provide new ideas and technical support for realizing low-carbon and intelligent design in the field of concrete.
[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will now be further explained in conjunction with the accompanying drawings, using two typical examples: low-carbon optimization design of concrete raw materials and optimization of the concrete mix proportion of the Hong Kong-Zhuhai-Macau Bridge piers. These examples visually demonstrate the advantages of this method in minimizing carbon emissions and material costs while meeting the mechanical performance and durability requirements of concrete structures. The specific implementation process is as follows:
[0056] 1. Case Study on Low-Carbon Optimization Design of Concrete Raw Materials Step 1: The concrete experimental studies shown in Table 2 are selected as the optimization benchmark. The aim is to explore how to optimize the concrete mix proportions by introducing low-carbon raw materials without using auxiliary cementitious materials. During this process, the compressive strength of the concrete must be maintained above 45 MPa, and its chloride ion diffusion coefficient must not exceed 5 × 10⁻⁶. - ¹² m² / s performance standard.
[0057] Table 2 Concrete mix proportions used in the experimental study Step 2: To meet the mechanical performance requirements of the low-carbon optimized design of concrete raw materials (i.e., compressive strength not less than 45 MPa), this case first establishes a 28-day compressive strength prediction model for concrete as an evaluation tool for mechanical performance constraints. This model is trained based on 1338 sets of concrete experimental data, and the data distribution characteristics are shown in Table 3. Input features include nine variables: cement, water-cement ratio, fine aggregate, coarse aggregate, silica fume, fly ash, blast furnace slag, water-reducing agent, and curing age. The output is a single objective variable—the 28-day compressive strength of concrete (MPa). The model structure uses a three-layer fully connected deep neural network with 128, 64, and 32 neurons in the hidden layers, respectively; the output layer consists of a single node, making it suitable for regression tasks. In each hidden layer, the input data is first linearly transformed (weighted summation) through the fully connected layer, and then normalized through a batch normalization layer, standardizing each feature to a mean of 0 and a variance of 1, thereby stabilizing the training process. Next, the data is activated using the ReLU activation function to introduce nonlinearity to enhance the model's ability to learn complex features. Finally, a dropout layer with a dropout rate of 0.2 was added to prevent overfitting and improve the model's generalization ability. Model parameters were initialized using the default settings implemented in PyTorch: fully connected layers were initialized with Kaiming (He) to accommodate the ReLU activation function. The weights of the batch normalized layers were initialized to 1, and the biases to 0, ensuring preservation of the original feature distribution. The loss function used was mean squared error (MSE) to measure the deviation between predicted and measured strength; the optimizer was Adam, with an initial learning rate of 0.001 to balance convergence speed and accuracy; the batch size was set to 64, and the number of iterations was 500 to ensure sufficient model convergence. This model can quickly predict the 28-day compressive strength of candidate mix proportions, providing a quantitative basis for subsequent judgments on whether the mechanical performance requirements of concrete raw materials are met.
[0058] Table 3 Data Feature Distribution of Compressive Strength Prediction Model Step 3, to meet the durability requirements of the low-carbon optimized design of concrete raw materials (i.e., the chloride ion diffusion coefficient must not exceed 5 × 10⁻⁶), -(¹² m² / s) In this embodiment, a text-enhanced deep learning model is constructed to predict the chloride ion diffusion coefficient of concrete, serving as an evaluation tool for durability constraints. The model is trained using 2912 experimental data points, each containing both numerical and textual features. The distribution of numerical features is shown in Table 4. For numerical features (cement, water-cement ratio, fine aggregate, coarse aggregate, silica fume, fly ash, blast furnace slag, water-reducing agent, exposure period, and test depth), Z-score standardization is used to eliminate dimensional differences and maintain consistent feature scales. Text data is converted into feature vectors using natural language processing techniques, specifically including text preprocessing, word embedding, and feature extraction. Text preprocessing transforms the original sentences into structured input; word embedding occurs in the model's embedding layer, mapping discrete word indices to continuous word vectors; then, feature extraction is performed in subsequent layers of the model, which take the word vector sequence as input to further extract semantic information from the text. Based on previous research, multi-head self-attention (MSA) can capture features from multiple perspectives and support parallel computation, exhibiting high prediction accuracy and fast execution speed. Therefore, the feature extraction layer in the text enhancement deep learning model used in this invention is MSA. Finally, the extracted text feature vectors are concatenated with the normalized numerical features along the feature dimension to form a unified input vector, which is then input into the DNN model for training (see [link to DNN model]). Figure 2 The model uses a batch size of 64, a loss function of MSE, and the Adam optimizer for parameter updates. During the prediction phase, textual information related to concrete raw materials from Table 1 needs to be input into the model to achieve targeted predictions. By introducing textual features, the model can effectively supplement complex influencing factors that are difficult to express with numerical features, thereby significantly improving the adaptability and accuracy of concrete durability prediction.
[0059] Table 4. Numerical data feature distribution in the text-enhanced deep learning model for predicting chloride ion diffusion coefficient. Step 4: To ensure that the concrete raw materials meet the performance standards, it is necessary to set constraints according to the relevant specifications, as detailed in formulas (10)-(12).
[0060] (10) Formula (10) represents the range constraint for each raw material in the concrete mixture. , ~ These represent the contents of cement, fine aggregate, coarse aggregate, silica fume, fly ash, blast furnace slag, and high-efficiency water-reducing agent, respectively.
[0061] (11) Formula (11) represents the proportional constraint in the concrete mixture. Indicates the water-to-glue ratio; Indicates the sand ratio.
[0062] (12) Formula (12) represents the performance constraints in the concrete mixture. It is compressive strength, which can be predicted using the developed compressive strength prediction model. It is the chloride ion diffusion coefficient, which can be calculated using an established text-enhanced deep learning model.
[0063] Step 5: Combining the established compressive strength prediction model and chloride ion diffusion coefficient prediction model, and using the improved NSGA-II algorithm, concrete mix proportions that meet performance constraints are screened. Simultaneously, the carbon emissions and material costs of each candidate scheme are calculated based on the data in Table 5. The specific process is as follows: First, 300 initial individuals satisfying all constraints are generated through real-number encoding and floating-point random sampling. Then, a binary tournament selection strategy is used for individual selection, and the SBX operator is used for crossover operations, with the crossover probability and distribution index set to 0.9 and 5, respectively. In addition, multinomial mutation is employed. The mutation probability is dynamically adjusted through a linear annealing strategy, linearly decreasing from 0.8 to 0.1, and the mutation distribution index is set to 5. In each iteration, 300 offspring individuals satisfying the constraints are generated. Then, the current parent and offspring populations are merged, and an elite strategy is applied. The merged population is sorted and selected using fast non-dominated sorting and crowding distance calculation, retaining the top 300 individuals as the new generation. After 100 iterations, the algorithm converged, ultimately obtaining 103 Pareto optimal solutions. Figure 5 This study demonstrates the evolution of the Pareto frontier in the process of low-carbon optimization design of concrete raw materials.
[0064] Table 5. Information related to concrete raw materials Step 6: The obtained Pareto front solutions were ranked using the TOPSIS method to determine the final optimal concrete mix design. To verify the optimization effect, the optimization results of this study were compared with concrete mix designs proposed by existing technologies (see Table 2, with carbon emissions and costs of 369.08 kgCO2 / m³ and ¥333.83 / m³, respectively). Table 6 lists three representative schemes: scheme A with the lowest cost, scheme C with the lowest carbon emissions, and scheme B, the comprehensive optimal scheme determined based on the TOPSIS method. The results show that, compared with the original mix design, the optimized scheme B achieves a significant improvement in carbon emissions (reduced by 20.35%) and cost (reduced by 12.18%) while meeting performance requirements.
[0065] Table 6 Detailed mix proportions of optimized concrete 2. Case Study on Optimized Concrete Mix Design for Hong Kong-Zhuhai-Macau Bridge Piers Step 1: To ensure that the concrete meets the 120-year design service life requirement of the Hong Kong-Zhuhai-Macau Bridge, the compressive strength of the pier concrete must reach 50 MPa, and the chloride ion diffusion coefficient must be controlled within 3.5 × 10⁻¹² m² / s.
[0066] Step 2 employs a 28-day compressive strength prediction model consistent with the low-carbon optimization design case of concrete raw materials, as well as a text-enhanced deep learning prediction model for the chloride ion diffusion coefficient of concrete. These models provide quantitative basis for subsequent judgments on whether candidate mix proportions meet the design requirements for the mechanical performance and durability of the Hong Kong-Zhuhai-Macau Bridge piers.
[0067] Step 3: To ensure the 120-year design service life of the Hong Kong-Zhuhai-Macau Bridge, constraints need to be set according to relevant specifications, as detailed in formulas (13)-(15).
[0068] (13) in, Represents the total content of cementitious materials; This indicates the corresponding moisture content.
[0069] (14) Formula (14) represents the proportional constraint in the concrete mixture. This indicates the proportion of auxiliary cementitious materials in the total cementitious material content.
[0070] (15) Step 4: Using the same prediction model and optimization algorithm as the low-carbon optimization design case for concrete raw materials, and inputting the text information related to the Hong Kong-Zhuhai-Macau Bridge pier structure as shown in Table 7 into the text-enhanced deep learning model for chloride ion diffusion coefficient, a Pareto front solution for the concrete optimization problem of this project is generated (see...). Figure 6Subsequently, the Pareto front solutions were ranked and optimized using the TOPSIS method to determine the final optimal mix proportion scheme. To verify the optimization effect, the optimization results of this study were compared with existing bridge pier concrete mix proportions (see Table 8, carbon emissions of 226.64 kgCO2 / m³, cost of ¥320.54 / m³). Table 9 shows three representative schemes: scheme A with the lowest cost, scheme C with the lowest carbon emissions, and scheme B, the comprehensive optimal scheme determined based on the TOPSIS method. The results show that, under the premise of simultaneously meeting the design requirements of compressive strength and durability, the optimal scheme B obtained by this study based on the TOPSIS method can achieve a significant optimization effect of reducing carbon emissions by 17.34% and reducing costs by 3%.
[0071] Table 7 Textual information related to the pier structure of the Hong Kong-Zhuhai-Macau Bridge Table 8 Concrete mix proportions for the Hong Kong-Zhuhai-Macau Bridge piers proposed in existing studies Table 9 Concrete Mix Proportion Scheme for the Hong Kong-Zhuhai-Macau Bridge The execution subject of the method provided by this invention can be a server, which can be a server set up on a business platform, or a device such as a desktop computer or laptop computer that can execute the solution in this specification.
[0072] When applying the text-enhanced concrete optimization design method for chloride-salt environments provided in this manual, it is not necessary to consider... Figure 1 The steps shown are executed in sequence. The specific execution order of each step can be determined as needed, and this manual does not impose any restrictions on it.
[0073] The above describes one or more embodiments of the method for optimizing concrete design in a chloride-salt environment based on text information enhancement. Based on the same idea, this specification also provides a corresponding device for optimizing concrete design in a chloride-salt environment based on text information enhancement, which includes: The acquisition module is used to acquire the concrete compressive strength prediction model and the text-enhanced deep learning model. The compressive strength prediction model is used to predict the compressive strength of concrete based on the raw material mix proportions of the concrete. The text-enhanced deep learning model is used to predict the chloride ion diffusion coefficient based on the raw material mix proportions of the concrete and text information. The text information includes the raw material characteristics, environmental conditions, and microstructural features of the concrete. The optimization module aims to minimize the carbon emissions and raw material costs of concrete, using the concrete mix proportions as optimization variables. Under the condition that the raw material mix proportions satisfy conventional and performance constraints, an improved non-dominated sorting genetic algorithm is employed for multi-objective optimization to determine the Pareto optimal front, i.e., the feasible solution set of the raw material mix proportions. Conventional constraints include range constraints, proportion constraints, and volume constraints of raw materials; performance constraints include compressive strength constraints and chloride ion diffusion coefficient constraints. The improved non-dominated sorting genetic algorithm dynamically adjusts the mutation rate during the population mutation process using a linear annealing mechanism and employs a reflection method to handle parameter out-of-bounds issues. The determination module is used to determine the optimal raw material mix proportion of concrete from the feasible solution set of raw material mix proportions by using the approximation ideal solution sorting method.
[0074] Specific limitations regarding the text-enhanced concrete optimization design device for chloride-enhanced environments can be found in the above-mentioned limitations on the text-enhanced concrete optimization design method for chloride-enhanced environments, and will not be repeated here. Each module in the aforementioned text-enhanced concrete optimization design device for chloride-enhanced environments can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.
[0075] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 A text-enhanced method for optimizing the design of concrete in chloride-salt environments is provided.
[0076] This instruction manual also provides Figure 7 The schematic diagram of the computer device shown is as follows: Figure 7 At the hardware level, the computer device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above-mentioned functions. Figure 1 A text-enhanced method for optimizing the design of concrete in chloride-salt environments is provided.
[0077] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.
[0078] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
Claims
1. A method for optimizing the design of concrete in a chloride-salt environment based on text-enhanced information, characterized in that, include: Obtain a concrete compressive strength prediction model and a text-enhanced deep learning model; The compressive strength prediction model is used to predict the compressive strength of concrete based on the raw material mix proportions; the text-enhanced deep learning model is used to predict the chloride ion diffusion coefficient based on the raw material mix proportions and text information of concrete; the text information includes the raw material characteristics, environmental conditions, and microstructural features of concrete. With the optimization objectives of minimizing concrete carbon emissions and cost, and the raw material mix proportions as optimization variables, an improved non-dominated sorting genetic algorithm is used for multi-objective optimization under the condition that the raw material mix proportions meet conventional and performance constraints. This algorithm determines the Pareto optimal front, i.e., the feasible solution set of the raw material mix proportions. Conventional constraints include the range, proportion, and absolute volume constraints of the raw materials; performance constraints include compressive strength constraints and chloride ion diffusion coefficient constraints. The improved non-dominated sorting genetic algorithm uses a linear annealing mechanism to dynamically adjust the mutation probability during the population mutation process and employs a reflection method to handle parameter out-of-bounds issues. The optimal raw material mix proportion for concrete is determined from the feasible solution set of raw material mix proportions by using the approximation ideal solution sorting method.
2. The method according to claim 1, characterized in that, The raw material proportions must meet the performance constraints, including: The raw material mix proportions and curing age are input into the compressive strength prediction model to obtain the compressive strength of the concrete; when the compressive strength of the concrete is greater than or equal to the compressive strength required by the structural design, it is determined that the raw material mix proportions meet the compressive strength constraints. The raw material mix proportions, exposure period, test depth, and text information are input into a text-enhanced deep learning model to obtain the chloride ion concentration in the concrete. Based on the chloride ion concentration in concrete, the chloride ion diffusion coefficient is derived using Fick's second law. When the chloride ion diffusion coefficient is less than or equal to the preset maximum chloride ion diffusion coefficient, the raw material mix proportion is determined to meet the chloride ion diffusion coefficient constraint.
3. The method according to claim 2, characterized in that, The stress resistance prediction model consists of three hidden layers connected in sequence, with 128, 64, and 32 neurons in the hidden layers, respectively. Each hidden layer includes a fully connected layer, a normalized layer, and a ReLU activation function connected in sequence. During the training of the stress resistance prediction model, the fully connected layer is initialized with Kaiming, the weights of the batch normalized layer are initialized to 1, the biases are initialized to 0, and a dropout layer with a dropout rate of 0.2 is added at the end of the last hidden layer.
4. The method according to claim 2, characterized in that, The text-enhanced deep learning model includes normalization units, natural language processing units, splicing units, and deep neural networks. Raw material mix proportions, exposure periods, testing depth, and text information are input into the text-enhanced deep learning model to obtain the chloride ion concentration of the concrete, including: The raw material mix ratio, exposure period, test depth and text information are input into the text enhancement deep learning model. The raw material mix ratio, exposure period and test depth are Z-score standardized by the standardized unit to obtain normalized numerical features. The text information is preprocessed, word embedded, and feature extracted sequentially by the natural language processing unit to obtain the text feature vector. The text feature vector and the normalized numerical features are concatenated by a concatenation unit, and the concatenated features are then input into a deep neural network to obtain the chloride ion concentration of the concrete.
5. The method according to claim 1, characterized in that, carbon emissions The calculation formula is: in, This refers to the quantity of raw materials used in concrete. Indicates the first in concrete The mass content of each raw material; Indicates the first in concrete Carbon emission factors of each raw material; Indicates the number of stages in the concrete production process; Indicates the first Energy consumption at each stage of production; Indicates the first Carbon emission factors of energy types used in each production stage; cost The calculation formula is: in, Indicates the first The actual market price per unit weight of raw materials; Indicates the first The unit price of the type of energy used in each stage of production.
6. The method according to claim 1, characterized in that, Under the condition that the raw material mix proportions meet conventional and performance constraints, an improved non-dominated sorting genetic algorithm is used for multi-objective optimization to determine the Pareto optimal front, including: Within a given range of materials, obtain a parent population that satisfies both general and performance constraints; each individual in the population represents a raw material mix proportion for concrete; the population size is N. Calculate the carbon emissions and raw material costs for each individual in the parent population; Based on the carbon emissions and raw material costs of each individual in the parent population, binary tournament selection, binary crossover, and polynomial mutation operations are performed on the individuals in sequence. If an individual after mutation exceeds the upper and lower bounds of the corresponding raw material, the corresponding offspring individual is reflected back into the upper and lower bounds of the raw material using the reflection method, so as to obtain an offspring population that satisfies the conventional constraints and performance constraints. Merge the parent and offspring populations, perform non-dominated sorting and crowding calculation, and select the N individuals with the highest crowding from the parent and offspring populations as the new parent population. Continue performing binary tournament selection, binary crossover, and polynomial mutation operations on the newly generated parent population until the maximum number of iterations is reached. Use all individuals in the parent population with a non-dominated level of 1 as the Pareto optimal frontier.
7. A concrete optimization design device based on text-enhanced information in a chloride-salt environment, characterized in that, include: The acquisition module is used to acquire the concrete compressive strength prediction model and the text-enhanced deep learning model; The compressive strength prediction model is used to predict the compressive strength of concrete based on the raw material mix proportions; the text-enhanced deep learning model is used to predict the chloride ion diffusion coefficient based on the raw material mix proportions and text information of concrete; the text information includes the raw material characteristics, environmental conditions, and microstructural features of concrete. The optimization module aims to minimize the carbon emissions and raw material costs of concrete, using the concrete mix proportions as optimization variables. Under the condition that the raw material mix proportions satisfy conventional and performance constraints, an improved non-dominated sorting genetic algorithm is employed for multi-objective optimization to determine the Pareto optimal front, i.e., the feasible solution set of the raw material mix proportions. Conventional constraints include range constraints, proportion constraints, and volume constraints of raw materials; performance constraints include compressive strength constraints and chloride ion diffusion coefficient constraints. The improved non-dominated sorting genetic algorithm dynamically adjusts the mutation rate during the population mutation process using a linear annealing mechanism and employs a reflection method to handle parameter out-of-bounds issues. The determination module is used to determine the optimal raw material mix proportion of concrete from the feasible solution set of raw material mix proportions by using the approximation ideal solution sorting method.
8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method described in any one of claims 1 to 6.
9. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any one of claims 1 to 6.