Intelligent concrete mix proportion dynamic regulation and control method and system based on multi-objective optimization

Abstract

The invention relates to an intelligent concrete mix proportion dynamic regulation and control method and system based on multi-objective optimization. The method comprises the following steps: acquiring a performance target parameter, a construction material performance parameter and a construction environment parameter associated with a current construction task; constructing a multi-objective optimization function according to the performance objective parameters; based on a multi-objective optimization function, inputting the performance objective parameters and the construction material performance parameters into a pre-trained multi-fidelity Bayesian joint optimization model to obtain a plurality of candidate mix proportions; and performing robustness disturbance planning on each candidate mix proportion according to the construction environment parameters, and determining the candidate mix proportion meeting the performance robustness and target tradeoff requirements as the construction concrete mix proportion. By the adoption of the method, under the condition that multiple requirements of strength, workability, economical efficiency and environmental protection performance are guaranteed, the concrete mixing proportion with high adaptability and controllable risk is dynamically provided for different construction tasks, and therefore the stability of engineering quality and the sustainability of construction are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of material engineering, and particularly relates to an intelligent concrete mix proportion dynamic regulation method and system based on multi-objective optimization. BACKGROUND

[0002] With the development of material science and engineering technology, material mix proportion optimization technology appears, which realizes the improvement of target performance by adjusting the types and proportions of raw materials, and has the characteristics of short research and development cycle, controllable cost, and customizable performance.

[0003] In traditional technology, machine learning, Bayesian optimization and other methods are used for material mix proportion design, which can obtain a material formula with better performance under the condition of fewer experimental samples. The optimization of material mix proportion usually relies on a low-fidelity performance prediction model to obtain prediction results at a lower experimental cost, and then selects some mix proportions for high-fidelity experimental verification according to the predicted performance indicators. This method can reduce the number of high-cost experiments to a certain extent and improve the research and development efficiency.

[0004] However, the current material mix proportion optimization method still has problems. There is an inevitable accuracy gap between the low-fidelity prediction model and the high-fidelity experimental results, resulting in a deviation between the predicted value and the true performance. Traditional methods often rely on a fixed initial candidate mix proportion set in iterative optimization, lack a mechanism for dynamically updating the initial candidate mix proportion in the optimization process, and limit the expansion of the search space and the efficiency of obtaining the global optimal solution. The utilization rate of high-fidelity verification data in model updating and acquisition function optimization is insufficient, and it is difficult to fully play its role in improving prediction accuracy and optimization efficiency. SUMMARY

[0005] Therefore, it is necessary to provide an intelligent concrete mix proportion dynamic regulation method and system based on multi-objective optimization which can combine low-fidelity prediction and high-fidelity verification.

[0006] In a first aspect, the application provides an intelligent concrete mix proportion dynamic regulation method based on multi-objective optimization, comprising:

[0007] obtaining performance target parameters, construction material performance parameters and construction environment parameters associated with a current construction task; the performance target parameters include target strength grade, workability index and environmental constraints;

[0008] constructing a multi-objective optimization function according to the performance target parameters; the multi-objective optimization function includes strength error minimization, workability deviation control, carbon footprint minimization and material cost minimization;

[0009] Based on a multi-objective optimization function, the performance objective parameters and construction material performance parameters are input into a pre-trained multi-fidelity Bayesian joint optimization model to obtain multiple candidate mix proportions.

[0010] Based on the construction environment parameters, robust disturbance planning is performed on each candidate mix proportion to determine the candidate mix proportion that meets the requirements of performance robustness and target trade-off as the construction concrete mix proportion.

[0011] In one embodiment, based on a multi-objective optimization function, performance objective parameters and construction material performance parameters are input into a pre-trained multi-fidelity Bayesian joint optimization model to obtain multiple candidate mix proportions, including:

[0012] The design space is obtained based on the performance target parameters and the performance parameters of the construction materials; the design space includes multiple design variables with clear boundaries; the design variables include water-cement ratio, cement dosage, fly ash content, sand ratio, and water-reducing agent content;

[0013] With the goal of converging the multi-fidelity Bayesian acquisition function to satisfy the multi-objective optimization function, the multi-fidelity Bayesian acquisition function is used to explore the values ​​of design variables in the design space and obtain multiple candidate mix proportions.

[0014] In one embodiment, with the goal of converging the multi-fidelity Bayesian acquisition function to satisfy a multi-objective optimization function, the multi-fidelity Bayesian acquisition function is used to explore the values ​​of design variables in the design space to obtain multiple candidate mix proportions, including:

[0015] A multi-fidelity Bayesian acquisition function is used to explore the values ​​of design variables in the design space to obtain initial candidate mix proportions; the initial candidate mix proportions include material properties and material ratios.

[0016] The performance indicators of each initial candidate mix proportion are predicted, and the performance indicator prediction results are obtained.

[0017] If the performance index prediction results do not meet the multi-objective optimization function, then optimize the multi-fidelity Bayesian acquisition function based on the prediction results, and use the optimized multi-fidelity Bayesian acquisition function to redetermine the initial candidate combination ratio.

[0018] If the performance index prediction results converge to the multi-objective optimization function, the initial candidate combination ratio of the corresponding multi-fidelity Bayesian acquisition function is determined as the candidate combination ratio.

[0019] In one embodiment, performance index prediction is performed on each initial candidate mix proportion to obtain performance index prediction results, including:

[0020] The performance index of each initial candidate mix ratio is predicted by a physical information neural network, and a performance prediction matrix is ​​obtained; the performance prediction matrix includes the performance index corresponding to each initial candidate mix ratio.

[0021] Based on a multi-objective optimization function, high-fidelity Bayesian optimization is used to identify high potential and verify high-fidelity performance prediction matrices, resulting in performance index predictions.

[0022] In one embodiment, based on a multi-objective optimization function, high-fidelity Bayesian optimization is used to perform high-potential identification and high-fidelity verification on the performance prediction matrix to obtain performance index prediction results, including:

[0023] Each performance index in the performance prediction matrix is ​​input into a multi-objective optimization function to calculate the degree of target achievement, and the performance achievement evaluation result is obtained.

[0024] Based on the strategy of maximizing the improvement value, a high-potential solution with better multi-objectives is determined from multiple initial candidate mix proportions according to the performance achievement evaluation results, and multiple high-potential mix proportion solutions are obtained.

[0025] High-fidelity simulations were performed on each high-potential mix design to obtain the predicted performance indicators of each initial candidate mix design.

[0026] In one embodiment, robust disturbance planning is performed on each candidate mix proportion based on construction environment parameters to determine the candidate mix proportion that meets the requirements of performance robustness and target trade-off as the construction concrete mix proportion, including:

[0027] Based on the construction environment parameters, a disturbance space is constructed for each candidate mix proportion; the disturbance space includes the disturbance amplitude, the disturbance direction, and the performance-sensitive parameters corresponding to the disturbance direction;

[0028] Based on interval fuzzy rules, multiple perturbation samples corresponding to candidate combination ratios are generated according to the perturbation space;

[0029] Each perturbation sample is input into a physical information neural network to predict its robustness performance, thereby obtaining the perturbation sample performance of the corresponding perturbation sample.

[0030] If the performance of the perturbation samples meets the requirements of the multi-objective optimization function within the preset tolerance range, the corresponding candidate mix proportion is determined as the construction concrete mix proportion.

[0031] In one embodiment, the multi-fidelity Bayesian joint optimization model is constructed using the following method:

[0032] A training set was constructed by acquiring a large amount of historical construction mix proportion data and corresponding measured performance index data;

[0033] A loss function is jointly constructed by multi-objective performance prediction error and physical constraint residual. The physical information neural network is trained end-to-end using the training set to obtain a physical information neural network with physical interpretability performance prediction capability. The physical information neural network includes mechanical constitutive equations and material aging constraint equations, and embeds mechanical equation constraint terms and engineering empirical rules as physical loss terms.

[0034] Acquire a large amount of experimental data on mixing ratios with different fidelity, and train a high-fidelity validation network based on the experimental data on mixing ratios;

[0035] Based on the collaborative kernel function, transfer modeling between data of different fidelity is realized, and a multifidelity Gaussian process network is obtained.

[0036] With the goal of maximizing the expected improvement of candidate solutions, a Bayesian optimizer is trained to obtain a multi-fidelity Bayesian acquisition function with filtering capabilities;

[0037] A multi-fidelity Bayesian joint optimization model is obtained by sequentially connecting a multi-fidelity Bayesian acquisition function, a physical information neural network, a high-fidelity verification network, and a multi-fidelity Gaussian process network.

[0038] Secondly, this application also provides an intelligent concrete mix proportion dynamic control system based on multi-objective optimization, comprising:

[0039] The data acquisition module is used to acquire the performance target parameters, construction material performance parameters, and construction environment parameters associated with the current construction task; the performance target parameters include the target strength level, workability index, and environmental constraints.

[0040] The optimization objective module is used to construct a multi-objective optimization function based on the performance objective parameters. The multi-objective optimization function includes minimizing strength error, controlling workability deviation, minimizing carbon footprint, and minimizing material cost.

[0041] The optimization module is used to input performance target parameters and construction material performance parameters into a pre-trained multi-fidelity Bayesian joint optimization model based on a multi-objective optimization function to obtain multiple candidate mix proportions.

[0042] The disturbance verification module is used to perform robust disturbance planning on each candidate mix proportion based on construction environment parameters, and determine the candidate mix proportion that meets the requirements of performance robustness and target trade-off as the construction concrete mix proportion.

[0043] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the steps of the above-mentioned intelligent concrete mix proportion dynamic control method based on multi-objective optimization.

[0044] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the above-described intelligent concrete mix proportion dynamic control methods based on multi-objective optimization.

[0045] The aforementioned intelligent concrete mix proportion dynamic control method and system based on multi-objective optimization comprehensively considers strength, workability, cost, and carbon emissions, avoiding the performance imbalance problem caused by traditional single-index optimization schemes. By integrating high- and low-fidelity experimental and simulation data through a multi-fidelity Bayesian optimization model, high prediction accuracy of performance indicators is achieved. Robust perturbation planning enables the mix proportion to adapt to changes in different construction site environments, reducing fluctuations in construction quality. The combination of a pre-trained model and multi-objective optimization realizes automated decision-making from parameter input to mix proportion output, reducing manual trial and error time. The objective optimization function and model contain explicit physical constraints and objective terms, improving the transparency of the results and ultimately rapidly generating multiple feasible concrete mix proportion schemes. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of the process of the intelligent concrete mix proportion dynamic control method based on multi-objective optimization of the present invention;

[0048] Figure 2 This is a flowchart illustrating the steps of step S103.

[0049] Figure 3 This is a flowchart illustrating the steps of step S202.

[0050] Figure 4 This is a structural diagram of the intelligent concrete mix proportion dynamic control system based on multi-objective optimization according to the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0052] In one embodiment, such as Figure 1As shown, a method for dynamic control of intelligent concrete mix proportions based on multi-objective optimization is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, or to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0053] S101. Obtain the performance target parameters, construction material performance parameters, and construction environment parameters associated with the current construction task; the performance target parameters include the target strength level, workability index, and environmental constraints.

[0054] In illustrative terms, performance target parameters refer to the engineering performance goals expected to be achieved during concrete preparation, including target strength grade, workability indicators, and environmental constraints. The target strength grade reflects the compressive strength standard that concrete can achieve within a specified age, determining the structure's load-bearing capacity and safety factor. Workability indicators reflect the concrete's fluidity, ease of use, and water retention during construction, typically related to quantifiable indicators such as slump and spread. Environmental constraints are limitations imposed by external factors such as climate, humidity, temperature, and transportation conditions at the construction site, such as requirements for controlling early water loss rates under high-temperature conditions or the need for frost resistance under cold conditions.

[0055] Construction material performance parameters refer to the physical, chemical, and mechanical properties of various raw materials used in concrete preparation, including cement, sand, aggregate, admixtures, and mineral admixtures. These parameters typically cover specific surface area, chemical composition, water absorption, mud content, and particle size distribution, providing fundamental material characteristic information for subsequent mix design optimization. Construction environment parameters reflect real-time external variables that may affect concrete performance under on-site construction conditions, such as ambient temperature, humidity, wind speed, altitude, and construction sequence. Optionally, data can be obtained through on-site sensor collection, meteorological data interfaces, or manual testing.

[0056] S102. Construct a multi-objective optimization function based on the performance target parameters; the multi-objective optimization function includes minimizing strength error, controlling workability deviation, minimizing carbon footprint, and minimizing material cost.

[0057] Specifically, a multi-objective optimization function is a mathematical model that can simultaneously consider multiple performance evaluation indicators, including four main optimization objectives: minimizing strength error, controlling workability deviation, minimizing carbon footprint, and minimizing material cost. Minimizing strength error aims to make the measured strength of the prepared concrete as close as possible to the design target strength, reducing the risk of quality fluctuations. Controlling workability deviation ensures that the difference between the actual flowability index and the design value is minimized, reducing problems such as pump blockage, segregation, or bleeding. Minimizing carbon footprint aims to reduce the proportion of high-energy-consuming and high-carbon-emission components in material selection and proportioning, thus conforming to the concept of green building. Minimizing material cost takes into account economic efficiency, ensuring that the mix design has a low procurement cost while meeting performance requirements. For example, the multi-objective optimization function can be constructed using multi-objective optimization strategies such as weighted summation and Pareto optimization. The weights can be adjusted based on project priorities, owner requirements, or regulatory standards to reflect the performance focus in different engineering contexts.

[0058] S103. Based on the multi-objective optimization function, the performance objective parameters and construction material performance parameters are input into the pre-trained multi-fidelity Bayesian joint optimization model to obtain multiple candidate mix proportions.

[0059] Indicatively, a multi-fidelity Bayesian joint optimization model is a statistical optimization model that combines high-fidelity and low-fidelity data sources. High-fidelity data typically comes from real experiments or precise numerical simulations, while low-fidelity data comes from simplified computational models or historical engineering experience databases. Bayesian optimization establishes a probability distribution model between input variables and the objective function. In each iteration, it selects the most promising mix design based on confidence intervals and a multi-fidelity Bayesian data collection function, thereby obtaining near-optimal results with fewer trials.

[0060] S104. Based on the construction environment parameters, robust disturbance planning is performed on each candidate mix proportion to determine the candidate mix proportion that meets the requirements of performance robustness and target trade-off as the construction concrete mix proportion.

[0061] Optionally, robust perturbation planning is performed on each candidate mix design based on construction environment parameters. This process simulates the impact of external environmental fluctuations on concrete performance, such as changes in hydration rate due to temperature variations, increased shrinkage risk due to humidity changes, and workability losses due to extended transportation time. By introducing a perturbation factor, multiple rounds of performance simulation and deviation evaluation are conducted on the candidate mix designs to determine which schemes can maintain performance stability under actual construction conditions. Specifically, if the performance deviation of a mix design under perturbation conditions exceeds a set threshold, it is deemed to have insufficient robustness and is eliminated. Finally, the mix designs that still meet the performance robustness requirements under environmental perturbation and achieve a reasonable trade-off across multiple performance indicators are selected as the concrete mix designs for actual construction.

[0062] In the aforementioned intelligent concrete mix proportion dynamic control method based on multi-objective optimization, the introduction of multi-dimensional index constraints into the performance objective parameters comprehensively reflects construction needs, ensuring not only the mechanical properties of concrete but also considering construction operability and environmental adaptability, thus improving the practicality and adaptability of the scheme. The multi-objective optimization function can achieve a comprehensive balance between performance, economy, and environmental impact during the design phase, avoiding significant deviations of other indicators from the target due to single performance optimization. A pre-trained multi-fidelity Bayesian joint optimization model is used to generate candidate mix proportions. Utilizing the transfer modeling capability of data at different fidelity levels significantly reduces experimental costs while improving mix proportion prediction accuracy and generalization ability. Robust perturbation planning, combined with construction environment parameters, pre-considers the impact of environmental fluctuations such as temperature and humidity on concrete performance, improving the stability and controllability of the mix proportion under different construction conditions. The final mix proportion selection mechanism based on performance robustness and objective trade-offs avoids selecting mix proportions solely based on the optimal theoretical value, ensuring that multi-objective requirements are still met in practical applications, thereby reducing construction risks.

[0063] In one embodiment, such as Figure 2 As shown, based on a multi-objective optimization function, the performance objective parameters and construction material performance parameters are input into a pre-trained multi-fidelity Bayesian joint optimization model to obtain multiple candidate mix proportions, including:

[0064] S201. The design space is obtained based on the performance target parameters and the performance parameters of the construction materials. The design space includes multiple design variables with clear boundaries. The design variables include the water-cement ratio, cement dosage, fly ash content, sand ratio, and water-reducing agent content.

[0065] Indicatively, based on the input performance target parameters, including compressive strength, flexural strength, and workability requirements, as well as the performance parameters of construction materials, including cement strength grade, fly ash fineness, sand moisture content, and water-reducing agent water reduction rate, and in conjunction with preset industry standards and engineering experience databases, the value range of each design variable is limited. Furthermore, while determining the value range, the constraints between design variables are established. For example, when the water-cement ratio increases, the water-reducing agent dosage needs to be adjusted accordingly to maintain fluidity; when the fly ash dosage increases, the cement dosage needs to be adjusted simultaneously to ensure early strength. When the water-cement ratio is limited to a range of 0.28-0.40, and when the water-cement ratio is greater than 0.35, the sand ratio needs to be controlled between 38% and 42% to meet the dual constraints of slump and bleeding rate.

[0066] The above method forms a multidimensional design space with multiple clear boundaries and variable constraints, providing a feasible region for subsequent multi-fidelity Bayesian optimization.

[0067] S202. With the goal of the multi-fidelity Bayesian acquisition function converging to satisfy the multi-objective optimization function, the multi-fidelity Bayesian acquisition function is used to explore the values ​​of design variables in the design space to obtain multiple candidate mix proportions.

[0068] Specifically, the design space is input as the search domain into the multi-fidelity Bayesian joint optimization model, with the termination condition being the convergence of the multi-fidelity Bayesian acquisition function value to satisfy the multi-objective optimization function. In each iteration, the multi-fidelity Bayesian acquisition function calculates the expected improvement value of each combination of design variables based on the currently known high-fidelity and low-fidelity data, and selects the optimal candidate points for simulation evaluation or low-fidelity verification. Based on the verification results, the posterior distribution of the multi-fidelity Bayesian model is dynamically updated, thereby adjusting the shape and search direction of the multi-fidelity Bayesian acquisition function. For example, in each iteration, new candidate combination ratios are generated, multi-fidelity evaluated, and the evaluation results are fed back to the model; when the improvement rate of the multi-fidelity Bayesian acquisition function is lower than a preset threshold or the performance objective is met, the loop terminates, resulting in multiple candidate combination ratios that meet the conditions.

[0069] In one embodiment, such as Figure 3 As shown, with the goal of converging the multi-fidelity Bayesian data acquisition function to satisfy the multi-objective optimization function, the multi-fidelity Bayesian data acquisition function is used to explore the values ​​of design variables in the design space, resulting in multiple candidate mix proportions, including:

[0070] S301. Use a multi-fidelity Bayesian acquisition function to explore the values ​​of design variables in the design space to obtain the initial candidate mix proportions; the initial candidate mix proportions include material properties and material ratios.

[0071] This illustration demonstrates how the iterative optimization process of the mix proportion is driven by the goal of converging a multi-fidelity Bayesian sampling function to satisfy a multi-objective optimization function. Specifically, the multi-fidelity Bayesian sampling function performs a global search and local refinement exploration of the possible values ​​of each design variable within the defined design space. It expands the search range through low-cost samples and ultimately obtains an initial candidate mix proportion through multiple rounds of sampling and updating. This initial candidate mix proportion includes not only material property parameters such as material type and particle size distribution, but also proportion parameters of cementitious materials, aggregates, water, and admixtures, thus comprehensively characterizing a feasible material mix design.

[0072] S302. Predict the performance indicators of each initial candidate mix proportion and obtain the performance indicator prediction results.

[0073] Furthermore, performance indicators are predicted based on a pre-built performance prediction model. These performance indicators typically cover multiple aspects of the material's performance in actual engineering, including compressive strength, flexural strength, elastic modulus, durability, shrinkage rate, flowability, and long-term stability. During this process, the prediction model utilizes historical experimental data and numerical simulation results, combined with high-fidelity and low-fidelity data fusion algorithms, to quantitatively evaluate the expected performance of the initial candidate mix proportions and output the predicted performance indicators.

[0074] S303. If the performance index prediction results do not meet the multi-objective optimization function, then optimize the multi-fidelity Bayesian acquisition function based on the prediction results, and use the optimized multi-fidelity Bayesian acquisition function to redetermine the initial candidate combination ratio.

[0075] Specifically, the prediction results are compared with a pre-defined multi-objective optimization function to determine the merits of the current candidate combinations. The multi-objective optimization function comprehensively considers multiple performance indicators and evaluates the overall fitness of candidate solutions through weighted or Pareto front analysis. If any key performance indicator in the prediction results fails to reach a set threshold, or if the overall fitness fails to enter the target region, it is determined that the multi-objective optimization requirements are not met. Furthermore, based on the prediction results, the parameters of the performance prediction model are re-anchored, and the parameters of the multi-fidelity Bayesian data acquisition function are adjusted and updated. Adjustments include increasing the search weight of variables related to unmet performance indicators, narrowing the range of unnecessary variable values, or introducing more targeted high-fidelity sample data, thereby guiding the multi-fidelity Bayesian data acquisition function to re-explore new candidate combinations in the design space.

[0076] S304. If the performance index prediction results converge to the multi-objective optimization function, determine the initial candidate mix ratio of the corresponding multi-fidelity Bayesian acquisition function as the candidate mix ratio.

[0077] When the performance index prediction results tend to stabilize after multiple rounds of iterative optimization, and all key performance indicators meet the constraints of the multi-objective optimization function, and the overall fitness is within the target region, it can be determined that the multi-fidelity Bayesian acquisition function has converged, and the corresponding initial candidate mix proportion will be determined as the final candidate mix proportion.

[0078] In one embodiment, performance index prediction is performed on each initial candidate mix proportion to obtain performance index prediction results, including:

[0079] S41. The performance indicators of each initial candidate mix ratio are predicted by the physical information neural network to obtain the performance prediction matrix; the performance prediction matrix includes the performance indicators corresponding to each initial candidate mix ratio.

[0080] In a schematic representation, the material property parameters and material proportion parameters of each initial candidate mix proportion are used as input vectors and fed into the input layer of a Physics-Informed Neural Network (PINN). The PINN not only contains a deep, fully connected network structure for learning data distribution characteristics, but also explicitly embeds the physical conservation equations, state equations, and empirical constitutive relations of material properties into the network's constraint layers. This ensures that the network simultaneously meets the dual requirements of physical consistency and data-driven accuracy during training and inference. After multi-layer nonlinear transformations and physical constraint calculations, the input vectors generate multi-dimensional performance prediction results for each initial candidate mix proportion at the output layer, such as key properties like strength, elastic modulus, durability, thermal conductivity, and density. The prediction results for all candidate mix proportions are summarized to form a performance prediction matrix, where rows correspond to different mix proportions and columns correspond to different performance indicators.

[0081] S42. Based on the multi-objective optimization function, high-fidelity Bayesian optimization is used to identify high potential and verify high-fidelity performance prediction matrix to obtain performance index prediction results.

[0082] Furthermore, the performance prediction matrix is ​​input into a high-fidelity Bayesian optimization module based on a multi-objective optimization function. This module first uses a probabilistic model to jointly evaluate the target achievement degree and uncertainty of each candidate combination ratio, calculating the potential score of each candidate combination ratio in the multi-objective optimization space. In the high-potential identification stage, the algorithm sorts the combination ratios according to their potential scores and selects the top-ranked ones with higher uncertainty as the high-potential sample set. Subsequently, it enters the high-fidelity verification stage, where additional high-precision experimental data or high-resolution numerical simulation data are introduced to retrain and re-predict the high-potential sample set, correcting the bias caused by low-fidelity prediction and significantly improving prediction accuracy. The results after high-fidelity verification form the performance index prediction results, which not only more realistically reflect the performance of each combination ratio but also provide a reliable basis for judging whether it satisfies the multi-objective optimization function.

[0083] In one embodiment, based on a multi-objective optimization function, high-fidelity Bayesian optimization is used to perform high-potential identification and high-fidelity verification on the performance prediction matrix to obtain performance index prediction results, including:

[0084] S51. Input each performance index in the performance prediction matrix into the multi-objective optimization function to calculate the degree of target achievement and obtain the performance achievement evaluation result.

[0085] Indicatively, the degree of achievement is calculated based on a multi-objective optimization function. Each performance index in the performance prediction matrix is ​​compared with the preset performance target value, and the deviation value of each performance index and its normalized distance in the target space are calculated. Specifically, let y be the performance index vector of the i-th candidate combination ratio in the performance prediction matrix. i =[y i1 ,y i2 ,…,y im ], where m is the number of performance metrics; for each performance metric y ij Calculate its value relative to the target value t j The degree of compliance d ij Exemplary Where ∈ is used to avoid division by zero; the pass rates of each indicator are weighted according to the weight vector w=[w1,w2,…,w m We perform a weighted summation to obtain the overall compliance rate of the i-th candidate combination ratio. The final performance compliance assessment result vector is formed as D = [D1, D2, ..., D...]. n ], where n is the number of candidate combination ratios, and this result is used to measure the degree of matching between each candidate combination ratio and the multi-objective performance.

[0086] S52. Based on the strategy of maximizing the improvement value, determine the high-potential solution with better multi-objective among multiple initial candidate mix proportions according to the performance evaluation results, and obtain multiple high-potential mix proportion solutions.

[0087] Specifically, the core idea of ​​the Expected Improvement (EI) strategy is to find candidate solutions that bring the greatest expected performance improvement based on known sample points, and then evaluate the performance achievement result D. i Combined with the prediction mean μ of the high-fidelity Bayesian optimization model i and the predicted standard deviation σ i Calculate the EI value. i =(μ i -D best )·Φ(Z i )+σ i ·φ(Z i ), where D best The best overall compliance rate known to date. Φ(·) is the cumulative distribution function (CDF) of the standard normal distribution, and φ(·) is the probability density function (PDF) of the standard normal distribution. For example, the larger the EI value, the higher the potential of the candidate mix design in improving the target performance; the mix design with the top k EI values ​​from multiple initial candidate mix designs is selected as the high-potential mix design solution that is better for multiple objectives.

[0088] S53. Perform high-fidelity simulation verification on each high-potential mix proportion solution to obtain the performance index prediction results of each initial candidate mix proportion.

[0089] Furthermore, the performance of high-potential mix design solutions is re-evaluated through more precise simulation. Specifically, for the selected high-potential mix design solutions, a high-fidelity numerical simulation tool, namely a high-fidelity verification network, is used to recalculate key indicators such as mechanical properties and durability. The high-fidelity verification network simulation considers more physical mechanisms and nonlinear factors, such as the chemical reaction kinetics between material components, the actual distribution of aggregate gradation, and changes in temperature and humidity conditions. The actual performance indicators obtained from the simulation are compared with the predicted values ​​in the original performance prediction matrix, and the performance indicator prediction results are updated. Optionally, if the high-fidelity results meet all preset performance targets, the mix design is included in the final candidate pool; otherwise, it is eliminated.

[0090] In one embodiment, robust disturbance planning is performed on each candidate mix proportion based on construction environment parameters to determine the candidate mix proportion that meets the requirements of performance robustness and target trade-off as the construction concrete mix proportion, including:

[0091] S61. Construct a disturbance space for each candidate mix proportion based on the construction environment parameters; the disturbance space includes the disturbance amplitude, the disturbance direction, and the performance-sensitive parameters corresponding to the disturbance direction.

[0092] This paper illustrates how key construction environment parameters are extracted based on real-time and historical environmental data from the construction site. Combining material mechanics properties and a concrete hardening kinetic model, the sensitivity of these parameters to various performance indicators is analyzed, and the corresponding set of disturbance-sensitive parameters is determined. Specifically, the disturbance amplitude is set based on the on-site fluctuation range and design margin of each sensitive parameter, determining the maximum and minimum amplitude of parameter disturbances. For each sensitive parameter, the possible effects of positive disturbances (performance enhancement) and negative disturbances (performance degradation) are considered. Sensitive parameter mapping establishes a mapping relationship between each sensitive parameter and its corresponding performance response coefficient, serving as a constraint condition for the disturbance space.

[0093] S62. Based on interval fuzzy rules, generate multiple perturbation samples corresponding to candidate matching ratios according to the perturbation space.

[0094] Specifically, interval fuzzy mathematics is used to fuzzify the representation of each sensitive parameter in the disturbance space within its interval. Optionally, a triangular or trapezoidal membership function is established for each sensitive parameter to express the uncertainty of its value. Within the fuzzy interval, multiple combinations of disturbance parameters are generated using Latin hypercube sampling (LHS) or Monte Carlo sampling methods. Substituting each combination of disturbance parameters into the original parameter set of the candidate mix proportions yields multiple disturbance mix proportion samples, which represent possible fluctuations under construction site conditions.

[0095] S63. Input each perturbation sample into the physical information neural network to predict the robustness performance and obtain the perturbation sample performance of the corresponding perturbation sample.

[0096] Similarly, using a pre-trained physical information neural network model, the generated perturbation mix ratio samples are used as input to predict their performance on various performance indicators.

[0097] S64. If the performance of the perturbation sample meets the requirements of the multi-objective optimization function within the preset tolerance range, the corresponding candidate mix proportion is determined as the construction concrete mix proportion.

[0098] Furthermore, the performance matrix of the perturbed samples is input into a multi-objective optimization function to calculate the target achievement status for each perturbed sample. If all perturbed samples fall within the preset tolerance range in terms of performance indicators, such as ±5% strength fluctuation and ±3% shrinkage rate fluctuation, and the comprehensive optimization objective function value does not exceed the threshold, then the candidate mix proportion is determined to have sufficient robustness under construction environment disturbances. The candidate mix proportion that meets the conditions will be finally confirmed as the construction concrete mix proportion for actual on-site pouring and construction.

[0099] In one embodiment, the multi-fidelity Bayesian joint optimization model is constructed using the following method:

[0100] S71. A training set is constructed by acquiring a large amount of historical construction mix proportion data and corresponding measured performance index data.

[0101] S72. A loss function is jointly constructed using the multi-objective performance prediction error and the physical constraint residual. The physical information neural network is trained end-to-end using the training set to obtain a physical information neural network with physical interpretability performance prediction capability. The physical information neural network contains mechanical constitutive equations and material aging constraint equations, and embeds mechanical equation constraint terms and engineering experience rules as physical loss terms.

[0102] Physical Information Neural Networks (PINNs) are used to map mix proportion parameters to predict physical or engineering performance, and enforce the physical laws and empirical constitutive relations of concrete materials during training and inference. The PINN network architecture typically consists of an input layer, several hidden layers, and multiple output layers. The input vectors include, but are not limited to, water-cement ratio, cementitious material dosage, sand ratio, admixture dosage, aggregate gradation characteristics, key properties of raw materials, and construction environment and curing conditions. The output vectors include target performance indicators such as age strength S(t), instantaneous slump, impermeability indicators, shrinkage rate, and electrical conductivity.

[0103] Optionally, the output of PINN is determined by a set of parameterized functions. Given θ, where x is the input vector and θ are the network parameters. The network minimizes the joint loss function consisting of data terms and physical terms. Conduct training The data error term is: Used to ensure consistency between network and experimental data; physical constraint terms Composed of the residuals of material dynamics and constitutive equations, exemplified by the hydration kinetics equation. Where α represents the degree of hydration, k(T) is an Arrhenius expression with temperature, and the hydration kinetic equation is related to the intensity evolution as S(t) ∝ [α(t)]. p The residual write loss, i.e. in This represents the residual operators derived from the physical equations, including hydration kinetic residuals, rheological constitutive residuals, Bingham model residuals, etc., and the regularization term. Control the norm of the weights to prevent overfitting. Weight λ data ,λ phys ,λ reg It can be adjusted according to task importance to balance data-driven and physical consistency.

[0104] Optionally, smooth activation functions are often used in hidden layers to facilitate gradient propagation of PDE residuals. The network depth and width are determined based on the input dimension and training samples. Training employs several rounds of pre-training with stochastic optimizers such as Adam to obtain better initial values, followed by fine-tuning convergence using an adaptive second-order optimizer. To obtain uncertainty estimates for use in the Bayesian acquisition function, PINN can provide prediction variance estimates through methods such as model ensembles.

[0105] S73. Obtain a large amount of experimental data on mixing ratios with different fidelity, and train a high-fidelity verification network based on the experimental data on mixing ratios.

[0106] To correct for biases in mid-fidelity PINNs and provide high-fidelity observations for MF-GP (Multi-Fidelity Gaussian Process), a high-fidelity validation network or a high-precision simulation pipeline needs to be built. The high-fidelity validation network can be a specially trained deep regression network, such as a ResNet (residual network) style network or a deep wide network, or it can be a digital twin module based on a detailed physical simulation, depending on the available samples and computational resources.

[0107] For example, a neural network is used. The network structure tends to capture complex nonlinear couplings, typically employing deep residual blocks, batch normalization, and stronger regularization techniques, along with more rigorous cross-validation. The training objective is to minimize the mean squared error of the high-fidelity data, while Bayesian neural network variational inference techniques can be used to directly output the prediction uncertainty. The loss used for training is the standard data fitting loss. Where g φ For high-fidelity network parameterization mapping, These are experimental or high-precision simulation observations.

[0108] S74. Based on the collaborative kernel function, the migration model between data of different fidelity is realized, and a multi-fidelity Gaussian process network is obtained.

[0109] Multi-fidelity Gaussian processes are used to model data of different fidelity levels—low fidelity data provided by PINN, medium fidelity data provided by simulation or a coarse model, and high fidelity data provided by simulation—in a unified manner, thereby improving prediction accuracy in cases of sparse samples. Multi-fidelity modeling employs an autoregressive collaborative model, the basic form of which is… Where f L For low-fidelity processes, f H For a high-fidelity process, ρ(x) is the scale coupling function, and δ(x) is a function independent of f. L The high-fidelity residual GP, its prior is The joint covariance is To achieve flexibility, the kernel function k uses a squared exponential (RBF) kernel. nuclear hyperparameter σ 2 and The coupling parameter ρ is obtained through maximum likelihood estimation (MLE) or maximization of the marginal likelihood logarithm.

[0110] Based on MF-GP, the joint posterior mean and variance of any candidate point can be obtained from the training data, which can be used for subsequent acquisition function calculation.

[0111] S75. With the goal of maximizing the expected improvement of candidate solutions, train a Bayesian optimizer to obtain a multi-fidelity Bayesian acquisition function with filtering capabilities.

[0112] In MF-BO (Multi-Fidelity Bayesian Optimization), the acquisition function determines the next sampling location and fidelity level, employing Expected Improvement (EI). For multi-fidelity objectives, Expected Hypervolume Improvement (EHVI) is used, or deterministic weighting is employed to transform multiple objectives into a single objective (scalarization). In multi-fidelity scenarios, the acquisition function also needs to consider the sampling cost c(s), maximizing the improvement value through cost normalization. In the multi-objective case, EHVI measures the hypervolume increment at a point under the current Pareto front; computation is complex but can be approximated numerically or estimated using Monte Carlo methods. For multi-fidelity EI, the EI value can be quickly estimated at low fidelity and multiplied by a fidelity confidence correction factor.

[0113] The sampling function is determined by tuning its parameters on the historical validation set and using Bayesian or cross-validation to identify the strategy that best improves the Pareto front within a given budget. The sampling function can be optimized by finding its maximum value in the design space using gradient or heuristic global optimizers to obtain the next sampling point.

[0114] S76. A multi-fidelity Bayesian joint optimization model is obtained by sequentially connecting a multi-fidelity Bayesian acquisition function, a physical information neural network, a high-fidelity verification network, and a multi-fidelity Gaussian process network.

[0115] The multi-fidelity Bayesian joint optimization model consists of PINN, a high-fidelity validation network, MF-GP, and a cost-aware acquisition function. A training set is constructed, and PINN is trained end-to-end. Then, the high-fidelity network is trained individually or in parallel. Next, MF-GP is fitted using all fidelity data. Finally, the acquisition function is constructed and its parameters are tuned based on the fitted MF-GP. During runtime, the acquisition function is maximized to generate new candidate points. These candidate points are first quickly evaluated by PINN, and if necessary, sent to high-fidelity validation according to the acquisition strategy. Validation data is fed back into MF-GP, and hyperparameters are re-estimated, forming a closed-loop iteration until the budget or convergence condition is met.

[0116] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0117] Based on the same inventive concept, this application also provides a system for dynamically controlling the mix proportion of intelligent concrete based on multi-objective optimization, used to implement the aforementioned method for dynamically controlling the mix proportion of intelligent concrete based on multi-objective optimization. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the system for dynamically controlling the mix proportion of intelligent concrete based on multi-objective optimization provided below can be found in the limitations of the method for dynamically controlling the mix proportion of intelligent concrete based on multi-objective optimization described above, and will not be repeated here.

[0118] In one exemplary embodiment, such as Figure 4 As shown, a smart concrete mix proportion dynamic control system based on multi-objective optimization is provided, including:

[0119] Data acquisition module 401 is used to acquire performance target parameters, construction material performance parameters, and construction environment parameters associated with the current construction task; performance target parameters include target strength level, workability index, and environmental constraints;

[0120] The optimization objective module 402 is used to construct a multi-objective optimization function based on the performance objective parameters; the multi-objective optimization function includes minimizing strength error, controlling workability deviation, minimizing carbon footprint, and minimizing material cost.

[0121] The optimization module 403 is used to input the performance target parameters and construction material performance parameters into a pre-trained multi-fidelity Bayes joint optimization model based on a multi-objective optimization function to obtain multiple candidate mix proportions.

[0122] The disturbance verification module 404 is used to perform robust disturbance planning on each candidate mix proportion based on construction environment parameters, and determine the candidate mix proportion that meets the requirements of performance robustness and target trade-off as the construction concrete mix proportion.

[0123] In one embodiment, the optimization module 403 is further configured to:

[0124] The design space is obtained based on the performance target parameters and the performance parameters of the construction materials; the design space includes multiple design variables with clear boundaries; the design variables include water-cement ratio, cement dosage, fly ash content, sand ratio, and water-reducing agent content;

[0125] With the goal of converging the multi-fidelity Bayesian acquisition function to satisfy the multi-objective optimization function, the multi-fidelity Bayesian acquisition function is used to explore the values ​​of design variables in the design space and obtain multiple candidate mix proportions.

[0126] In one embodiment, it includes:

[0127] The exploration module is used to explore the values ​​of design variables in the design space using a multi-fidelity Bayesian acquisition function to obtain initial candidate mix proportions.

[0128] The performance prediction module is used to predict the performance indicators of each initial candidate mix proportion and obtain the performance indicator prediction results.

[0129] The iterative optimization module is used to optimize the multi-fidelity Bayesian acquisition function based on the prediction results if the performance index prediction results do not meet the multi-objective optimization function, and to redetermine the initial candidate combination ratio using the optimized multi-fidelity Bayesian acquisition function.

[0130] The optimization module 403 is also used to determine the initial candidate combination ratio of the corresponding multi-fidelity Bayesian acquisition function as the candidate combination ratio if the performance index prediction result converges to the multi-objective optimization function.

[0131] In one embodiment, it includes:

[0132] The performance prediction module is also used to predict the performance indicators of each initial candidate mix ratio through a physical information neural network, and obtain the performance prediction matrix.

[0133] The high-fidelity verification module is used to identify and verify the high potential of the performance prediction matrix based on a multi-objective optimization function through high-fidelity Bayesian optimization, and obtain the performance index prediction results.

[0134] In one embodiment, it includes:

[0135] The evaluation module is used to input each performance index in the performance prediction matrix into a multi-objective optimization function to calculate the degree of target achievement and obtain the performance achievement evaluation result.

[0136] The high-potential module is used to determine the better high-potential solution for multiple objectives from multiple initial candidate mix proportions based on the strategy of maximizing the improvement value and the performance achievement evaluation results, thereby obtaining multiple high-potential mix proportion solutions.

[0137] The high-fidelity verification module is also used to perform high-fidelity simulation verification on each high-potential mix design solution to obtain the performance index prediction results of each initial candidate mix design.

[0138] In one embodiment, the disturbance verification module 404 is further configured to:

[0139] Based on the construction environment parameters, a disturbance space is constructed for each candidate mix proportion; the disturbance space includes the disturbance amplitude, the disturbance direction, and the performance-sensitive parameters corresponding to the disturbance direction;

[0140] Based on interval fuzzy rules, multiple perturbation samples corresponding to candidate combination ratios are generated according to the perturbation space;

[0141] Each perturbation sample is input into a physical information neural network to predict its robustness performance, thereby obtaining the perturbation sample performance of the corresponding perturbation sample.

[0142] If the performance of the perturbation samples meets the requirements of the multi-objective optimization function within the preset tolerance range, the corresponding candidate mix proportion is determined as the construction concrete mix proportion.

[0143] In one embodiment, a model training module is also included, for:

[0144] A training set was constructed by acquiring a large amount of historical construction mix proportion data and corresponding measured performance index data;

[0145] A loss function is jointly constructed by multi-objective performance prediction error and physical constraint residual. The physical information neural network is trained end-to-end using the training set to obtain a physical information neural network with physical interpretability performance prediction capability. The physical information neural network includes mechanical constitutive equations and material aging constraint equations, and embeds mechanical equation constraint terms and engineering empirical rules as physical loss terms.

[0146] Acquire a large amount of experimental data on mixing ratios with different fidelity, and train a high-fidelity validation network based on the experimental data on mixing ratios;

[0147] Based on the collaborative kernel function, transfer modeling between data of different fidelity is realized, and a multifidelity Gaussian process network is obtained.

[0148] With the goal of maximizing the expected improvement of candidate solutions, a Bayesian optimizer is trained to obtain a multi-fidelity Bayesian acquisition function with filtering capabilities;

[0149] A multi-fidelity Bayesian joint optimization model is obtained by sequentially connecting a multi-fidelity Bayesian acquisition function, a physical information neural network, a high-fidelity verification network, and a multi-fidelity Gaussian process network.

[0150] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps in the above method embodiments.

[0151] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0152] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0153] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for dynamic control of intelligent concrete mix proportions based on multi-objective optimization, characterized in that, The method includes: Obtain the performance target parameters, construction material performance parameters, and construction environment parameters associated with the current construction task; the performance target parameters include the target strength level, workability index, and environmental constraints. A multi-objective optimization function is constructed based on the performance target parameters; the multi-objective optimization function includes minimizing strength error, controlling workability deviation, minimizing carbon footprint, and minimizing material cost. Based on the multi-objective optimization function, the performance objective parameters and the construction material performance parameters are input into a pre-trained multi-fidelity Bayesian joint optimization model to obtain multiple candidate mix proportions; Based on the construction environment parameters, robust disturbance planning is performed on each of the candidate mix proportions to determine the candidate mix proportion that meets the requirements of performance robustness and target trade-off as the construction concrete mix proportion.

2. The method according to claim 1, characterized in that, Based on the multi-objective optimization function, the performance objective parameters and the construction material performance parameters are input into a pre-trained multi-fidelity Bayesian joint optimization model to obtain multiple candidate mix proportions, including: The design space is obtained based on the performance target parameters and the construction material performance parameters; the design space includes multiple design variables with clear boundaries; the design variables include water-cement ratio, cement dosage, fly ash content, sand ratio, and water-reducing agent content; With the goal of converging the multi-fidelity Bayesian acquisition function to satisfy the multi-objective optimization function, the multi-fidelity Bayesian acquisition function is used to explore the values ​​of design variables in the design space to obtain multiple candidate mix proportions.

3. The method according to claim 2, characterized in that, The objective is to converge the multi-fidelity Bayesian acquisition function to satisfy the multi-objective optimization function. The multi-fidelity Bayesian acquisition function is used to explore the values ​​of design variables in the design space to obtain multiple candidate mix proportions, including: The multi-fidelity Bayesian acquisition function is used to explore the values ​​of design variables in the design space to obtain an initial candidate mix proportion; the initial candidate mix proportion includes material properties and material proportions. The performance index prediction results are obtained by predicting the performance index of each of the initial candidate mix proportions. If the performance index prediction result does not meet the multi-objective optimization function, then the multi-fidelity Bayesian acquisition function is optimized according to the prediction result, and the initial candidate combination ratio is re-determined using the optimized multi-fidelity Bayesian acquisition function; If the performance index prediction result converges to the multi-objective optimization function, the initial candidate combination ratio of the corresponding multi-fidelity Bayesian acquisition function is determined as the candidate combination ratio.

4. The method according to claim 3, characterized in that, The process of predicting performance indicators for each of the initial candidate mix proportions to obtain performance indicator prediction results includes: A physical information neural network is used to predict the performance indicators of each of the initial candidate mix proportions to obtain a performance prediction matrix; the performance prediction matrix includes the performance indicators corresponding to each of the initial candidate mix proportions. Based on the multi-objective optimization function, the performance prediction matrix is ​​subjected to high-potential identification and high-fidelity verification through high-fidelity Bayesian optimization to obtain the performance index prediction results.

5. The method according to claim 4, characterized in that, The process of performing high-potential identification and high-fidelity verification of the performance prediction matrix based on the multi-objective optimization function, to obtain the performance index prediction results, includes: Each performance index in the performance prediction matrix is ​​input into a multi-objective optimization function to calculate the degree of target achievement, and the performance achievement evaluation result is obtained. Based on the strategy of maximizing the improvement value, a high-potential solution with better multi-objectives is determined from multiple initial candidate mix ratios according to the performance achievement evaluation results, thereby obtaining multiple high-potential mix ratio solutions. High-fidelity simulations were performed on each of the high-potential mix proportions to obtain the predicted performance indicators of each of the initial candidate mix proportions.

6. The method according to claim 4, characterized in that, The step of performing robust disturbance planning on each of the candidate mix proportions based on the construction environment parameters, and determining the candidate mix proportion that meets the requirements of performance robustness and target trade-off as the construction concrete mix proportion, includes: Based on the construction environment parameters, a disturbance space is constructed for each of the candidate mix proportions; the disturbance space includes the disturbance amplitude, the disturbance direction, and the performance sensitive parameters corresponding to the disturbance direction; Based on the interval fuzzy rule, multiple perturbation samples corresponding to the candidate combination ratio are generated according to the perturbation space; Each of the perturbation samples is input into the physical information neural network for robust performance prediction, thereby obtaining the perturbation sample performance corresponding to the perturbation sample; If the performance of the perturbation sample meets the requirements of the multi-objective optimization function within the preset tolerance range, the corresponding candidate mix proportion is determined as the construction concrete mix proportion.

7. The method according to claim 1, characterized in that, The multi-fidelity Bayesian joint optimization model is constructed using the following method: A training set was constructed by acquiring a large amount of historical construction mix proportion data and corresponding measured performance index data; A loss function is jointly constructed using multi-objective performance prediction error and physical constraint residual. The physical information neural network is then trained end-to-end using the training set to obtain a physical information neural network with physically interpretable performance prediction capabilities. The physical information neural network includes mechanical constitutive equations and material aging constraint equations, and embeds mechanical equation constraint terms and engineering empirical rules as physical loss terms. A large amount of experimental data on mixing ratios with different fidelity was obtained, and a high-fidelity verification network was trained based on the experimental data on the mixing ratios. Based on the collaborative kernel function, transfer modeling between data of different fidelity is realized, and a multifidelity Gaussian process network is obtained. With the goal of maximizing the expected improvement of candidate solutions, a Bayesian optimizer is trained to obtain a multi-fidelity Bayesian acquisition function with filtering capabilities; The multi-fidelity Bayesian joint optimization model is obtained by sequentially connecting the multi-fidelity Bayesian acquisition function, the physical information neural network, the high-fidelity verification network, and the multi-fidelity Gaussian process network.

8. A smart concrete mix proportion dynamic control system based on multi-objective optimization, characterized in that, The system includes: The data acquisition module is used to acquire the performance target parameters, construction material performance parameters, and construction environment parameters associated with the current construction task; the performance target parameters include target strength level, workability index, and environmental constraints. An optimization objective module is used to construct a multi-objective optimization function based on the performance objective parameters; the multi-objective optimization function includes minimizing strength error, controlling workability deviation, minimizing carbon footprint, and minimizing material cost. The optimization module is used to input the performance target parameters and the construction material performance parameters into a pre-trained multi-fidelity Bayesian joint optimization model based on the multi-objective optimization function to obtain multiple candidate mix proportions; The disturbance verification module is used to perform robust disturbance planning on each of the candidate mix proportions based on the construction environment parameters, and determine the candidate mix proportion that meets the requirements of performance robustness and target trade-off as the construction concrete mix proportion.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images