Material ratio intelligent optimization method and device fusing domain knowledge and machine learning

Abstract

The invention is applicable to the field of building materials, and provides an intelligent material ratio optimization method and device fusing domain knowledge and machine learning, multiple groups of training data sets of a fiber cement-based material are acquired, and each group of training data set at least comprises training fiber data; determining a target first regression coefficient of the fiber cement-based material based on the multiple groups of training data sets and a compressive strength formula of the cement-based material; constructing a first target prediction model of the fiber cement-based material according to the multiple groups of training data sets and the target first regression coefficient; and constructing a second target prediction model of the fiber cement-based material and calculating the mix proportion of the doped material in the fiber cement-based material. According to the method, fiber data are fused into a compressive strength formula to determine two regression coefficients suitable for the fiber cement-based material, and then a prediction model for predicting the compressive strength of the fiber cement-based material is constructed by adopting the two determined regression coefficients, so that the influence of fiber doping is considered when the material mix proportion is calculated; and the mix proportion design precision is improved.

Description

Technical Field

[0001] The present application belongs to the field of building materials technology, and in particular relates to a method and device for intelligent optimization of material ratios that integrates domain knowledge and machine learning. Background Art

[0002] Fiber cement-based materials, as a new type of building material, are a general term for composite materials composed of fibers and cement-based materials. Fiber cement-based materials can be considered as materials composed of cement, mortar, etc. as the base material and fibers as the reinforcing material. They can significantly improve the ductility, crack resistance and toughness of traditional cement-based materials, and have important application value in engineering scenarios such as earthquake resistance and impact resistance.

[0003] However, the performance of fiber cement-based materials is affected by the combined influence of matrix components (such as cement, fly ash, mineral powder, etc.), fiber properties (such as type, length, volume fraction, etc.), and mix parameters (such as water-binder ratio, sand-binder ratio, etc.), showing a complex nonlinear relationship.

[0004] The regression coefficients in traditional methods for determining the material mix ratio of fiber cement-based materials are mostly empirical values ​​or fixed values. For example, the material mix ratio of fiber cement-based materials is determined directly using the regression coefficients determined by experimental data of traditional cement-based materials. The influence of fiber incorporation on the material mix ratio is not considered, resulting in insufficient accuracy in mix ratio design and difficulty in achieving precise control of the performance of fiber cement-based materials. Summary of the Invention

[0005] The embodiments of the present application provide a method and device for intelligent optimization of material proportions that integrate domain knowledge and machine learning. By considering the impact of fiber incorporation on the material proportion, the accuracy of the proportion design is improved, and precise regulation of the performance of fiber cement-based materials is achieved.

[0006] In a first aspect, the embodiments of the present application provide a material ratio intelligent optimization method that integrates domain knowledge and machine learning, including: Acquire multiple sets of training data sets of fiber cement-based materials, wherein each set of training data sets includes at least: training cement ratio, training sand-binder ratio and training fiber data of the fiber cement-based materials; Determining a target first regression coefficient of the fiber cement-based material based on multiple training data sets and a compressive strength formula of the cement-based material, wherein the compressive strength formula is determined based on at least the first regression coefficient and the second regression coefficient of the cement-based material, and the target first regression coefficient is used to calculate the actual compressive strength of the fiber cement-based material; A first target prediction model for the fiber cement-based material is constructed based on multiple sets of training data sets and target first regression coefficients. The first target prediction model is used to predict the actual second regression coefficient of the fiber cement-based material. The actual second regression coefficient is used to calculate the actual compressive strength of the fiber cement-based material. Constructing a second target prediction model for the fiber cement-based material based on the multiple training data sets and the first target prediction model, wherein the second target prediction model is used to predict the actual compressive strength of the fiber cement-based material; Based on the objective function and constraints of the fiber cement-based material, the mix ratio of the added materials in the fiber cement-based material is calculated, wherein at least part of the objective function is determined based on the actual compressive strength and target compressive strength output by the second objective prediction model.

[0007] In a second aspect, the embodiments of the present application provide a material ratio intelligent optimization device that integrates domain knowledge and machine learning, including: An acquisition module is used to acquire multiple sets of training data sets of fiber cement-based materials, wherein each set of training data sets at least includes: training cement ratio, training sand-binder ratio and training fiber data of the fiber cement-based materials; a regression coefficient determination module, configured to determine a target first regression coefficient of the fiber cement-based material based on multiple sets of training data sets and a compressive strength formula of the cement-based material, wherein the compressive strength formula is determined based on at least the first regression coefficient and the second regression coefficient of the cement-based material, and the target first regression coefficient is used to calculate the actual compressive strength of the fiber cement-based material; A first model building module is used to build a first target prediction model for the fiber cement-based material based on multiple sets of training data sets and the target first regression coefficient, the first target prediction model is used to predict the actual second regression coefficient of the fiber cement-based material, and the actual second regression coefficient is used to calculate the actual compressive strength of the fiber cement-based material; A second model building module is used to build a second target prediction model for the fiber cement-based material based on the multiple sets of training data sets and the first target prediction model, wherein the second target prediction model is used to predict the actual compressive strength of the fiber cement-based material; A calculation module is used to calculate the mix ratio of the added materials in the fiber cement-based material based on the objective function and constraints of the fiber cement-based material, wherein at least part of the objective function is determined based on the actual compressive strength and target compressive strength output by the second objective prediction model.

[0008] In a third aspect, an embodiment of the present application provides a terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the methods of the first aspect when executing the computer program.

[0009] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method of any one of the first aspects.

[0010] In a fifth aspect, an embodiment of the present application provides a computer program product, which, when executed on a terminal device, enables the terminal device to execute any one of the methods in the first aspect above.

[0011] The embodiment of the present application provides a material ratio intelligent optimization method and device that integrates domain knowledge and machine learning. The method includes: obtaining multiple sets of training data sets of fiber cement-based materials, wherein each set of training data sets includes at least: training cement ratio, training sand-glue ratio and training fiber data of the fiber cement-based materials; determining a target first regression coefficient of the fiber cement-based material based on the multiple sets of training data sets and the compressive strength formula of the cement-based material, the compressive strength formula is determined based on at least the first regression coefficient and the second regression coefficient of the cement-based material, and the target first regression coefficient is used to calculate the actual compressive strength of the fiber cement-based material; according to the multiple sets of training data sets and the target first regression coefficient A first target prediction model for fiber cement-based materials is constructed based on multiple training data sets and the first target prediction model, and the first target prediction model is used to predict the actual second regression coefficient of the fiber cement-based materials, and the actual second regression coefficient is used to calculate the actual compressive strength of the fiber cement-based materials; a second target prediction model for fiber cement-based materials is constructed based on multiple training data sets and the first target prediction model, and the second target prediction model is used to predict the actual compressive strength of the fiber cement-based materials; based on the objective function and constraints of the fiber cement-based materials, the mix ratio of the added materials in the fiber cement-based materials is calculated, wherein at least part of the objective function is determined based on the actual compressive strength and target compressive strength output by the second target prediction model. Utilizing the above technical solution, two regression coefficients applicable to fiber cement-based materials are determined by integrating fiber data into the compressive strength formula, and then a prediction model for predicting the compressive strength of fiber cement-based materials is constructed by using the two determined regression coefficients. This allows the influence of fiber incorporation to be taken into account when calculating the material mix ratio, improves the design accuracy of the mix ratio, and thus achieves precise control of the performance of fiber cement-based materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0013] Figure 1 This is a flow chart of a material ratio intelligent optimization method that integrates domain knowledge and machine learning, provided in one embodiment of the present application; Figure 2 This is a flow chart of a material ratio intelligent optimization method that integrates domain knowledge and machine learning, provided in another embodiment of the present application; Figure 3 is a second regression coefficient provided in an embodiment of the present application Schematic diagram of the comparison between the true value and the predicted value; Figure 4 This is an embodiment of the present application. Schematic diagram of the comparison between the true value and the predicted value; Figure 5 This is a structural block diagram of a material ratio intelligent optimization device that integrates domain knowledge and machine learning, provided in one embodiment of the present application; Figure 6 This is a structural diagram of a terminal device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0014] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0015] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0016] It will also be understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0017] As used in this specification and the appended claims, the term "if" can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0018] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0019] References to "one embodiment" or "some embodiments" in this specification mean that a particular feature, structure, or characteristic described in conjunction with that embodiment is included in one or more embodiments of the present application. Thus, phrases such as "in one embodiment," "in some embodiments," "in other embodiments," and "in other embodiments" appearing in various places in this specification do not necessarily refer to the same embodiment, but rather mean "one or more but not all embodiments," unless otherwise specifically emphasized. The terms "including," "comprising," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0020] It can be considered that the traditional method of determining the material mix ratio of fiber cement-based materials has the following limitations: 1. Limitations of domain knowledge application: In the compressive strength formula of traditional cement-based materials, the regression coefficient fails to fully consider the effect of fiber incorporation on the microstructure and macroscopic mechanical properties of the material, making it difficult for theoretical models to accurately predict the performance of fiber cement-based materials. In addition, the core regression coefficient in the existing formula is usually a fixed value and cannot be dynamically adjusted according to key parameters such as fiber type, dosage, and mix ratio, resulting in a significant deviation between theoretical calculation results and actual performance. Among them, cement-based materials can be understood as a type of composite material formed by hardening after mixing cement with water, aggregates (sand, stone, etc.) and admixtures. Cement-based materials are one of the most widely used materials in modern construction projects, mainly including concrete, mortar, cement paste, etc.

[0021] 2. Deficiencies of Purely Data-Driven Models: In recent years, machine learning algorithms (such as neural networks and random forests) have been used to predict the properties of fiber-cement-based materials. However, these machine learning algorithms rely on large amounts of labeled data and are prone to overfitting or loss of physical meaning. For example, purely data-driven models can output predictions that violate basic material mechanics (e.g., strength increases indefinitely with increasing water-cement ratio). Furthermore, purely data-driven models lack interpretability, making them difficult to guide mix optimization.

[0022] 3. Complexity of multi-objective optimization: The design of fiber-cement-based materials must simultaneously meet the comprehensive requirements of strength (28-day compressive strength), economy (material cost), and low carbon footprint. Traditional methods for determining the material mix ratio of fiber-cement-based materials, mainly through trial-and-error or single-objective optimization, struggle to balance the trade-offs between multiple objectives. For example, simply pursuing high strength will lead to increased cement usage, which not only increases costs but also carbon emissions, and is inconsistent with the development trend of green buildings.

[0023] 4. Lack of dynamic parameter correction: In the traditional method of determining the material mix ratio of fiber cement-based materials, the key parameters that affect the compressive strength (such as the core regression coefficient) are mostly empirical values ​​or fixed values. The dynamic influence of fiber properties (such as diameter, aspect ratio, volume fraction) and the proportion of matrix components on the compressive strength are not considered. As a result, the design accuracy of the material mix ratio is insufficient, making it difficult to achieve precise control of material properties.

[0024] Based on this, the embodiment of the present application provides an intelligent optimization method for material ratios that integrates domain knowledge and machine learning. Through the dual-driven design that integrates domain knowledge and machine learning, it can use the physical law constraint model to predict, dynamically correct key parameters, and optimize multi-objective trade-off relationships to achieve efficient, accurate, and green design of fiber cement-based materials.

[0025] Figure 1 This is a flow chart of a material ratio intelligent optimization method that integrates domain knowledge and machine learning, provided in one embodiment of the present application. As an example and not a limitation, this method can be applied to terminal devices, such as Figure 1 As shown, the method includes: S101. Obtain multiple training data sets of fiber cement-based materials.

[0026] Each training dataset includes at least training cement ratios, training sand-binder ratios, and training fiber data for fiber-cement-based materials. Other types of training data can also be included in addition to the aforementioned training data to enrich the content of the training dataset. For example, each training dataset may also include training water-binder ratios, training cement strength, and training compressive strength.

[0027] S102. Determine a target first regression coefficient of the fiber cement-based material based on multiple training data sets and a compressive strength formula of the cement-based material.

[0028] The compressive strength formula is determined based on at least a first regression coefficient and a second regression coefficient of the cement-based material, and the target first regression coefficient is used to calculate the actual compressive strength of the fiber cement-based material.

[0029] It should be noted that the first regression coefficient and the second regression coefficient in the existing compressive strength formula of cement-based materials are obtained based on the experimental data of cement-based materials. Therefore, the first regression coefficient and the second regression coefficient are mainly suitable for calculating the compressive strength of cement-based materials, and are not suitable for calculating the compressive strength of fiber cement-based materials. Therefore, this step can first perform statistical analysis and parameter fitting on multiple sets of training data sets of fiber cement-based materials to fit a set of regression coefficients suitable for fiber cement-based materials, and use the fitted regression coefficients to correct the compressive strength formula of cement-based materials, so that the corrected compressive strength formula can adapt to the characteristics of fiber cement-based materials, providing a data basis for subsequent improvement of model prediction effects.

[0030] The fitted regression coefficient may, for example, include a target first regression coefficient, i.e., a first regression coefficient applicable to fiber cement-based materials. The specific process for determining the target first regression coefficient is not limited herein. For example, a pre-configured neural network model may be used to directly input multiple training data sets and the compressive strength formula of cement-based materials into the neural network model, so that the neural network model outputs the corresponding target first regression coefficient. Alternatively, the target first regression coefficient may be obtained by setting initial values ​​for the first and second regression coefficients and then continuously modifying and iterating the initial values ​​of the first and second regression coefficients using an iterative optimization algorithm.

[0031] As a feasible implementation method, based on multiple sets of training data sets and the compressive strength formula of cement-based materials, the target first regression coefficient of the fiber cement-based material is determined, including: determining a first initial value as a current first regression coefficient of the fiber cement-based material, and determining a second initial value as a current second regression coefficient of the fiber cement-based material; For each training data set, the training water-cement ratio, training cement strength, current first regression coefficient, and current second regression coefficient of each training data set are input into the compressive strength formula to obtain the theoretical compressive strength of each training data set; Based on the theoretical compressive strength and training compressive strength of each training data set, calculate the current residual sum of squares of multiple training data sets; According to the current residual sum of squares, the current first regression coefficient and the current second regression coefficient are corrected, and the step of inputting the training water-cement ratio, training cement strength, current first regression coefficient, and current second regression coefficient of each training data set into the compressive strength formula to obtain the theoretical compressive strength of each training data set is returned to execution for each training data set, until the current residual sum of squares reaches convergence or the number of corrections to the current first regression coefficient reaches the maximum number of iterations; The current first regression coefficient corresponding to when the current residual sum of squares reaches convergence or the number of corrections to the current first regression coefficient reaches the maximum number of iterations is determined as the target first regression coefficient of the fiber cement-based material.

[0032] The first and second initial values ​​can be understood as the starting points of an iterative optimization algorithm. They help the algorithm converge to the global optimal solution more quickly and avoid fitting failures or biased results caused by inappropriate initial values. For example, if the initial value differs significantly from the true value, the iteration may not converge (an error message) or produce meaningless parameters. The specific values ​​of the first and second initial values ​​can be determined based on empirical data.

[0033] Optionally, the iterative optimization algorithm may be a Levenberg-Marquardt (LM) algorithm, and the first initial value and the second initial value may be 2 and 1, respectively. The current first regression coefficient and the current second regression coefficient are the first regression coefficient and the second regression coefficient corresponding to each current iteration cycle.

[0034] Furthermore, the compressive strength formula can be ,in, Represents compressive strength, Represents the strength of cement, W / B represents the water-to-binder ratio, Represent the first regression coefficient and the second regression coefficient respectively. It can be seen that the compressive strength formula is a nonlinear model because the parameters The existence of a multiplicative relationship means that the relationship between the parameter and the dependent variable cannot be expressed by a simple linear combination.

[0035] In a specific embodiment, the above compressive strength formula is first determined as a model function fcu_model, and it can be seen that It is the unknown parameter that needs to be solved and the target to be fitted.

[0036] The training water-cement ratio and cement strength data sets from multiple training datasets can then be combined into a two-dimensional array with a shape of (2, n), using np.vstack. This conforms to the curve_fit function's requirement for multiple independent variable inputs, where n is the number of training datasets. The curve_fit function can be used for nonlinear least squares fitting.

[0037] Then, the curve_fit function can find the optimal parameters by minimizing the "sum of squares of the residuals between the true value and the model predicted value". The specific steps may include: in the first iteration round, the initial values ​​[2, 1] of the two-dimensional array, the first initial value and the second initial value can be substituted into the model fcu_model to calculate the predicted value (i.e., the theoretical compressive strength) corresponding to each set of training data sets; then calculate The residual between the predicted value and the true value (i.e., the training compressive strength) is calculated, and the current residual sum of squares of multiple training data sets is solved; then the optimization algorithm (such as the least squares method) is called inside the curve_fit function to correct the current residual square The value of ; then in subsequent iterations, the two-dimensional array, the corrected Substitute it into the model fcu_model, output the predicted value (theoretical compressive strength) corresponding to each set of training data sets, and continue to calculate the current residual sum of squares of multiple sets of training data sets until the current residual sum of squares reaches the minimum or the current number of iterations meets the conditions, then the iteration ends. The final parameter value is the optimal solution that makes the model prediction value closest to the true value. Among them, the final parameter value is an array containing the optimal parameters obtained by fitting , the code can be passed Extract these two optimal parameters, such as the optimal solution ,in, This is the target first regression coefficient of fiber cement-based materials.

[0038] Furthermore, this embodiment can also use the sklearn.metrics.r2_score function to compare the true value (such as ) and the model prediction value (such as ) comparison, we get Value, where Value = 1-(residual sum of squares / total sum of squares of deviations), that is, the proportion of variation that the model can explain after deducting the error, The value range is [0,1]. The closer to 1, the stronger the model's ability to explain the data, and the closer the predicted value is to the true value. In practical applications, The value is retained to 4 decimal places and printed to visually show the quality of the model fit, such as The goodness of fit output can be 0.9039, indicating that the model can explain 90.39% of the data variation and the fitting effect is good.

[0039] S103: Constructing a first target prediction model for fiber cement-based materials based on multiple sets of training data sets and target first regression coefficients.

[0040] The first target prediction model is used to predict the actual second regression coefficient of the fiber cement-based material, and the actual second regression coefficient is used to calculate the actual compressive strength of the fiber cement-based material.

[0041] Although the above steps have produced a set of The optimal solution is obtained, but the specific values ​​need to be further modified. For example, this step can consider the proportion of cement in the cementitious material, sand-binder ratio, W / B, fiber data (such as type, diameter, aspect ratio, volume fraction) The dynamic influence of To dynamically correct , so for each group of different cement-based material mix ratios, based on comprehensive consideration of fiber-related factors and other factors, are different.

[0042] Specifically, this step can comprehensively consider the training water-cement ratio, training cement strength, training cement ratio, training sand-cement ratio and training fiber data of each training data set, as well as the target first regression coefficient obtained in the previous step, to construct a first target prediction model for the fiber cement-based material. There is no limit to the method for constructing the first target prediction model. For example, the first preliminary prediction model can be selected according to actual needs, and the first preliminary prediction model can be trained by using multiple sets of training data sets and the target first regression coefficient to obtain the final first target prediction model. Alternatively, this step can also adopt other methods to construct the first target prediction model, as long as the first target prediction model can be obtained in the end.

[0043] S104: Construct a second target prediction model for fiber cement-based materials based on the multiple training data sets and the first target prediction model.

[0044] The second objective prediction model is used to predict the actual compressive strength of fiber cement-based materials.

[0045] Similar to step S103, this embodiment does not limit the method of constructing the second target prediction model. For example, a second preliminary prediction model can be selected according to actual needs, and the second preliminary prediction model can be trained by using multiple sets of training data sets and the first target prediction model to obtain the final second target prediction model. Alternatively, this step can also adopt other methods to construct the second target prediction model, as long as the second target prediction model can be obtained in the end.

[0046] S105. Based on the objective function and constraints of the fiber cement-based material, calculate the mix ratio of the materials added to the fiber cement-based material.

[0047] At least part of the objective function is determined based on the actual compressive strength and the target compressive strength output by the second objective prediction model.

[0048] This step can calculate the mix ratio of the added materials in the fiber cement-based material by clarifying the objective function and constraints of the fiber cement-based material. For example, different objective functions can be configured according to different design requirements of the fiber cement-based material. At the same time, the number of configured objective functions is not limited. For example, a single objective function can be configured, such as the objective function can be aimed at minimizing the difference between the actual compressive strength output by the second objective prediction model and the target compressive strength; multiple objective functions can also be configured according to design requirements, and the mix ratio of the added materials in the fiber cement-based material can be comprehensively calculated by balancing the trade-off relationship between multiple objective functions.

[0049] As a feasible implementation, the objective function includes a first objective function for minimizing the compressive strength error, a second objective function for minimizing carbon emissions, and a third objective function for minimizing material costs, wherein the first objective function is determined based on the actual compressive strength and the target compressive strength predicted by the second objective prediction model; Based on the objective function and constraints of the fiber cement-based material, the mix ratio of the material added to the fiber cement-based material is calculated, including: respectively determining the first function weight of the first objective function, the second function weight of the second objective function, and the third function weight of the third objective function; according to the first objective function, the second objective function, the third objective function, the first function weight, the second function weight, the third function weight, and the constraints, a non-dominated sorting genetic algorithm is used to calculate the mix ratio of the material added to the fiber cement-based material, where the mix ratio includes at least target cement strength, target component ratio, target water-reducing agent ratio, target water-cement ratio, target sand-cement ratio, and target fiber data.

[0050] Specifically, the first objective function is ,in, The actual compressive strength predicted by the model can be used for the second objective, The target compressive strength can be configured according to engineering requirements.

[0051] The goal of the second objective function can be to minimize carbon emissions. For example, the second objective function can be obtained by weighted accumulation based on the unit carbon emission coefficient of each component material (such as cement, mineral powder, fly ash, etc.) according to the feed amount to measure the impact of the fiber cement-based material ratio on the environment. For example, the second objective function can be specifically .

[0052] The third objective function can be used to calculate the total cost according to the amount of each material used to reflect the economic goal, that is, .

[0053] Among them, density Mi It can be the unit volume usage of each raw material / fiber in the fiber cement-based material, representing the mass of the i-th component in each cubic meter of fiber cement-based material, It can represent the mass of carbon dioxide emitted by the i-th component in each kilogram of fiber cement-based material (i.e., unit carbon emission coefficient), unit price It can be the unit price of the i-th component in the fiber cement-based material, which can be further considered as the unit cost of each raw material / fiber based on the market price or project price. Fibers can be a general term for fibers in the fiber cement-based material, such as basalt fiber, PE fiber, etc. For example, as shown in Table 1, the characteristic parameters of the component materials in the fiber cement-based material are listed, including the density, and unit price P.

[0054] Table 1 Characteristic parameters of the components of fiber cement-based materials

[0055] Furthermore, the constraints can be configured according to actual conditions, such as the ratio constraints and range constraints of the various components of the fiber cement-based material, wherein the ratio constraints can be , used to indicate that the sum of the mass proportions of each cementitious material is close to 1 (total amount of unit cementitious material), with a small error allowed , such as constant .

[0056] Table 2 shows the range constraints of the various components of the fiber cement-based material. OPC / B, FA / B, GGBFS / B, and SF / B represent the mass ratios of ordinary Portland cement, fly ash, slag powder, and silica fume in the cementitious material, respectively, which help reflect the active component structure of the cementitious material; SP / B represents the mass ratio of the water reducer to the cementitious material; S / B represents the mass ratio of the sand to the cementitious material; W / B represents the water-binder ratio; Vf represents the fiber volume fraction; df and Lf represent the fiber diameter and length, respectively.

[0057] Table 2 Range constraints

[0058] In specific applications, corresponding weights can be assigned to different objective functions based on actual project requirements. For example, in the construction of the core tube of a super-high-rise building, structural safety is the primary criterion, and the compressive strength must reach C80 or above. In this case, the engineer can increase the compressive strength weight to 0.6, and set the carbon footprint and cost weights to 0.2 and 0.2, respectively. The algorithm will then prioritize ensuring that the strength predicted by the second objective prediction model meets the standard, and on this basis, minimize carbon emissions and costs. The final solution can be based on high-grade cement, supplemented by a small amount of high-performance mineral admixtures, to ensure that the strength meets the super-high-rise load requirements. At the same time, the carbon footprint increase can be controlled through the rational combination of mineral admixtures to avoid excessive cost increases.

[0059] For example, for municipal road base projects, strength requirements are relatively relaxed (e.g., C30 is sufficient). However, due to the large scale of the project, the sensitivity of total carbon emissions, and the need to strictly control project costs, engineers can set the carbon footprint weight to 0.5, the cost weight to 0.4, and the compressive strength weight to just 0.1. In this case, the algorithm will focus on selecting mix solutions with low cement content (e.g., by incorporating large amounts of fly ash, slag, and other industrial waste). By sacrificing a small amount of non-critical strength margin, cement content (cement is the main contributor to the carbon footprint) can be significantly reduced. At the same time, low-cost admixtures are used to control material costs. The resulting solution can achieve "medium strength, low carbon emissions, and low cost" to meet the economic and environmental requirements of municipal projects.

[0060] For example, in green building demonstration projects, carbon footprint indicators are included in the mandatory assessment system. However, if the project is located in a high-seismic zone, the structural strength must still meet a high standard (such as C40). Engineers can set the compressive strength weighting to 0.4, the carbon footprint weighting to 0.4, and the cost weighting to 0.2. In this case, the algorithm will prioritize cementitious material combinations with lower carbon emission coefficients (such as a low-carbon cement + metakaolin composite system) while ensuring that the strength does not fall below the design value. At the same time, it will balance the price premium of low-carbon materials in the material procurement process. The resulting solution not only meets the carbon emission reduction requirements of green buildings but also ensures structural safety. Although the cost increases slightly, it is kept within the acceptable green technology increment range for the project.

[0061] Specifically, this embodiment can use a non-dominated sorting genetic algorithm to first randomly generate an initial parent population according to the design parameter range; then use the second target prediction model to predict the compressive strength , carbon footprint calculation formula and material cost calculation formula as the objective function, and perform non-dominated sorting and crowding calculation on the population in combination with the set constraints; generate offspring populations through selection, crossover and mutation operations, and repeat iterations until the maximum evolutionary generation is reached, and finally obtain multiple groups of Pareto optimal solutions to reflect the trade-off relationship between various objectives. For example, the Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) can be used to continuously adjust the compressive strength weight, carbon footprint and cost weight as needed, and select the final solution from multiple groups of Pareto optimal solutions.

[0062] The final solution is the mix ratio of materials added to fiber cement-based materials, including cement strength , the mass ratios of ordinary Portland cement, fly ash, slag powder and silica fume in cementitious materials OPC / B, FA / B, GGBFS / B, SF / B (i.e., target component ratios), the mass ratio of sand to cementitious materials S / B (i.e., target sand-binder ratio), the mass ratio of water-reducing agent to cementitious materials SP / B (i.e., target water-reducing agent proportion), the water-binder ratio W / B (i.e., target water-binder ratio), and the detailed properties of the fiber (i.e., target fiber data). The detailed properties of the fiber include, for example, fiber type, diameter df, length Lf, aspect ratio, density, tensile strength, elastic modulus, percentage and other data.

[0063] This embodiment provides an intelligent material ratio optimization method that integrates domain knowledge and machine learning. By integrating fiber data into the compressive strength formula, two regression coefficients applicable to fiber cement-based materials are determined. Then, a prediction model for predicting the compressive strength of fiber cement-based materials is constructed by using the two determined regression coefficients. This allows the influence of fiber incorporation to be taken into account when calculating the material ratio, improves the design accuracy of the ratio, and thus achieves precise regulation of the performance of fiber cement-based materials.

[0064] Figure 2 This is a flow chart of a material ratio intelligent optimization method that integrates domain knowledge and machine learning, provided by another embodiment of the present application. Each training data set in this embodiment also includes: training water-cement ratio, training cement strength, and training compressive strength. The compressive strength formula is ,in, Represents compressive strength, Represents the strength of cement, W / B represents the water-to-binder ratio, Represent the first regression coefficient and the second regression coefficient respectively; and according to multiple sets of training data sets and the target first regression coefficient, the first target prediction model of fiber cement-based materials is further optimized as follows: for each set of training data sets, the training water-cement ratio, training cement strength, training compressive strength and target first regression coefficient of each training data set are input into the compressive strength formula to obtain the theoretical second regression coefficient corresponding to each training data set; the first preliminary prediction model of fiber cement-based materials is established, and the first preliminary prediction model is a random forest regressor; according to the training water-cement ratio, training cement strength, training cement ratio, training sand-cement ratio and training fiber data of multiple sets of training data sets, and the theoretical second regression coefficient corresponding to each training data set, the first preliminary prediction model is trained to obtain the first target prediction model of fiber cement-based materials. Figure 2 As shown, the method includes: S201. Obtain multiple sets of training data sets of fiber cement-based materials.

[0065] S202. Determine a target first regression coefficient of the fiber cement-based material based on multiple training data sets and a compressive strength formula of the cement-based material.

[0066] S203. For each training data set, input the training water-cement ratio, training cement strength, training compressive strength, and target first regression coefficient of each training data set into the compressive strength formula to obtain a theoretical second regression coefficient corresponding to each training data set.

[0067] Specifically, each set of training data sets collected in the above step may include the ratio of cement to cementitious materials (i.e., training cement ratio), sand-binder ratio (i.e., training sand-binder ratio), W / B (i.e., training water-binder ratio), fiber type, fiber diameter, fiber length, fiber aspect ratio, fiber volume percentage (i.e., training fiber data), (i.e. training cement strength), (i.e., training compressive strength), etc. This step can extract the W / B, 、 These three parameters, and let =1.61 (i.e. the target first regression coefficient), and then the theoretical second regression coefficient corresponding to each training data set can be calculated according to the compressive strength formula, that is .

[0068] S204: Establish a first preliminary prediction model for fiber cement-based materials, where the first preliminary prediction model is a random forest regressor.

[0069] In this step, a random forest of fiber cement-based materials can be established based on the above dataset. Prediction model (i.e. the first preliminary prediction model), it can be preliminarily determined that the input features of the first preliminary prediction model are the proportion of cement in cementitious materials, sand-binder ratio, W / B, fiber type, fiber diameter, fiber length, fiber aspect ratio, fiber volume percentage, , the output feature is the theoretical second regression coefficient .

[0070] Furthermore, the input features can be preprocessed before model training. For example, the input features and output features can be spliced ​​into a feature matrix in the form of column vectors in advance, the correlation coefficient matrix between the input feature columns (excluding the last column) can be calculated, and the correlation between the features can be visualized through a heat map (seaborn.heatmap) to facilitate a preliminary understanding of the variable relationship; the classification feature FT (i.e. fiber type) can be one-hot encoded (pd.get_dummies) to convert it into a numerical variable that can be processed by the model, and the output feature column can be confirmed again. Whether it exists in the data.

[0071] S205. Based on the training water-binder ratio, training cement strength, training cement ratio, training sand-binder ratio, and training fiber data of multiple training data sets, and the theoretical second regression coefficient corresponding to each training data set, the first preliminary prediction model is trained to obtain a first target prediction model for the fiber cement-based material.

[0072] During the specific model training process, the input feature data set and the output feature data can first be divided into a training set (X_train, Y_train) and a test set (X_test, Y_test) in a 9:1 ratio to ensure the objectivity of the model evaluation.

[0073] Then, you can initialize the RandomForestRegressor and perform hyperparameter tuning through grid search (GridSearchCV). The parameter grid can include key parameters such as the number of trees (n_estimators), maximum depth (max_depth), split and minimum number of leaf nodes, etc., and use 5-fold cross validation (cv=5) to The score is used as an evaluation indicator to screen out the optimal parameter combination.

[0074] Finally, the random forest model with the optimal parameters (i.e., the first target prediction model) can be used to make predictions on the training set and the test set, respectively, and the mean square error (MSE) and Goodness of fit: evaluate the model's fitting effect and generalization ability; further verify the model's stability through 5-fold cross validation, and output the average cross validation result. The score and standard deviation comprehensively reflect the performance of the model.

[0075] Figure 3 is a second regression coefficient provided in an embodiment of the present application The comparison diagram of the true value and the predicted value is as follows: Figure 3 As shown, the horizontal axis is the second regression coefficient The true value of the first target prediction model is shown on the ordinate, and the predicted value of the second regression coefficient output by the first target prediction model is shown on the ordinate. This embodiment calculates the prediction results of the first target prediction model in the training set and the test set respectively, compares the prediction results with their true values, and uses The model was evaluated using the score as the evaluation index. The evaluation indexes of the first target prediction model in the training set and the test set were =0.991 and =0.949.

[0076] S206: Construct a second target prediction model for fiber cement-based materials based on the multiple training data sets and the first target prediction model.

[0077] S207. Based on the objective function and constraints of the fiber cement-based material, calculate the mix ratio of the materials added to the fiber cement-based material.

[0078] This embodiment provides a material ratio intelligent optimization method that integrates domain knowledge and machine learning. By calculating the theoretical second regression coefficient corresponding to each group of training data sets, and then based on the training water-binder ratio, training cement strength, training cement ratio, training sand-binder ratio and training fiber data of multiple training data sets, as well as the theoretical second regression coefficient corresponding to each group of training data sets, a first preliminary prediction model is trained to obtain a first target prediction model of the fiber cement-based material. This improves the accuracy of the first target prediction model, provides an accurate model basis for the subsequent calculation of the mix ratio of the materials added to the fiber cement-based material, and further realizes the precise regulation of the performance of the fiber cement-based material.

[0079] As a feasible implementation method, a second target prediction model for fiber cement-based materials is constructed based on multiple sets of training data sets and the first target prediction model, including: For each training data set, the training water-binder ratio, training cement strength, training cement ratio, training sand-binder ratio, and training fiber data of each training data set are input into the first target prediction model to obtain the actual second regression coefficient corresponding to each training data set; Establishing a second preliminary prediction model for fiber cement-based materials, wherein the second preliminary prediction model is a multi-layer perceptron model, the multi-layer perceptron model has two hidden layers, each hidden layer contains 64 neurons, and the activation function of the multi-layer perceptron model is a ReLU function; According to each set of training data and the actual second regression coefficient corresponding to each set of training data, the second preliminary prediction model is trained to obtain a second target prediction model for the fiber cement-based material.

[0080] In the process of constructing the second target prediction model, the training data and benchmark model required for model training can be prepared in advance. For example, the training data set for prediction can be extracted from each set of training data. The feature data (i.e., the second regression coefficient) includes, for example, training water-cement ratio, training cement strength, training cement ratio, training sand-cement ratio, and training fiber data. By inputting the extracted feature data into the first target prediction model, the actual second regression coefficient corresponding to each training dataset can be obtained. Furthermore, a multilayer perceptron (MLP) model (i.e., the second preliminary prediction model) can be defined. Specifically, the MLPModel class can be defined, inheriting from nn.Module, to construct a multilayer perceptron model with two hidden layers (64 neurons per layer) and a ReLU activation function. The input dimension can be determined based on the feature dimension of the training set, and the output dimension is 1. This model is used to predict the 28-day compressive strength of fiber-cement-based materials.

[0081] Subsequently, the second preliminary prediction model can be trained according to each set of training data and the actual second regression coefficient corresponding to each set of training data, so as to obtain the second target prediction model of the fiber cement-based material. The model training process can, for example, first calculate the corresponding theoretical compressive strength based on the actual second regression coefficient corresponding to each set of training data, calculate the predicted compressive strength output by the second preliminary prediction model based on the specific content of each set of training data, and then gradually train the second preliminary prediction model so that the output predicted compressive strength is close to the theoretical compressive strength, that is, gradually reduce the error between the predicted compressive strength and the theoretical compressive strength. When the number of training times reaches the maximum number of iterations, the final second target prediction model can be obtained, which can be used for subsequent prediction of the actual compressive strength of the fiber cement-based material.

[0082] As a feasible implementation method, each training data set also includes: training group ratio and training water reducer ratio; According to each set of training data and the actual second regression coefficient corresponding to each set of training data, the second preliminary prediction model is trained to obtain a second target prediction model of the fiber cement-based material, including: determining the second preliminary prediction model as a second current prediction model for the fiber cement-based material; The training group ratio, training water-reducing agent ratio, training water-binder ratio, training sand-binder ratio, training fiber data and training cement strength of each training data set are input into the second current prediction model to obtain the predicted compressive strength corresponding to each training data set; The training water-cement ratio, training cement strength, target first regression coefficient and actual second regression coefficient of each training data set are input into the compressive strength formula to obtain the theoretical compressive strength corresponding to each training data set; updating the model parameters of the second current prediction model according to the predicted compressive strength and the theoretical compressive strength corresponding to each set of training data sets; Return to executing the step of inputting the training group ratio, training water-reducing agent ratio, training water-binder ratio, training sand-binder ratio, training fiber data, and training cement strength of each training data set into the second current prediction model to obtain the predicted compressive strength corresponding to each training data set, until the number of updates of the second current prediction model reaches the maximum number of iterations; The second current prediction model corresponding to the maximum number of iterations is determined as the second target prediction model of the fiber cement-based material.

[0083] During the specific process of each training round, the model parameters of the second current prediction model can be gradually updated according to the predicted compressive strength and theoretical compressive strength corresponding to each set of training data sets. The specific process of updating the model parameters is not limited. For example, the model parameters of the second current prediction model can be updated directly based on the size between the predicted compressive strength and the theoretical compressive strength, or a series of loss calculations can be performed on the predicted compressive strength and the theoretical compressive strength, and then the model parameters can be updated according to the results of the specific loss calculations. Alternatively, the model parameters of the second current prediction model can be updated according to other methods, and this embodiment does not further limit this.

[0084] As a feasible implementation method, the model parameters of the second current prediction model are updated according to the predicted compressive strength and the theoretical compressive strength corresponding to each set of training data sets, including: Based on the predicted compressive strength and theoretical compressive strength corresponding to each training data set, the physical loss value and data loss value of each training data set are calculated respectively; Based on the physical loss value and data loss value of each training data set, calculate the total loss value of multiple training data sets; The model parameters of the second current prediction model are updated according to the total loss value.

[0085] During the update process of specific model parameters, the update of model parameters can be comprehensively guided by calculating the physical loss value and the data loss value in each training round, and the training of the second current prediction model can be gradually realized. Among them, the physical loss value can reflect the physical consistency loss of multiple sets of training data sets. By relying on the physical consistency loss, the model output can not only match the observed data, but also conform to the dynamic behavior of the physical system. For example, the physical loss value can be the error between the predicted compressive strength and the theoretical compressive strength. The data loss value can reflect the supervised learning loss of multiple sets of training data sets. By relying on the difference between the model prediction results and the true labels, the model parameters can be optimized to reduce the error. For example, the data loss value can be the error between the predicted compressive strength and the true strength.

[0086] The following is an exemplary description of the main process of constructing the second target prediction model: (1) First, a multilayer perceptron (MLP) model can be selected as the main model to read and preprocess the training data of the MLP model. For example, the input features of the MLP model can be extracted from multiple sets of training data sets: various component features of fiber cement-based materials, including the proportion of cement, fly ash, mineral powder, silica fume and other cementitious materials, sand-binder ratio, water-binder ratio, high-efficiency water reducer percentage, cement strength, etc. , as well as detailed properties of the fiber (fiber type, diameter, length, aspect ratio, density, tensile strength, elastic modulus, percentage), the output feature can be 28-day compressive strength (i.e. target variable); the training data can then be divided into a feature matrix X_raw (composed of input features, excluding columns) and the target variable Y_raw( The feature matrix is ​​further distinguished between numerical and categorical types. The categorical features (FT) are one-hot encoded using OneHotEncoder. The encoded categorical features can then be concatenated with the numerical features to obtain the complete feature matrix X_all. The preliminarily processed data is split into training and test sets at a ratio of 9:1. The feature matrix and target variable are standardized to make the data have zero mean and unit variance, which is convenient for subsequent model training.

[0087] (2) Next, you can prepare random forest related variables. The specific steps may include: 1) First, there is a trained random forest model file "rf_alpha_a.pkl" (i.e., the first target prediction model). You can load the random forest model file into the code through the "joblib.load" function to obtain the model object "rf_alpha_a".

[0088] 2) Can be clearly used for prediction The characteristic names of the materials are as follows: C represents the proportion of cement in cementitious materials, SCR represents the sand-binder ratio, W / B represents the water-binder ratio, FDia represents the fiber diameter, FL represents the fiber length, FAR represents the fiber aspect ratio, and FV represents the fiber volume fraction. Represents cement strength, FT_BF represents BF fiber, FT_PE represents PE fiber, FT_POM represents POM fiber, FT_PP represents PP fiber, FT_PVA represents PVA fiber, and FT_carbon fiber represents carbon fiber. The above feature names are stored in the "alpha_a_features" list and need to correspond to the features used in random forest model training.

[0089] 3) The code determines the index position of the above feature among all preprocessed features. Because features include both numerical and one-hot-encoded categorical features, we can first record the number of numerical features, n_numerical , and then iterate over each feature in alpha_a_features . If the feature is in the numerical feature column, numerical_cols , its index in numerical_cols is added to the list alpha_a_feature_indices . If the feature is not in the numerical feature column, we iterate over the one-hot-encoded categorical feature names, cat_feature_names . When a name beginning with that feature is found, its corresponding index (the number of numerical features plus the current index position) is added to the list alpha_a_feature_indices .

[0090] 4) Based on the column names of all features "X_feature_names" (consisting of numerical feature column names and categorical feature names after one-hot encoding), the index recorded in "alpha_a_feature_indices" can be used to obtain the predictions for the random forest model The input feature column names are stored in the "rf_feature_names" list for subsequent random forest model Be prepared for predictions and operations such as combining physical model calculations.

[0091] (3) Then, we can build the dataset and data loader: define the ConcreteDataset class, which inherits from Dataset. In the class, we perform tensor conversion on the input features and target variables, and implement methods for obtaining the dataset length and obtaining data samples by index. For example, we can create an instance of this class called train_dataset based on the training dataset, and then encapsulate it with DataLoader. We set the batch size to 32 and shuffle the data order to facilitate loading data in batches during training.

[0092] (4) Then, a physical loss function can be defined and used to calculate the physical loss: for example, a physical loss function named “physics_loss” is defined, whose core function is to introduce the physical theoretical laws of the strength of fiber cement-based materials into model training, and constrain the prediction results of the neural network to be consistent with domain knowledge.

[0093] In the process of calculating physical loss, the physics_loss function first converts the input feature tensor into a NumPy array through "detach().cpu().numpy()" to adapt to the input format of the subsequent random forest model. Then, the "alpha_a_feature_indices" index can be used to extract the features for prediction from the converted array. The feature data is collected and organized into a DataFrame with column names corresponding to the "rf_feature_names" used in random forest training. Ensure that the input format matches.

[0094] The pre-trained random forest model file can then be used to predict the above features and obtain the key parameters in the physical formula At the same time, the cementitious material strength ( ) and water-binder ratio (W / B) values ​​are adjusted to column vector form to meet subsequent calculation requirements.

[0095] In order to implement tensor operations, the random forest model file can be used to predict Convert to a PyTorch tensor, adjust the shape, and transfer to the same device as the input features (GPU or CPU). After that, the theoretical strength can be calculated according to the compressive strength formula of cement-based materials. ,in, It is a fixed parameter set based on domain knowledge (default 1.61).

[0096] Finally, the model prediction strength can be calculated by the Mean Squared Error Loss (MSE Loss) function and theoretical strength The difference between gets physical loss and returns.

[0097] (5) Finally, the model can be trained in conjunction with physical loss: First, the code automatically selects the training device based on the hardware conditions, prioritizing the use of the GPU to improve computational efficiency. If a GPU is not available, the CPU is used. Then, the multi-layer perceptron (MLP) model can be initialized and its parameters migrated to the selected device. At the same time, the Adam optimizer (learning rate of 0.001) and the mean squared error loss function (MSELoss) can be configured for subsequent parameter optimization and data loss calculation. Before the training process begins, 100 training rounds can be set in advance (e.g., num_epochs=100), and lambda_ can be set to balance the weights of data loss and physical loss. Among them, the optimal training parameter lambda_=0.0064 can be obtained by manual parameter adjustment.

[0098] In each training round, the model can be switched to training mode, and the cumulative variables of total loss, data loss, and physical loss can be initialized; the training data loader (train_loader) can be used to read data in batches, each batch of data can contain 32 samples, and the input features and true labels can be migrated to the training device.

[0099] For each batch of data, the model predicts the input features to obtain the strength prediction value of the fiber cement-based material. The total loss can then be calculated by calculating the data loss and physical loss separately: the data loss (loss_data) can refer to the mean square error between the predicted value and the true label to ensure that the model fits the training data; the physical loss (loss_phys) can be calculated by calling the defined physics_loss function to constrain the model prediction to conform to the theoretical law of the strength of fiber cement-based materials; the total loss is the sum of the data loss and the physical loss multiplied by the weight, such as .

[0100] After the loss calculation is complete, the optimizer clears the previous gradient (optimizer.zero_grad()), backpropagates the total loss to calculate the gradient (loss.backward()), and updates the model parameters of the multilayer perceptron model based on the gradient (optimizer.step()). Simultaneously, the code accumulates the loss values ​​for each batch and prints the current total loss, data loss, and physical loss every 10 epochs, making it easier to monitor training progress and loss trends.

[0101] After 100 rounds of training, the code will output a "Training Completed" prompt. The entire training process combines data-driven and physical constraints to guide model training. This not only ensures the model's ability to fit the data, ensuring that the prediction results are consistent with the actual data, but also ensures that the prediction results conform to the physical laws of fiber cement-based materials, thereby improving the model's generalization ability and predictive rationality.

[0102] Furthermore, after the model training is completed, the model can be set to evaluation mode, and predictions can be made on the test set and training set respectively. When predicting, the data can be converted into tensors and moved to the device. After the model outputs the results, it is converted back to NumPy arrays and the actual predicted values ​​are obtained through normalized inverse transformation. Then, the coefficient of determination (R2_score) is calculated. ) Evaluate the goodness of fit of the model to the data and use Mean_Absolute_Error to calculate the mean absolute error (MAE) to measure the average deviation between the predicted value and the true value; finally, you can print the evaluation results on the training set and test set.

[0103] Figure 4 This is an embodiment of the present application. The comparison diagram of the true value and the predicted value is as follows: Figure 4 As shown in (a) and (b), the outputs of different second target prediction models are compared for the training set and the test set. The difference between the predicted value and the true value of The model was evaluated using the score as the evaluation index, where Figure 4 The second target prediction model in (a) is a neural network (NN) model that does not consider physical constraints, that is, the predicted value in (a) is the result of the compressive strength neural network prediction without incorporating physical formulas; Figure 4 The second target prediction model in (b) is the Physics-Informed Neural Networks (PINN) model that takes physical constraints into account. That is, the predicted value in (b) is the result of the compressive strength neural network prediction that incorporates physical formulas. It can be seen that for both the training set and the test set, Figure 4 The prediction effect of (b) is improved compared with (a), such as the evaluation index of the training set. The evaluation index of the test set increased from 0.894 to 0.966. The PINN model improved from 0.839 to 0.925, and the goodness of fit of the data to the PINN model was higher than that of the NN model, and the error between the predicted intensity results and the actual intensity results was smaller.

[0104] From the above description, it can be seen that the material proportion intelligent optimization method that integrates domain knowledge and machine learning provided in this embodiment involves a dual-driven design that integrates professional knowledge in the field of cement-based materials and machine learning algorithms. It is suitable for the mix design, performance prediction and multi-objective optimization of fiber-reinforced cement-based materials. It can be widely used in fields such as building structures, bridge engineering, municipal construction, etc. that have comprehensive requirements for material ductility, strength, economy and low carbon, and provide technical support for the research and development and application of high-performance, green cement-based materials.

[0105] For example, this embodiment builds a flexible and scalable multi-objective optimization framework, rather than being limited to a single fixed solution. Engineers can dynamically adjust optimization directions based on the core requirements of specific projects in actual engineering scenarios. For example, by fine-tuning the weights of compressive strength, carbon footprint, and cost, multiple mix ratios tailored to different priorities can be quickly generated. Ultimately, the optimal balance between performance compliance, green and low-carbon development, and economic controllability is found, providing engineers with multiple options for specific design scenarios.

[0106] Whether it is a special structure that focuses on ensuring strength, a livelihood project that pursues the ultimate cost-effectiveness, or a benchmark project with low carbon as the core, the appropriate proportion direction can be quickly locked in through weight adjustment, and the specific parameters corresponding to multiple sets of Pareto optimal solutions (such as the amount of each material, predicted strength value, carbon footprint value, cost details, etc.) can be simultaneously output. This allows engineers to grasp the core needs in the solution selection while taking into account the balance of other dimensions, significantly improving the efficiency and accuracy of the mix design.

[0107] Corresponding to the material ratio intelligent optimization method integrating domain knowledge and machine learning in the above embodiment, Figure 5 This is a structural block diagram of a material ratio intelligent optimization device that integrates domain knowledge and machine learning, provided in one embodiment of the present application. For ease of explanation, only the parts related to the embodiment of the present application are shown.

[0108] Reference Figure 5 , the device comprises: An acquisition module 301 is configured to acquire multiple sets of training data sets of fiber cement-based materials, wherein each set of training data sets includes at least: a training cement ratio, a training sand-binder ratio, and training fiber data of the fiber cement-based materials; A regression coefficient determination module 302 is configured to determine a target first regression coefficient for the fiber cement-based material based on multiple training data sets and a compressive strength formula for the cement-based material, wherein the compressive strength formula is determined based on at least the first regression coefficient and the second regression coefficient for the cement-based material, and the target first regression coefficient is used to calculate the actual compressive strength of the fiber cement-based material; A first model building module 303 is configured to build a first target prediction model for the fiber cement-based material based on multiple sets of training data sets and the target first regression coefficient, wherein the first target prediction model is used to predict the actual second regression coefficient of the fiber cement-based material, and the actual second regression coefficient is used to calculate the actual compressive strength of the fiber cement-based material; A second model building module 304 is configured to build a second target prediction model for the fiber cement-based material based on the multiple training data sets and the first target prediction model, wherein the second target prediction model is configured to predict the actual compressive strength of the fiber cement-based material; The calculation module 305 is used to calculate the mix ratio of the added materials in the fiber cement-based material based on the objective function and constraints of the fiber cement-based material, wherein at least part of the objective function is determined based on the actual compressive strength and target compressive strength output by the second objective prediction model.

[0109] The present embodiment provides a material ratio intelligent optimization device that integrates domain knowledge and machine learning. The device obtains multiple sets of training data sets of fiber cement-based materials through an acquisition module, wherein each set of training data sets includes at least: training cement ratio, training sand-glue ratio and training fiber data of the fiber cement-based materials; the regression coefficient determination module determines the target first regression coefficient of the fiber cement-based material based on the multiple sets of training data sets and the compressive strength formula of the cement-based material. The compressive strength formula is determined based on at least the first regression coefficient and the second regression coefficient of the cement-based material. The target first regression coefficient is used to calculate the actual compressive strength of the fiber cement-based material; the first model construction module determines the target first regression coefficient based on the multiple sets of training data sets and the target second regression coefficient. A first target prediction model for a fiber cement-based material is constructed using a regression coefficient. The first target prediction model is used to predict the actual second regression coefficient of the fiber cement-based material. The actual second regression coefficient is used to calculate the actual compressive strength of the fiber cement-based material. A second model construction module constructs a second target prediction model for the fiber cement-based material based on multiple training data sets and the first target prediction model. The second target prediction model is used to predict the actual compressive strength of the fiber cement-based material. A calculation module calculates the mix ratio of the added materials in the fiber cement-based material based on the objective function and constraints of the fiber cement-based material, wherein at least part of the objective function is determined based on the second target prediction model and the target compressive strength. Using this device, two regression coefficients applicable to the fiber cement-based material are determined by integrating fiber data into the compressive strength formula. Then, a prediction model for predicting the compressive strength of the fiber cement-based material is constructed using the two determined regression coefficients. This allows the influence of fiber incorporation to be taken into account when calculating the material mix ratio, improving the design accuracy of the mix ratio and thus achieving precise control of the performance of the fiber cement-based material.

[0110] Optionally, each training data set also includes: training water-cement ratio, training cement strength, training compressive strength, and the compressive strength formula is ,in, Represents compressive strength, Represents the strength of cement, W / B represents the water-to-binder ratio, represent the first and second regression coefficients, respectively; The first model building module is specifically used to: For each training data set, the training water-cement ratio, training cement strength, training compressive strength and target first regression coefficient of each training data set are input into the compressive strength formula to obtain the theoretical second regression coefficient corresponding to each training data set; Establishing a first preliminary prediction model for fiber cement-based materials, the first preliminary prediction model is a random forest regressor; According to the training water-binder ratio, training cement strength, training cement ratio, training sand-binder ratio and training fiber data of multiple training data sets, as well as the theoretical second regression coefficient corresponding to each training data set, the first preliminary prediction model is trained to obtain the first target prediction model of fiber cement-based materials.

[0111] Optionally, the second model building module includes: An input unit is used to input the training water-binder ratio, training cement strength, training cement ratio, training sand-binder ratio, and training fiber data of each training data set into the first target prediction model to obtain an actual second regression coefficient corresponding to each training data set; A model building unit is used to establish a second preliminary prediction model of the fiber cement-based material, wherein the second preliminary prediction model is a multi-layer perceptron model, the multi-layer perceptron model has two hidden layers, each hidden layer includes 64 neurons, and the activation function of the multi-layer perceptron model is a ReLU function; The model training unit is used to perform model training on the second preliminary prediction model according to each set of training data and the actual second regression coefficient corresponding to each set of training data, so as to obtain a second target prediction model of the fiber cement-based material.

[0112] Optionally, each training data set further includes: a training group ratio and a training water reducer ratio; The model training unit includes: a first determining subunit, configured to determine the second preliminary prediction model as a second current prediction model for the fiber cement-based material; The first input subunit is used to input the training group ratio, training water-reducing agent ratio, training water-binder ratio, training sand-binder ratio, training fiber data and training cement strength of each training data set into the second current prediction model to obtain the predicted compressive strength corresponding to each training data set; The second input subunit is used to input the training water-binder ratio, training cement strength, target first regression coefficient and actual second regression coefficient of each training data set into the compressive strength formula to obtain the theoretical compressive strength corresponding to each training data set; an updating subunit, configured to update the model parameters of the second current prediction model according to the predicted compressive strength and the theoretical compressive strength corresponding to each set of training data sets; Returning the execution subunit, which is used to return to the step of inputting the training group ratio, training water-reducing agent ratio, training water-binder ratio, training sand-binder ratio, training fiber data, and training cement strength of each training data set into the second current prediction model to obtain the predicted compressive strength corresponding to each training data set, until the number of updates of the second current prediction model reaches the maximum number of iterations; The second determining subunit is used to determine the second current prediction model corresponding to the maximum number of iterations as the second target prediction model of the fiber cement-based material.

[0113] Optionally, the updating subunit is specifically configured to: Based on the predicted compressive strength and theoretical compressive strength corresponding to each training data set, the physical loss value and data loss value of each training data set are calculated respectively; Based on the physical loss value and data loss value of each training data set, calculate the total loss value of multiple training data sets; The model parameters of the second current prediction model are updated according to the total loss value.

[0114] Optionally, the regression coefficient determination module is specifically used to: determining a first initial value as a current first regression coefficient of the fiber cement-based material, and determining a second initial value as a current second regression coefficient of the fiber cement-based material; For each training data set, the training water-cement ratio, training cement strength, current first regression coefficient, and current second regression coefficient of each training data set are input into the compressive strength formula to obtain the theoretical compressive strength of each training data set; Based on the theoretical compressive strength and training compressive strength of each training data set, calculate the current residual sum of squares of multiple training data sets; According to the current residual sum of squares, the current first regression coefficient and the current second regression coefficient are corrected, and the step of inputting the training water-cement ratio, training cement strength, current first regression coefficient, and current second regression coefficient of each training data set into the compressive strength formula to obtain the theoretical compressive strength of each training data set is returned to execution for each training data set, until the current residual sum of squares reaches convergence or the number of corrections to the current first regression coefficient reaches the maximum number of iterations; The current first regression coefficient corresponding to when the current residual sum of squares reaches convergence or the number of corrections to the current first regression coefficient reaches the maximum number of iterations is determined as the target first regression coefficient of the fiber cement-based material.

[0115] Optionally, the objective function includes a first objective function aiming at minimizing the compressive strength error, a second objective function aiming at minimizing carbon emissions, and a third objective function aiming at minimizing material costs, wherein the first objective function is determined based on the actual compressive strength and the target compressive strength predicted by the second objective prediction model; The calculation module is specifically used for: respectively determining a first function weight of the first objective function, a second function weight of the second objective function, and a third function weight of the third objective function; According to the first objective function, the second objective function, the third objective function, the first function weight, the second function weight, the third function weight, and the constraints, a non-dominated sorting genetic algorithm is used to calculate the mix ratio of materials added to the fiber cement-based material. The mix ratio includes at least target cement strength, target component ratio, target water-reducing agent ratio, target water-binder ratio, target sand-binder ratio and target fiber data.

[0116] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of this application. Their specific functions and technical effects can be found in the method embodiment section and will not be repeated here.

[0117] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0118] Figure 6 This is a schematic diagram of the structure of a terminal device provided by an embodiment of the present application. Figure 6 As shown, the terminal device 500 of this embodiment includes: at least one processor 502 ( Figure 6 Only one is shown in the figure) a processor, a memory 501, and a computer program 503 stored in the memory 501 and executable on at least one processor 502. When the processor 502 executes the computer program 503, the steps in the control method embodiment of any of the above-mentioned applications are implemented.

[0119] The terminal device 500 can be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The terminal device may include, but is not limited to, a processor 502 and a memory 501. Those skilled in the art will understand that Figure 6 This is merely an example of the terminal device 500 and does not constitute a limitation on the terminal device 500 . The terminal device 500 may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the terminal device 500 may also include input and output devices, network access devices, etc.

[0120] The processor 502 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor.

[0121] In some embodiments, the memory 501 may be an internal storage unit of the terminal device 500, such as a hard drive or memory of the terminal device 500. In other embodiments, the memory 501 may also be an external storage device of the terminal device 500, such as a plug-in hard drive, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. equipped on the terminal device 500. Furthermore, the memory 501 may include both an internal storage unit of the terminal device 500 and an external storage device. The memory 501 is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of a computer program. The memory 501 may also be used to temporarily store data that has been output or is about to be output.

[0122] The embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by the processor 502, the steps in the above-mentioned method embodiments can be implemented.

[0123] An embodiment of the present application provides a computer program product. When the computer program product is run on a terminal device, the terminal device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0124] If the integrated unit is implemented as a software functional unit and sold or used as a standalone product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application can implement all or part of the process steps in the above-mentioned method embodiments by using a computer program to instruct the relevant hardware. The computer program can be stored in a computer-readable storage medium. When executed by processor 502, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. A computer-readable storage medium can include at least: any entity or device capable of carrying computer program code to a device / terminal equipment, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signals, telecommunication signals, and software distribution media. Examples include USB flash drives, removable hard drives, magnetic disks, or optical disks. In some jurisdictions, based on legislation and patent practice, computer-readable storage media cannot be electric carrier signals or telecommunication signals.

[0125] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0126] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0127] In the embodiments provided in this application, it should be understood that the disclosed devices / terminal devices and methods can be implemented in other ways. For example, the device / terminal device embodiments described above are merely illustrative. For example, the division of modules or units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0128] Units described as separate components may or may not be physically separate, and 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 these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0129] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A material ratio intelligent optimization method integrating domain knowledge and machine learning, characterized in that: include: Acquire multiple sets of training data sets of fiber cement-based materials, wherein each set of the training data sets at least includes: a training cement ratio, a training sand-binder ratio, and training fiber data of the fiber cement-based materials; Determining a target first regression coefficient of the fiber cement-based material based on the multiple sets of training data sets and a compressive strength formula of the cement-based material, wherein the compressive strength formula is determined based on at least the first regression coefficient and the second regression coefficient of the cement-based material, and the target first regression coefficient is used to calculate the actual compressive strength of the fiber cement-based material; constructing a first target prediction model for the fiber cement-based material based on the multiple sets of training data sets and the target first regression coefficient, wherein the first target prediction model is used to predict an actual second regression coefficient of the fiber cement-based material, and the actual second regression coefficient is used to calculate an actual compressive strength of the fiber cement-based material; Constructing a second target prediction model for the fiber cement-based material based on the multiple sets of training data sets and the first target prediction model, wherein the second target prediction model is used to predict the actual compressive strength of the fiber cement-based material; Based on the objective function and constraints of the fiber cement-based material, the mix ratio of the materials added to the fiber cement-based material is calculated, wherein at least part of the objective function is determined based on the actual compressive strength and target compressive strength output by the second objective prediction model.

2. The intelligent optimization method for material ratios integrating domain knowledge and machine learning according to claim 1 is characterized in that: Each set of training data sets also includes: training water-cement ratio, training cement strength, and training compressive strength. The compressive strength formula is: ,in, Represents compressive strength, Represents the strength of cement, W / B represents the water-to-binder ratio, represent the first and second regression coefficients, respectively; The step of constructing a first target prediction model for the fiber cement-based material based on the multiple sets of training data sets and the target first regression coefficient includes: For each set of the training data sets, the training water-cement ratio, the training cement strength, the training compressive strength, and the target first regression coefficient of each set of the training data sets are input into the compressive strength formula to obtain a theoretical second regression coefficient corresponding to each set of the training data sets; Establishing a first preliminary prediction model for the fiber cement-based material, wherein the first preliminary prediction model is a random forest regressor; According to the training water-binder ratio, the training cement strength, the training cement ratio, the training sand-binder ratio and the training fiber data of the multiple sets of training data sets, and the theoretical second regression coefficient corresponding to each set of the training data sets, the first preliminary prediction model is trained to obtain the first target prediction model of the fiber cement-based material.

3. The intelligent optimization method for material ratios integrating domain knowledge and machine learning according to claim 2 is characterized in that: The step of constructing a second target prediction model for the fiber cement-based material based on the multiple sets of training data sets and the first target prediction model includes: For each set of the training data sets, the training water-binder ratio, the training cement strength, the training cement ratio, the training sand-binder ratio, and the training fiber data of each set of the training data sets are input into the first target prediction model to obtain an actual second regression coefficient corresponding to each set of the training data sets; Establishing a second preliminary prediction model of the fiber cement-based material, wherein the second preliminary prediction model is a multi-layer perceptron model, the multi-layer perceptron model has two hidden layers, each hidden layer includes 64 neurons, and the activation function of the multi-layer perceptron model is a ReLU function; According to each group of the training data sets and the actual second regression coefficient corresponding to each group of the training data sets, the second preliminary prediction model is trained to obtain the second target prediction model of the fiber cement-based material.

4. The intelligent optimization method for material ratios by integrating domain knowledge and machine learning as claimed in claim 3 is characterized in that: Each set of training data sets also includes: training group ratio and training water reducer ratio; The second preliminary prediction model is trained according to each set of the training data sets and the actual second regression coefficient corresponding to each set of the training data sets to obtain the second target prediction model of the fiber cement-based material, including: determining the second preliminary prediction model as a second current prediction model of the fiber cement-based material; Inputting the training group ratio, the training water-reducing agent ratio, the training water-binder ratio, the training sand-binder ratio, the training fiber data, and the training cement strength of each training data set into the second current prediction model to obtain the predicted compressive strength corresponding to each training data set; Inputting the training water-cement ratio, the training cement strength, the target first regression coefficient, and the actual second regression coefficient of each set of the training data sets into the compressive strength formula to obtain the theoretical compressive strength corresponding to each set of the training data sets; updating the model parameters of the second current prediction model according to the predicted compressive strength and the theoretical compressive strength corresponding to each set of the training data sets; Returning to the step of inputting the training group ratio, the training water-reducing agent ratio, the training water-binder ratio, the training sand-binder ratio, the training fiber data, and the training cement strength of each training data set into the second current prediction model to obtain the predicted compressive strength corresponding to each training data set, until the number of updates of the second current prediction model reaches a maximum number of iterations; The second current prediction model corresponding to the maximum number of iterations is determined as the second target prediction model of the fiber cement-based material.

5. The intelligent optimization method for material ratios by integrating domain knowledge and machine learning as claimed in claim 4 is characterized in that: The updating of the model parameters of the second current prediction model according to the predicted compressive strength and the theoretical compressive strength corresponding to each set of the training data sets includes: Based on the predicted compressive strength and the theoretical compressive strength corresponding to each set of the training data sets, respectively calculating the physical loss value and the data loss value of each set of the training data sets; Calculating a total loss value of the plurality of training data sets based on the physical loss value and the data loss value of each set of the training data sets; The model parameters of the second current prediction model are updated according to the total loss value.

6. The intelligent optimization method for material ratios integrating domain knowledge and machine learning according to claim 2, characterized in that: The determining of a target first regression coefficient of the fiber cement-based material based on the multiple sets of training data sets and the compressive strength formula of the cement-based material includes: determining a first initial value as a current first regression coefficient of the fiber cement-based material, and determining a second initial value as a current second regression coefficient of the fiber cement-based material; For each set of the training data sets, inputting the training water-cement ratio, the training cement strength, the current first regression coefficient, and the current second regression coefficient of each set of the training data sets into the compressive strength formula to obtain the theoretical compressive strength of each set of the training data sets; Calculating the current residual sum of squares of the multiple sets of training data sets based on the theoretical compressive strength and the training compressive strength of each set of training data sets; According to the current residual sum of squares, the current first regression coefficient and the current second regression coefficient are corrected, and the step of inputting the training water-cement ratio, the training cement strength, the current first regression coefficient, and the current second regression coefficient of each training data set into the compressive strength formula to obtain the theoretical compressive strength of each training data set is returned to execution, until the current residual sum of squares reaches convergence or the number of corrections to the current first regression coefficient reaches a maximum number of iterations; The current first regression coefficient corresponding to when the current residual sum of squares reaches convergence or the number of corrections of the current first regression coefficient reaches the maximum number of iterations is determined as the target first regression coefficient of the fiber cement-based material.

7. The intelligent optimization method for material ratios by integrating domain knowledge and machine learning according to any one of claims 1 to 6, characterized in that: The objective function includes a first objective function with the goal of minimizing the compressive strength error, a second objective function with the goal of minimizing carbon emissions, and a third objective function with the goal of minimizing material costs, wherein the first objective function is determined based on the actual compressive strength and the target compressive strength predicted by the second objective prediction model; The calculating the mix ratio of the materials added to the fiber cement-based material based on the objective function and constraint conditions of the fiber cement-based material includes: respectively determining a first function weight of the first objective function, a second function weight of the second objective function, and a third function weight of the third objective function; According to the first objective function, the second objective function, the third objective function, the first function weight, the second function weight, the third function weight, and the constraints, a non-dominated sorting genetic algorithm is used to calculate the mix ratio of the materials added to the fiber cement-based material, where the mix ratio includes at least target cement strength, target component ratio, target water-reducing agent ratio, target water-binder ratio, target sand-binder ratio, and target fiber data.

8. A material ratio intelligent optimization device that integrates domain knowledge and machine learning, characterized in that: include: An acquisition module is used to acquire multiple sets of training data sets of fiber cement-based materials, wherein each set of the training data sets at least includes: a training cement ratio, a training sand-binder ratio and training fiber data of the fiber cement-based materials; a regression coefficient determination module, configured to determine a target first regression coefficient of the fiber cement-based material based on the multiple training data sets and a compressive strength formula of the cement-based material, wherein the compressive strength formula is determined based on at least the first regression coefficient and the second regression coefficient of the cement-based material, and the target first regression coefficient is used to calculate the actual compressive strength of the fiber cement-based material; a first model building module, configured to build a first target prediction model for the fiber cement-based material based on the multiple sets of training data sets and the target first regression coefficient, wherein the first target prediction model is used to predict an actual second regression coefficient of the fiber cement-based material, and the actual second regression coefficient is used to calculate an actual compressive strength of the fiber cement-based material; A second model building module is used to build a second target prediction model of the fiber cement-based material according to the multiple sets of training data sets and the first target prediction model, wherein the second target prediction model is used to predict the actual compressive strength of the fiber cement-based material; A calculation module is used to calculate the mix ratio of the added materials in the fiber cement-based material based on the objective function and constraints of the fiber cement-based material, wherein at least part of the objective function is determined based on the actual compressive strength and target compressive strength output by the second target prediction model.

9. A terminal device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the terminal device implements the method according to any one of claims 1 to 7.

10. A computer program product, characterized in that When the computer program product is run on a terminal device, the terminal device is caused to execute the method according to any one of claims 1 to 7.

Images