A Smart Test Decision-Making Method and Device for Foamed Lightweight Soil Proportioning
By constructing a standardized dataset and a pre-trained model, and combining SHAP value analysis and genetic algorithm optimization, the problems of low efficiency and insufficient transparency in foam lightweight soil mix design were solved, achieving efficient and accurate multi-objective optimization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA RAILWAY ENG CONSULTING GRP CO LTD
- Filing Date
- 2026-03-23
- Publication Date
- 2026-06-02
AI Technical Summary
Traditional foam lightweight soil mix design relies on experience and physical experiments, which is time-consuming and makes it difficult to reveal the relationship between material properties and mix parameters. Machine learning models have insufficient predictive performance in small sample scenarios and their logic is not transparent.
A standardized tabular dataset is constructed, and a network model is fitted using pre-trained tabular prior data for learning. Combined with SHAP value analysis and a non-dominated sorting genetic algorithm, the mix proportion of foamed lightweight soil is optimized to achieve multi-objective optimization.
It significantly improves modeling efficiency and prediction accuracy, provides quantifiable feature contribution relationships and multi-criteria decision support, and automatically searches for Pareto optimal matching schemes.
Smart Images
Figure CN122135850A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of building material design technology, and more specifically, to a method and apparatus for intelligent experimental decision-making on the proportioning of foamed lightweight soil. Background Technology
[0002] The mix design of foamed lightweight soil is a crucial step in determining its mechanical properties, lightweight characteristics, and economic cost. Traditional design methods rely on engineers' experience and trial-and-error through physical experiments, approximating the target by repeatedly adjusting component ratios. This process is time-consuming and cumbersome, and it is difficult to systematically reveal the complex relationship between material properties and mix parameters. Recent data-driven methods, primarily based on machine learning, often employ traditional models such as gradient boosting trees. Their performance heavily depends on large-scale, high-quality training data. However, experimental data in the field of foamed lightweight soil typically has small sample sizes, leading to low prediction accuracy in traditional models. Furthermore, the predictive logic of these models is opaque, making it difficult to translate the model's predictions into clearly defined physical guidelines for mix design. Summary of the Invention
[0003] The purpose of this invention is to provide an intelligent experimental decision-making method and device for foamed lightweight soil mix proportioning, in order to improve the above-mentioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows: Firstly, this application provides an intelligent experimental decision-making method for the proportioning of foamed lightweight soil, including: Experimental data of multi-source foamed lightweight soil were obtained and preprocessed to obtain a standardized tabular dataset containing input features and corresponding performance values; A standardized tabular dataset is used as context input to a pre-trained tabular prior data fitting network model. The tabular prior data fitting network model learns and predicts the input features to obtain the corresponding performance prediction value. The SHAP value analysis method is used to calculate the contribution of the input features to the performance prediction value in the tabular prior data fitting network model, thereby obtaining the contribution relationship between the input features and the performance prediction value. The network model fitted with the aforementioned tabular prior data is used as the performance prediction model, and a multi-objective optimization model is established by combining material cost and material density. Constraints are set for the multi-objective optimization model based on the aforementioned contribution relationship. The multi-objective optimization model is solved using a non-dominated sorting genetic algorithm to search for a set of Pareto optimal matching schemes.
[0004] Secondly, this application also provides an intelligent test decision-making device for foamed lightweight soil proportioning, comprising: The data acquisition module is used to acquire experimental data of multi-source foamed lightweight soil and preprocess it to obtain a standardized tabular dataset containing input features and corresponding performance values. The model prediction module is used to input a standardized tabular dataset as context to a pre-trained tabular prior data fitting network model, which learns and predicts the input features to obtain the corresponding performance prediction value. The contribution analysis module is used to calculate the contribution of the input features in the tabular prior data fitting network model to the performance prediction value using the SHAP value analysis method, so as to obtain the contribution relationship between the input features and the performance prediction value. The optimization modeling module is used to fit a network model with the tabular prior data as a performance prediction model, and to establish a multi-objective optimization model by combining material cost and material density, and to set constraints for the multi-objective optimization model according to the contribution relationship. The optimization solution module is used to solve the multi-objective optimization model using a non-dominated sorting genetic algorithm to search for a set of Pareto optimal matching schemes.
[0005] Thirdly, this application also provides an intelligent test decision-making device for foamed lightweight soil proportioning, comprising: Memory, used to store computer programs; A processor is used to implement the steps of the intelligent test decision-making method for foamed lightweight soil proportioning when executing the computer program.
[0006] Fourthly, this application also provides a readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described intelligent test decision-making method based on foamed lightweight soil proportioning.
[0007] The beneficial effects of this invention are as follows: 1. This invention constructs a standardized tabular dataset containing input features and corresponding performance values, and directly inputs it into a pre-trained tabular prior data to fit a network model for context learning. This eliminates the need for traditional large-scale model training and hyperparameter tuning, greatly reducing the dependence on the amount of data in the target domain. It solves the bottleneck of insufficient model generalization ability in small sample scenarios and significantly improves modeling efficiency and prediction accuracy.
[0008] 2. This invention uses the SHAP value analysis method to calculate the contribution value of input features to the performance prediction value, thereby obtaining the contribution relationship between input features and performance prediction value. This transforms the prediction logic of complex models into quantifiable feature contribution degree and direction, and enables the setting of reasonable constraints for subsequent model optimization based on the contribution relationship.
[0009] 3. This invention uses a tabular prior data to fit a network model as a performance prediction model, combines material cost and material density to establish a multi-objective optimization model, and applies a non-dominated sorting genetic algorithm to solve it, automatically searching for a set of Pareto optimal ratio schemes, which significantly improves the efficiency of ratio design and provides support for multi-criteria decision-making in different application scenarios.
[0010] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description
[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0012] Figure 1 This is a schematic diagram of the intelligent test decision-making method for foamed lightweight soil proportioning described in this embodiment of the invention; Figure 2 This is a diagram of the tabular prior data fitting network model structure described in this embodiment of the invention. Figure 3 This is a schematic diagram of the intelligent test decision-making device for foamed lightweight soil proportioning described in this embodiment of the invention. Figure 4 This is a schematic diagram of the intelligent test and decision-making device for foamed lightweight soil proportioning described in this embodiment of the invention.
[0013] Marked in the image: 800. Intelligent test and decision-making equipment for foamed lightweight soil proportioning; 801. Processor; 802. Memory; 803. Multimedia component; 804. I / O interface; 805. Communication component. Detailed Implementation
[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.
[0015] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0016] Example 1: This embodiment provides an intelligent test decision-making method for foamed lightweight soil mix proportions, which is applicable to the situation where designers need to quickly determine the mix proportion scheme that meets multiple objectives such as strength, density and cost in new construction projects.
[0017] See Figure 1 The figure shows that this method includes: S1. Obtain experimental data of multi-source foamed lightweight soil and preprocess it to obtain a standardized tabular dataset containing input features and corresponding performance values; Specifically, step S1 includes: S11. Filter out experimental data of foamed lightweight soil from the database, including the chemical composition, physical properties, mix proportions of raw materials, and corresponding curing age and performance indicators; Specifically, using the search terms "foamed lightweight soil," "foamed concrete," "mix ratio," "compressive strength," and "wet density," we screened out experimental research literature on foamed lightweight soil using various industrial solid wastes, including fly ash, slag, silica fume, and tailings, as admixtures. These research documents included experimental data on the chemical composition, physical properties, mix ratio, curing age, and corresponding performance values of the raw materials.
[0018] The chemical composition includes at least the percentages of bonded oxides such as SiO2, Al2O3, and CaO; the physical properties include the specific surface area of the cementitious materials and admixtures to characterize their reactivity; and the mix proportion includes the dosage of each component per cubic meter, in kg / m³. 3 .
[0019] S12. The experimental data of the foamed lightweight soil are cleaned and formatted. Specifically, the cleaning and formatting process includes unifying measurement units, correcting recording errors, and performing descriptive statistical analysis on continuous numerical characteristics, transforming scattered and heterogeneous experimental data into experimental data with unified format, comprehensive features, and complete information.
[0020] S13. Summarize the input features of the cleaned experimental data into material intrinsic property parameters, mix design parameters, and process condition parameters, and use the performance indicators as output targets to form the standardized tabular dataset; Specifically, the intrinsic material properties include the specific surface area and CaO, SiO2, and Al2O3 content of cement, as well as the specific surface area and main active oxide content of various admixtures. Its input characteristic can be expressed as: specific surface area 380m². 2 / kg, SiO2 content 52.3%, Al2O3 content 28.6%, CaO content 5.2%.
[0021] Specifically, the mix design parameters include the specific amounts of cement, various admixtures, water, foaming agents, and water-reducing agents per cubic meter of the mixture, all in kg / m³. 3 For components not used in the mix proportions, their usage characteristics are filled with "0" values to maintain data consistency. Their input characteristics can be represented as: cement 280kg / m³ 3 fly ash 120kg / m³ 3 Slag 0kg / m 3 Water 160kg / m 3 Foaming agent 2.5kg / m 3 Water-reducing agent 3.2kg / m 3 .
[0022] Specifically, the process condition parameters are the curing period, in days (d), and their input characteristics can be expressed as: 7d, 14d, 28d, etc.
[0023] In this embodiment, the output target is the mechanical performance index measured corresponding to the input features. The performance index can be expressed as: compressive strength 2.8 MPa, wet density 650 kg / m³. 3 Flowability 180mm.
[0024] Based on the above embodiments, this method further includes: S2. Use the standardized tabular dataset as context input to fit the network model to the pre-trained tabular prior data, such as... Figure 2As shown, the network model is fitted with the tabular prior data to learn and predict the input features, thereby obtaining the corresponding performance prediction value; Specifically, step S2 includes: S21. The standardized tabular dataset is randomly divided into a training subset and a test subset according to a preset ratio; In this embodiment, the standardized tabular dataset is randomly divided into a training subset and a test subset in an 8:2 ratio.
[0025] S22. The training subset is used as context input to the pre-trained tabular prior data fitting network model. The tabular prior data fitting network model predicts the test samples in the test subset based on the correspondence between the input features of the training subset and the performance indicators, and outputs the performance prediction value corresponding to the test samples. Specifically, step S2 includes: S221. Encode the training samples in the training subset and the test samples in the test subset together into an input sequence; Specifically, for a test sample The TabPFN model will use the training subset. and the sample to be tested They are jointly encoded into an input sequence. Wherein, the... Indicates a training subset. This represents the input features of the training samples. This represents the performance metrics of the training samples. Indicates the sequence number of the training sample. This indicates the total number of training samples.
[0026] In this embodiment, the input sequence includes the input features and corresponding performance metrics of all training samples in the training subset, as well as the input features of the sample to be tested.
[0027] S222. Using a bidirectional attention mechanism, calculate the association weight between any two elements in the input sequence. Specifically, the formula for calculating the association weight is as follows: ; in, Indicates the association weight. These represent the query matrix, key matrix, and value matrix, respectively, all of which are obtained from the input sequence through linear transformations. For transpose, It is the dimension of the key matrix.
[0028] In this embodiment, each element in the association weight matrix represents the association strength between two elements in the input sequence. The association weight between the test sample and each training sample in the training subset is extracted from the association weight matrix and denoted as . ,in, This represents the index of the training sample. The larger the association weight, the more important the training sample is to the current test sample.
[0029] S223. Based on the association weights, the relationship between the input features and performance of each training sample in the training subset is weighted and aggregated to obtain the performance prediction value corresponding to the sample to be tested: ; in, Represents the probability distribution function. This represents the predicted performance value of the sample to be tested. Indicates the sample to be tested. Indicates a training subset; This represents the generation of the hypothesis space from all possible data. This is based on a specific hypothesis, namely, a model of the relationship between a potential feature and a target variable. Indicates that under a given hypothesis Under the condition of given input features At that time, performance value The probability of occurrence.
[0030] It is after observing the training data Then, assuming This represents the posterior probability of the real relationship model. Indicates all Accumulate. This represents the function used to fit the pre-trained tabular prior data to the TabPFN model. Represents the pre-trained parameters. The TabPFN model represents the input features based on a subset of the training set and the test samples, and its output... The approximate posterior prediction distribution.
[0031] S23. Calculate the prediction accuracy of the tabular prior data fitting network model on the test subset based on the performance prediction value; In this embodiment, the error between the predicted performance value of each test sample and the corresponding true performance index in the test subset is calculated to obtain the prediction accuracy of the TabPFN model on the test subset.
[0032] S24. Using a preset regression evaluation index, verify the predictive performance of the tabular prior data fitting network model based on the prediction accuracy; To verify the effectiveness of the TabPFN model, the same training and testing subsets were used to train and predict XGBoost and LSTM models. Regression evaluation metrics were employed for comparison and verification; preferably, these metrics could be root mean square error, mean absolute error, or the A20 exponent. In this embodiment, the comparison results show that the TabPFN model outperforms other machine learning models in all evaluation metrics.
[0033] Based on the above embodiments, this method further includes: S3. Using the SHAP value analysis method, calculate the contribution value of the input features in the tabular prior data fitting network model to the performance prediction value, obtain the contribution relationship between the input features and the performance prediction value, quantify and visualize the degree and direction of the influence of each input feature on the performance prediction result, and reveal the intrinsic relationship between key design variables and performance.
[0034] In this embodiment, the SHAP value analysis method is used to provide a quantitative interpretation framework for the TabPFN model, transforming the predictive logic of the TabPFN model into domain knowledge that can be understood and trusted by materials engineers. The basic principle of the SHAP method is based on the Shapley value framework of cooperative game theory, and for input features... Its SHAP value is obtained by calculating the weighted average of the marginal contributions generated when the input feature is added to all possible feature subsets, and its definition is as follows: ; In the formula, Representing input features SHAP value, Indicates that it does not contain features Any feature subset ( ), Represents the complete set consisting of all features. and Representing sets and The cardinality (i.e., the number of features included). To define in the feature subset The value function on the feature set represents the use of only a subset of features. When considering the features in the model, the model's expected prediction value for the current instance is given.
[0035] Input features Add to feature subset The marginal contribution increment brought about by time, weighting coefficient This ensures that the order in which all possible input features are added is considered fairly.
[0036] Specifically, step S3 includes: S31. Calculate the contribution of each input feature to the predicted performance value for each sample; In this embodiment, the SHAP calculation library is used to calculate the TabPFN model and the standardized tabular dataset. Preferably, the SHAP calculation library is the shap library in Python. For each sample in the dataset, the SHAP value of each input feature of the sample is calculated using the SHAP value analysis method.
[0037] In this embodiment, the sum of the SHAP values of all input features of an input sample is equal to the deviation between the model prediction value of the sample and the average prediction value of the entire dataset, thereby ensuring the additivity and consistency of the interpretation.
[0038] S32. Aggregate the contribution values of each input feature for all samples to generate a global ranking of the contribution of the input features to the performance prediction values; Specifically, the average absolute SHAP value of the input feature is obtained by taking the absolute value of each input feature for all samples and then averaging them. ; In the formula, Representing input features The average absolute SHAP value, Represents the total number of samples. Indicates the first Input features of each sample The SHAP value.
[0039] Furthermore, all input features are sorted in descending order to generate a global feature importance map. This global feature importance map shows the main material properties and proportioning parameters that affect the performance of foamed lightweight soil in all predictions. For example, the water-cement ratio, cement content, and fly ash specific surface area are key material properties and proportioning parameters.
[0040] S33. Based on the global contribution ranking, generate a chart showing the variation of the contribution of the input features to the performance prediction value under different values; Specifically, a feature dependency contribution map is plotted with the actual value of a single input feature on the x-axis and the corresponding SHAP value on the y-axis. Each point in the map represents a sample, with the x-axis representing the SHAP value of the feature (i.e., its contribution) and the y-axis representing the actual value of the feature. This feature dependency contribution map can demonstrate the nonlinear relationship and interaction between a single feature and the model prediction. For example, it can reveal the law that the marginal contribution of the specific surface area of the admixture to the strength improvement begins to decrease after exceeding a threshold.
[0041] This embodiment enhances the transparency and interpretability of the TabPFN model by performing SHAP analysis on the input features, clarifying the extent to which adjusting specific ratio parameters will affect the target performance and how this effect interacts with other parameters.
[0042] Based on the above embodiments, this method further includes: S4. Fit the network model with the tabular prior data as the performance prediction model, and establish a multi-objective optimization model by combining material cost and material density, and set constraints for the multi-objective optimization model according to the contribution relationship; In this embodiment, by embedding the TabPFN model into a multi-objective decision framework, the constraint conditions are set by using the contribution relationship between input features and performance prediction values, thereby automatically searching for a set of proportion schemes that can simultaneously satisfy multiple objectives within the proportion design space.
[0043] Specifically, step S4 includes: S41. Define the objective function to maximize mechanical properties, minimize material density, and minimize material cost; Specifically, with the goal of maximizing unconfined compressive strength, a trained TabPFN model is used. Predict compressive strength; among which, Indicates compressive strength. Given a fixed vector of intrinsic material properties, the chemical composition and specific surface area of cement and admixtures. This refers to the ratio vector to be optimized, such as the dosage of each component per cubic meter. The maintenance period is usually 28 days.
[0044] Specifically, with the goal of minimizing wet density, an objective function is established. , wherein Indicates wet density; Specifically, with the goal of minimizing material costs, an objective function is established. ,in and The first The unit mass cost of various components, such as cement, admixtures, water, additives, etc., and their role in the mix proportions. Dosage in the formula.
[0045] Combining the above three objectives, the final objective function is: .
[0046] In the formula, This represents the final objective function. For compressive strength, Wet density, For material costs.
[0047] S42. Based on the degree of contribution of each input feature in the contribution relationship to the performance prediction value, apply constraints to the mix proportion to be optimized, including component dosage range constraints and material ratio constraints; In this embodiment, by ranking the global contribution of input features to the performance prediction values, the key features that have the greatest impact on the performance of foamed lightweight soil, such as cement content, water-cement ratio, and fly ash specific surface area, can be identified. Based on the nonlinear laws revealed by the feature dependency contribution plot, the reasonable value range for each key feature can be determined. For example, the feature dependency contribution plot shows that the water-cement ratio has a significant positive contribution to compressive strength in the range of 0.4 to 0.5, and the contribution tends to level off after exceeding 0.5. Therefore, the upper limit of the search for the water-cement ratio can be set to 0.5, thereby ensuring that the optimization process is carried out within the performance stable region.
[0048] Specifically, the constraints on the range of component dosages are as follows: ; In the formula, Indicates the first Dosage of each component Indicates the lower limit. Indicates the upper limit.
[0049] The proportional relationship between the materials is constrained as follows: ; In the formula, It is the lower limit of the material ratio. It refers to the component dosage of water. It refers to the component dosage of cement. This refers to the dosage of other auxiliary cementitious materials. It is the upper limit of the material ratio.
[0050] S43. The objective function and the constraints are combined to construct the multi-objective optimization model.
[0051] Based on the above embodiments, this method further includes: S5. Solve the multi-objective optimization model using a non-dominated sorting genetic algorithm to search for a set of Pareto optimal matching schemes.
[0052] Specifically, step S5 includes: S51. Set the population size, maximum number of generations, crossover probability, and mutation probability. Randomly generate the initial population. Each individual in the initial population represents a candidate foamed lightweight soil mix design, i.e., the mix design vector of the constraints. .
[0053] S52. For the initial population Each individual in Calculate the three corresponding objective function values for each individual based on the objective function. The objective function values form a three-dimensional vector. .
[0054] S53. Perform a non-dominated ranking of all individuals in the current population, and divide the individuals into multiple frontier ranks according to Pareto dominance. , …,in, The first frontier contains the set of optimal solutions in the current population that are not dominated by any other individual; The second frontier contains the solution set dominated only by individuals in F1, and so on.
[0055] S54. Calculate the crowding distance for each individual within the same non-dominated front, and perform selection, crossover, and mutation operations based on the crowding distance and non-dominated ranking to generate a progeny population.
[0056] S55. Merge the offspring population with the current population, re-perform non-dominated sorting and crowding distance calculation on the merged population, and select individuals to form a new generation population based on the crowding distance.
[0057] S56. Repeat steps S53-S55 to gradually evolve the population towards the global Pareto optimal front. The evolutionary process terminates when the preset maximum number of generations is reached. The population at the first non-dominated front... All individuals constitute the Pareto optimal solution set. The Pareto optimal solution set forms a line or a surface in the multidimensional space formed by the objective function values, namely the "Pareto front". Each point on the front represents an optimal trade-off design.
[0058] In this embodiment, each individual in the Pareto optimal solution set represents a foamed lightweight soil mix design that achieves the best balance between compressive strength, wet density, and material cost. Each foamed lightweight soil mix design includes the mass of cement, admixtures, water, foaming agent, and additives per cubic meter of foamed lightweight soil; as well as the predicted compressive strength, wet density, and material cost.
[0059] Example 2: like Figure 3 As shown, this embodiment provides an intelligent test decision-making device for foamed lightweight soil proportioning. The device includes: The data acquisition module is used to acquire experimental data of multi-source foamed lightweight soil and preprocess it to obtain a standardized tabular dataset containing input features and corresponding performance values. The model prediction module is used to input a standardized tabular dataset as context to a pre-trained tabular prior data fitting network model, which learns and predicts the input features to obtain the corresponding performance prediction value. The contribution analysis module is used to calculate the contribution of the input features in the tabular prior data fitting network model to the performance prediction value using the SHAP value analysis method, so as to obtain the contribution relationship between the input features and the performance prediction value. The optimization modeling module is used to fit a network model with the tabular prior data as a performance prediction model, and to establish a multi-objective optimization model by combining material cost and material density, and to set constraints for the multi-objective optimization model according to the contribution relationship. The optimization solution module is used to solve the multi-objective optimization model using a non-dominated sorting genetic algorithm to search for a set of Pareto optimal matching schemes.
[0060] Based on the above embodiments, the data acquisition module includes: The data filtering unit is used to filter out experimental data of foamed lightweight soil from the database, which includes the chemical composition, physical properties, mix proportions of raw materials, as well as the corresponding curing age and performance indicators. The data cleaning unit is used to clean and format the experimental data of the foamed lightweight soil. The data summarization unit is used to summarize the input characteristics of the cleaned experimental data into material intrinsic property parameters, mix design parameters, and process condition parameters, and to use the performance indicators as output targets to form the standardized tabular dataset.
[0061] Based on the above embodiments, the model prediction module includes: A data partitioning unit is used to randomly divide the standardized tabular dataset into a training subset and a test subset according to a preset ratio; The prediction execution unit is used to input the training subset as context into the pre-trained tabular prior data fitting network model, and the tabular prior data fitting network model predicts the test samples in the test subset based on the correspondence between the input features of the training subset and the performance indicators, and outputs the performance prediction value corresponding to the test sample. The accuracy calculation unit is used to calculate the prediction accuracy of the tabular prior data fitting network model on the test subset based on the performance prediction value; The performance verification unit is used to verify the predictive performance of the network model fitted to the tabular prior data based on the predicted accuracy using a preset regression evaluation index.
[0062] Based on the above embodiments, the prediction execution unit includes: A sequence encoding subunit is used to encode the training samples in the training subset and the test samples in the test subset into an input sequence. The weight calculation subunit is used to calculate the association weight between any two elements in the input sequence through a bidirectional attention mechanism. The association weight includes the association weight between the test sample and each training sample in the training subset. The weighted aggregation subunit is used to perform weighted aggregation on the relationship between the input features and performance of each training sample in the training subset according to the association weight, so as to obtain the performance prediction value corresponding to the test sample.
[0063] Based on the above embodiments, the contribution analysis module includes: A contribution calculation unit is used to calculate the contribution value of each input feature to the performance prediction value of each sample; The aggregation statistics unit is used to aggregate the contribution values of each input feature for all samples and generate a global ranking of the contribution of the input features to the performance prediction values. The chart generation unit is used to generate a chart showing the variation of the contribution values of the input features to the performance prediction values under different values, based on the global contribution ranking.
[0064] Based on the above embodiments, the optimization modeling module includes: The objective definition unit is used to define objective functions to maximize mechanical properties, minimize material density, and minimize material cost. The constraint setting unit is used to apply constraints to the mix proportion to be optimized based on the degree of contribution of each input feature in the contribution relationship to the performance prediction value. The constraints include component dosage range constraints and material ratio relationship constraints. The model building unit is used to construct the multi-objective optimization model by combining the objective function and the constraints.
[0065] It should be noted that the specific manner in which each module performs its operation in the apparatus described in the above embodiments has been described in detail in the embodiments of the method, and will not be elaborated here.
[0066] Example 3: Corresponding to the above method embodiments, this embodiment also provides an intelligent test decision device for foamed lightweight soil proportioning. The intelligent test decision device for foamed lightweight soil proportioning described below and the intelligent test decision method for foamed lightweight soil proportioning described above can be referred to each other.
[0067] Figure 4 This is a block diagram illustrating an intelligent experimental decision-making device 800 for foamed lightweight soil proportioning, according to an exemplary embodiment. Figure 4 As shown, the intelligent test decision-making device 800 for foamed lightweight soil proportioning may include: a processor 801 and a memory 802. The intelligent test decision-making device 800 for foamed lightweight soil proportioning may also include one or more of the following: a multimedia component 803, an I / O interface 804, and a communication component 805.
[0068] The processor 801 controls the overall operation of the intelligent test decision-making device 800 for foamed lightweight soil proportioning to complete all or part of the steps in the aforementioned intelligent test decision-making method for foamed lightweight soil proportioning. The memory 802 stores various types of data to support the operation of the intelligent test decision-making device 800 for foamed lightweight soil proportioning. This data may include, for example, instructions for any application or method operating on the intelligent test decision-making device 800 for foamed lightweight soil proportioning, as well as application-related data such as contact data, sent and received messages, images, audio, video, etc. The memory 802 can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in the memory 802 or transmitted via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. I / O interface 804 provides an interface between processor 801 and other interface modules, such as a keyboard, mouse, and buttons. These buttons can be virtual or physical. Communication component 805 is used for wired or wireless communication between the foam lightweight soil proportioning intelligent test decision device 800 and other devices. Wireless communication includes Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination thereof. Therefore, the corresponding communication component 805 may include a Wi-Fi module, a Bluetooth module, and an NFC module.
[0069] In an exemplary embodiment, the foam lightweight soil mix proportioning intelligent test decision device 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to execute the above-described foam lightweight soil mix proportioning intelligent test decision method.
[0070] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, which, when executed by a processor, implement the steps of the above-described intelligent test decision method for foamed lightweight soil proportioning. For example, the computer-readable storage medium may be the memory 802 including the program instructions, which may be executed by the processor 801 of the intelligent test decision device 800 for foamed lightweight soil proportioning to complete the above-described intelligent test decision method for foamed lightweight soil proportioning.
[0071] Example 4: Corresponding to the above method embodiments, this embodiment also provides a readable storage medium. The readable storage medium described below and the intelligent test decision method for foamed lightweight soil proportioning described above can be referred to each other.
[0072] A readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the intelligent test decision-making method for foamed lightweight soil proportioning described in the above method embodiments.
[0073] Specifically, the readable storage medium can be a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or any other readable storage medium capable of storing program code.
[0074] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
[0075] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A smart experimental decision-making method for the proportioning of foamed lightweight soil, characterized in that, include: Experimental data of multi-source foamed lightweight soil were obtained and preprocessed to obtain a standardized tabular dataset containing input features and corresponding performance values; A standardized tabular dataset is used as context input to a pre-trained tabular prior data fitting network model. The tabular prior data fitting network model learns and predicts the input features to obtain the corresponding performance prediction value. The SHAP value analysis method is used to calculate the contribution of the input features to the performance prediction value in the tabular prior data fitting network model, thereby obtaining the contribution relationship between the input features and the performance prediction value. The network model fitted with the aforementioned tabular prior data is used as the performance prediction model, and a multi-objective optimization model is established by combining material cost and material density. Constraints are set for the multi-objective optimization model based on the aforementioned contribution relationship. The multi-objective optimization model is solved using a non-dominated sorting genetic algorithm to search for a set of Pareto optimal matching schemes.
2. The intelligent experimental decision-making method for foamed lightweight soil proportioning according to claim 1, characterized in that, Experimental data of multi-source foamed lightweight soil were acquired and preprocessed to obtain a standardized tabular dataset containing input features and corresponding performance values, including: Experimental data on foamed lightweight soil, including the chemical composition, physical properties, mixing ratio, and corresponding curing age and performance indicators of raw materials, were selected from the database. The experimental data of the foamed lightweight soil were cleaned and formatted. The input features of the cleaned experimental data are categorized into intrinsic material properties, mix design parameters, and process conditions parameters. The performance indicators are then used as output targets to form the standardized tabular dataset.
3. The intelligent experimental decision-making method for foamed lightweight soil mix proportioning according to claim 1, characterized in that, A standardized tabular dataset is used as context input to a pre-trained tabular prior data fitting network model. The tabular prior data fitting network model learns and predicts the input features to obtain the corresponding performance prediction values, including: The standardized tabular dataset is randomly divided into a training subset and a test subset according to a preset ratio; The training subset is used as context input to the pre-trained tabular prior data fitting network model. The tabular prior data fitting network model predicts the test samples in the test subset based on the correspondence between the input features of the training subset and the performance indicators, and outputs the performance prediction value corresponding to the test sample. The prediction accuracy of the tabular prior data fitting network model on the test subset is calculated based on the performance prediction values. Using a preset regression evaluation index, the predictive performance of the network model fitted with the tabular prior data is verified based on the predicted accuracy.
4. The intelligent experimental decision-making method for foamed lightweight soil mix proportioning according to claim 3, characterized in that, The network model fits the correspondence between the input features and performance metrics of the training subset using the tabular prior data, predicts the performance of the test samples in the test subset, and outputs the predicted performance value of the test samples, including: The training samples in the training subset and the test samples in the test subset are jointly encoded into an input sequence; The association weight between any two elements in the input sequence is calculated using a bidirectional attention mechanism. The association weight includes the association weight between the test sample and each training sample in the training subset. The relationship between the input features and performance of each training sample in the training subset is weighted and aggregated according to the association weights to obtain the performance prediction value corresponding to the test sample.
5. The intelligent experimental decision-making method for foamed lightweight soil mix proportioning according to claim 1, characterized in that, The process involves fitting a network model to the tabular prior data as a performance prediction model, establishing a multi-objective optimization model based on material cost and material density, and setting constraints for the multi-objective optimization model according to the contribution relationship, including: The objective function is defined as maximizing mechanical properties, minimizing material density, and minimizing material cost; Based on the degree of contribution of each input feature in the contribution relationship to the performance prediction value, constraints are applied to the mix proportion to be optimized. The constraints include constraints on the range of component dosage and constraints on the proportional relationship between materials. The objective function and the constraints are combined to construct the multi-objective optimization model.
6. A smart experimental decision-making device for foamed lightweight soil proportioning, characterized in that, include: The data acquisition module is used to acquire experimental data of multi-source foamed lightweight soil and preprocess it to obtain a standardized tabular dataset containing input features and corresponding performance values. The model prediction module is used to input a standardized tabular dataset as context to a pre-trained tabular prior data fitting network model, which learns and predicts the input features to obtain the corresponding performance prediction value. The contribution analysis module is used to calculate the contribution of the input features in the tabular prior data fitting network model to the performance prediction value using the SHAP value analysis method, so as to obtain the contribution relationship between the input features and the performance prediction value. The optimization modeling module is used to fit a network model with the tabular prior data as a performance prediction model, and to establish a multi-objective optimization model by combining material cost and material density, and to set constraints for the multi-objective optimization model according to the contribution relationship. The optimization solution module is used to solve the multi-objective optimization model using a non-dominated sorting genetic algorithm to search for a set of Pareto optimal matching schemes.
7. The intelligent test decision-making device for foamed lightweight soil proportioning according to claim 6, characterized in that, The data acquisition module includes: The data filtering unit is used to filter out experimental data of foamed lightweight soil from the database, which includes the chemical composition, physical properties, mix proportions of raw materials, as well as the corresponding curing age and performance indicators. The data cleaning unit is used to clean and format the experimental data of the foamed lightweight soil. The data summarization unit is used to summarize the input characteristics of the cleaned experimental data into material intrinsic property parameters, mix design parameters, and process condition parameters, and to use the performance indicators as output targets to form the standardized tabular dataset.
8. The intelligent test decision-making device for foamed lightweight soil proportioning according to claim 6, characterized in that, The model prediction module includes: A data partitioning unit is used to randomly divide the standardized tabular dataset into a training subset and a test subset according to a preset ratio; The prediction execution unit is used to input the training subset as context into the pre-trained tabular prior data fitting network model, and the tabular prior data fitting network model predicts the test samples in the test subset based on the correspondence between the input features of the training subset and the performance indicators, and outputs the performance prediction value corresponding to the test sample. The accuracy calculation unit is used to calculate the prediction accuracy of the tabular prior data fitting network model on the test subset based on the performance prediction value; The performance verification unit is used to verify the predictive performance of the network model fitted to the tabular prior data based on the predicted accuracy using a preset regression evaluation index.
9. The intelligent test decision-making device for foamed lightweight soil proportioning according to claim 6, characterized in that, The prediction execution unit includes: A sequence encoding subunit is used to encode the training samples in the training subset and the test samples in the test subset into an input sequence. The weight calculation subunit is used to calculate the association weight between any two elements in the input sequence through a bidirectional attention mechanism. The association weight includes the association weight between the test sample and each training sample in the training subset. The weighted aggregation subunit is used to perform weighted aggregation on the relationship between the input features and performance of each training sample in the training subset according to the association weight, so as to obtain the performance prediction value corresponding to the test sample.
10. The intelligent test decision-making device for foamed lightweight soil proportioning according to claim 6, characterized in that, The optimization modeling module includes: The objective definition unit is used to define objective functions to maximize mechanical properties, minimize material density, and minimize material cost. The constraint setting unit is used to apply constraints to the mix proportion to be optimized based on the degree of contribution of each input feature in the contribution relationship to the performance prediction value. The constraints include component dosage range constraints and material ratio relationship constraints. The model building unit is used to construct the multi-objective optimization model by combining the objective function and the constraints.