Method for determining mix proportion of underwater repair material

Abstract

The present application relates to the technical field of patching material mix ratio determination, and particularly relates to an underwater patching material mix ratio determination method. A plurality of initial underwater patching material mix ratios are obtained; initial compressive strength, initial splitting tensile strength and initial fluidity corresponding to each initial underwater patching material mix ratio are determined; a multi-objective fitness function is constructed based on the initial compressive strength, the initial splitting tensile strength and the initial fluidity corresponding to each initial underwater patching material mix ratio; and each initial underwater patching material mix ratio is optimized based on the multi-objective fitness function to obtain a target underwater patching material mix ratio. The traditional trial-and-error test method is replaced, the test workload is greatly reduced, and the mix ratio research and optimization efficiency is improved. The engineering adaptability and application reliability of the target underwater patching material mix ratio are ensured, and the core engineering goals of the underwater grouting repair, defect repair and bearing capacity recovery are achieved.

Description

Technical Field

[0001] This invention relates to the technical field of determining the mix proportion of repair materials, and specifically to a method for determining the mix proportion of underwater repair materials. Background Technology

[0002] Hydraulic structures operate in underwater environments for extended periods. Compared to onshore structures, their surfaces and interiors are more susceptible to erosion from water flow and suspended particles, leading to cracks and defects that severely impact structural safety and durability. Current underwater repair cement-based grouting materials primarily utilize aluminate cement, while research on ordinary silicate cement-based underwater grouting materials remains weak and the systems are incomplete.

[0003] At the mix design level, the performance of underwater remediation grouting materials is influenced by multiple factors, including water-cement ratio, thickening components, water-reducing components, mineral admixtures, and biological components, exhibiting significant nonlinear and strongly coupled characteristics. Traditional empirical iterative or simple statistical analysis methods are insufficient to accurately characterize the coupling relationships among these multiple factors. Extensive and time-consuming trial-and-error experiments are required to approximate the required mix proportions and process windows, resulting in low efficiency in research and optimization.

[0004] In summary, how to accurately and quickly determine the mix proportions of underwater repair materials has become an urgent problem to be solved. Summary of the Invention

[0005] This invention provides a method for determining the mix proportion of underwater repair materials, in order to solve the problem of how to accurately and quickly determine the mix proportion of underwater repair materials.

[0006] In a first aspect, the present invention provides a method for determining the mix proportion of underwater repair materials, the method comprising: Multiple initial underwater repair material mix proportions are obtained; the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to each initial underwater repair material mix proportion are determined; based on the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to each initial underwater repair material mix proportion, a multi-objective fitness function is constructed; based on the multi-objective fitness function, each initial underwater repair material mix proportion is optimized to obtain the target underwater repair material mix proportion.

[0007] In one optional implementation, determining the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to each initial underwater repair material mix proportion includes: for each initial underwater repair material mix proportion, obtaining the initial input features corresponding to the initial underwater repair material mix proportion; the initial input features include water-cement ratio, diatomaceous earth content, HPMC content, water-reducing agent content, yeast extract content, urea concentration, calcium chloride concentration, bacterial concentration, curing age, hydration time, curing method, underwater flow velocity, and applied water pressure; inputting the initial input features into a preset index prediction model; the preset index prediction model evaluates the initial input features... The process involves feature identification to determine the core features of the mix proportion, as well as the process and operating conditions; feature extraction to construct at least one coupled feature; fusion of the coupled feature with the initial input feature to obtain the initial fused feature; attribute information labeling of each fused sub-feature in the initial fused feature; redundancy removal operation to obtain the target fused feature corresponding to the initial fused feature based on the attribute information corresponding to each fused sub-feature; and feature identification of the target fused feature to determine the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to the initial underwater repair material mix proportion.

[0008] In one optional implementation, the coupling features include at least one of the following: stable synergistic features, reaction matching features, mineralization efficiency features, interface enhancement features, and hydration regulation features. Feature extraction is performed on the initial input features to construct at least one coupling feature, including: constructing stable synergistic features based on bacterial concentration and HPMC dosage; constructing reaction matching features based on urea concentration and hydration time; constructing mineralization efficiency features based on bacterial concentration and calcium chloride concentration; constructing operating condition adaptation features based on underwater flow velocity and HPMC dosage; constructing interface enhancement features based on diatomaceous earth dosage and bacterial concentration; and constructing hydration regulation features based on water-reducing agent dosage and water-cement ratio.

[0009] In one optional implementation, based on the attribute information corresponding to each fusion sub-feature, a redundancy removal operation is performed on the initial fusion feature to obtain the target fusion feature corresponding to the initial fusion feature. This includes: combining each fusion sub-feature in pairs according to the attribute information corresponding to each fusion sub-feature to generate multiple feature pairs; calculating the Pearson correlation coefficient and Spearman correlation coefficient of each feature pair respectively; determining feature pairs whose Pearson correlation coefficient and Spearman correlation coefficient are both greater than a preset correlation coefficient threshold as target feature pairs; and removing the first target fusion sub-feature from each target feature pair to determine the first target feature to be removed. The first remaining feature after fusing the sub-features corresponds to the first compressive strength, first splitting tensile strength, and first flowability; the second target fused sub-feature in the target feature pair is removed, and the second remaining feature after removing the second target fused sub-feature is determined to correspond to the second compressive strength, second splitting tensile strength, and second flowability; the first compressive strength, first splitting tensile strength, and first flowability are compared with the second compressive strength, second splitting tensile strength, and second flowability to determine whether the first target fused sub-feature and the second target fused sub-feature in the target feature pair have unique contributions; based on the comparison results, a redundancy removal operation is performed on the target feature pair to obtain the target fused feature.

[0010] In one optional implementation, a multi-objective fitness function is constructed based on the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to each initial underwater repair material mix proportion, including: The initial compressive strength, initial splitting tensile strength, and initial flowability of each initial underwater repair material mix proportion are checked to see if they are all within the corresponding preset index ranges. Each initial underwater repair material mix proportion within the preset index range is identified as a candidate underwater repair material mix proportion. Each candidate underwater repair material mix proportion is evaluated to determine each backup underwater repair material mix proportion. Based on the backup compressive strength, backup splitting tensile strength, and backup flowability corresponding to each backup underwater repair material mix proportion, a multi-objective fitness function is constructed.

[0011] In one optional implementation, the mix proportions of each candidate underwater repair material are evaluated to determine each backup underwater repair material mix proportion. This includes: calculating a baseline value based on the candidate compressive strength, candidate splitting tensile strength, and candidate flowability corresponding to each candidate underwater repair material mix proportion; for each candidate underwater repair material mix proportion, based on the baseline value, calculating the candidate local SHAP value of each initial input feature for the candidate underwater repair material mix proportion; for each initial input feature, calculating the candidate global SHAP value of the initial input feature for all candidate underwater repair material mix proportions based on the candidate local SHAP value of the initial input feature for the candidate underwater repair material mix proportions; based on the candidate global SHAP value, constructing a correlation between feature value changes and performance contribution changes; based on the correlation, checking whether each candidate underwater repair material mix proportion conforms to the global rule verification; and based on the candidate local SHAP value, checking whether each candidate underwater repair material mix proportion conforms to the optimal layout interval verification. Based on the verification results, the candidate underwater repair material mix proportions that conform to both the global rule verification and the optimal layout interval verification are determined as the backup underwater repair material mix proportions.

[0012] In one optional implementation, a multi-objective fitness function is constructed based on the backup compressive strength, backup splitting tensile strength, and backup flowability corresponding to each backup underwater repair material mix proportion. This includes: obtaining the application scenarios corresponding to each backup underwater repair material mix proportion; determining the scenario weights corresponding to each backup underwater repair material mix proportion based on the application scenarios; constructing a dynamic operating condition constraint function based on the underwater flow velocity and applied water pressure corresponding to each backup underwater repair material mix proportion; obtaining the backup global SHAP values ​​of each initial input feature in each backup underwater repair material mix proportion for all backup underwater repair material mix proportions; constructing a feature contribution constraint function based on the backup global SHAP values; setting a performance hard constraint function based on the engineering specification requirements and the backup global SHAP values; and constructing a basic objective function based on the scenario weights corresponding to each backup underwater repair material mix proportion. Based on the dynamic operating condition constraint function, the feature contribution constraint function, and the performance hard constraint function, a constraint penalty function is constructed; based on the basic objective function and the constraint penalty function, a multi-objective fitness function is constructed.

[0013] In one optional implementation, the initial underwater repair material mix ratios are optimized based on a multi-objective fitness function to obtain the target underwater repair material mix ratios. This includes: determining each spare underwater repair material mix ratio as an initial individual in the initial population; determining the target parent population from the initial population; and performing crossover and mutation operations on the target parent population to obtain updated individuals. Based on the multi-objective fitness function, the update fitness corresponding to each update individual is calculated; according to the update fitness corresponding to each update individual, crossover and mutation operations are performed on each update individual until the preset stopping condition is met, and the candidate individual set is output; from the candidate individual set, the target underwater repair material mix ratio is determined.

[0014] In one optional implementation, determining the target parent population from the initial population includes: traversing each initial individual in the initial population to determine the dominance relationship between each initial individual; identifying first-level initial individuals that are not dominated by any other initial individuals based on the dominance relationship; removing the first-level initial individuals and identifying second-level initial individuals that are not dominated by any other initial individuals in the remaining population; and so on, completing the grading of all initial individuals; for each initial individual of the same level, sorting them by each performance indicator from smallest to largest; calculating the crowding distance between each sorted initial individual and its adjacent initial individuals for each indicator; calculating the crowding degree corresponding to each initial individual based on the crowding distance; removing initial individuals with a crowding degree less than a first preset crowding degree threshold from each initial individual in each level; marking initial individuals with a crowding degree greater than a second preset crowding degree threshold as rare and high-quality individuals; and selecting the first-level initial individuals... Each rare and high-quality individual is retained for the next iteration. For the first remaining initial individuals (excluding rare and high-quality individuals), pair them into multiple pairs. For each pair, calculate the initial fitness of each remaining initial individual based on a multi-objective fitness function. Compare the initial fitness values ​​of the two remaining initial individuals in the pair, and mark the one with the higher initial fitness value as the tournament winner. If the initial fitness values ​​of the two remaining initial individuals are equal, mark the one with the higher crowding as the tournament winner. Repeat this process a preset number of times to obtain each tournament winner. Combine each rare and high-quality individual with the tournament winners to form the initial parent population. If the variance of the target feature in the initial parent population is less than a preset variance threshold, find a preset number of diverse individuals from the second remaining initial individuals (excluding rare and high-quality individuals and tournament winners) whose target feature values ​​differ from a preset difference threshold. Add these diverse individuals to the initial parent population to obtain the target parent population.

[0015] In one optional implementation, determining the target underwater repair material mix ratio from the candidate individual set includes: detecting whether each candidate individual in the candidate individual set meets preset biocompatibility core indicators and preset hydration synergy indicators; deleting each candidate individual that does not meet the preset biocompatibility core indicators and / or preset hydration synergy indicators to obtain the remaining candidate individuals; calculating the core performance score, biocompatibility score, and constraint satisfaction score corresponding to each remaining candidate individual; obtaining the comprehensive score corresponding to each remaining candidate individual based on the core performance score, biocompatibility score, and constraint satisfaction score; determining the remaining candidate individual with the highest comprehensive score as the target individual; and determining the target underwater repair material mix ratio based on the target individual.

[0016] The underwater repair material mix proportion determination method provided in this application obtains multiple initial underwater repair material mix proportions, providing a sufficient sample base for subsequent optimization. This avoids the optimization results from getting stuck in local optima due to a single initial sample, ensuring the breadth of exploration for the target underwater repair material mix proportion. It determines the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to each initial mix proportion, focusing on the most critical strength and flowability dimensions of underwater grouting materials, eliminating irrelevant indicators, and ensuring that the subsequent optimization direction is highly consistent with the actual needs of underwater engineering. Simultaneously, through a unified performance testing standard, it enables the performance of each initial mix proportion to be comparable and screenable, laying a quantitative evaluation foundation for optimization. A multi-objective fitness function is constructed based on core performance indicators. The multi-dimensional engineering requirements of underwater grouting materials, such as anti-dispersion, flowability, and strength, are transformed into a calculable and optimizable mathematical function, realizing a quantitative evaluation of the comprehensive performance of the mix proportion. This makes the evaluation results more aligned with the multi-objective engineering requirements of underwater repair, while providing a unified target guidance for subsequent algorithm optimization, improving the scientific rigor and relevance of the optimization process. The target underwater repair material mix ratio is obtained by optimizing the multi-objective fitness function, replacing the traditional trial-and-error experimental method, which greatly reduces the workload of experiments and improves the efficiency of mix ratio research and optimization. Through the constraint and weight design of the function, the optimized target underwater repair material mix ratio can take into account the core requirements of anti-dispersion, fluidity, strength and durability under underwater disturbance conditions. The optimization process can accurately match the underwater service characteristics of hydraulic structures, ensuring the engineering adaptability and application reliability of the target underwater repair material mix ratio, and achieving the core engineering goals of underwater grouting repair: seepage prevention, defect repair and load-bearing capacity restoration. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the first process of determining the mix proportion of underwater repair materials according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the second process for determining the mix proportion of underwater repair materials according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the Spearman correlation coefficient according to an embodiment of the present invention. Detailed Implementation

[0019] 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 embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] According to an embodiment of the present invention, an embodiment of a method for determining the mix proportion of underwater repair materials is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0021] This embodiment provides a method for determining the mix proportion of underwater repair materials, which can be used in electronic devices. Figure 1 This is a flowchart of a method for determining the mix proportion of underwater repair materials according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain various initial underwater repair material mix ratios.

[0022] Specifically, the electronic device can receive multiple initial underwater repair material mix ratios input by the user, or it can receive multiple initial underwater repair material mix ratios sent by other devices. This application does not specify the particular method for obtaining multiple initial underwater repair material mix ratios.

[0023] Step S102: Determine the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to each initial underwater repair material mix ratio.

[0024] Specifically, the electronic device can input the mix proportions of each initial underwater repair material into a preset index prediction model to determine the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to each initial underwater repair material mix proportion.

[0025] This step will be explained in detail below.

[0026] Step S103: Based on the initial compressive strength, initial splitting tensile strength and initial flowability corresponding to each initial underwater repair material mix ratio, construct a multi-objective fitness function.

[0027] Specifically, the electronic device can construct a multi-objective fitness function based on the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to each initial underwater repair material mix ratio.

[0028] This step will be explained in detail below.

[0029] Step S104: Based on the multi-objective fitness function, optimize the mix proportions of each initial underwater repair material to obtain the target underwater repair material mix proportion.

[0030] Specifically, the electronic device can solve the multi-objective fitness function, optimize the mix proportions of each initial underwater repair material, and obtain the target underwater repair material mix proportion.

[0031] This step will be explained in detail below.

[0032] The underwater repair material mix design method provided in this embodiment obtains multiple initial underwater repair material mix designs, providing a sufficient sample base for subsequent optimization. This avoids the optimization results from getting stuck in local optima due to a single initial sample, ensuring the breadth of exploration for the target underwater repair material mix design. The method determines the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to each initial mix design, focusing on the most critical strength and flowability dimensions of underwater grouting materials. Irrelevant indicators are eliminated, ensuring that the subsequent optimization direction is highly aligned with the actual needs of underwater engineering. Simultaneously, by using a unified performance testing standard, the performance of each initial mix design can be compared and screened, laying a quantitative evaluation foundation for optimization. A multi-objective fitness function is constructed based on core performance indicators. The multi-dimensional engineering requirements of underwater grouting materials, such as resistance to dispersion, flowability, and strength, are transformed into a calculable and optimizable mathematical function, achieving a quantitative evaluation of the comprehensive performance of the mix design. This makes the evaluation results more aligned with the multi-objective engineering requirements of underwater repair, while providing a unified target guidance for subsequent algorithm optimization, improving the scientific rigor and relevance of the optimization process. The target underwater repair material mix ratio is obtained by optimizing the multi-objective fitness function, replacing the traditional trial-and-error experimental method, which greatly reduces the workload of experiments and improves the efficiency of mix ratio research and optimization. Through the constraint and weight design of the function, the optimized target underwater repair material mix ratio can take into account the core requirements of anti-dispersion, fluidity, strength and durability under underwater disturbance conditions. The optimization process can accurately match the underwater service characteristics of hydraulic structures, ensuring the engineering adaptability and application reliability of the target underwater repair material mix ratio, and achieving the core engineering goals of underwater grouting repair: seepage prevention, defect repair and load-bearing capacity restoration.

[0033] This embodiment provides a method for determining the mix proportion of underwater repair materials, which can be used in electronic devices. Figure 2 This is a flowchart of a method for determining the mix proportion of underwater repair materials according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Obtain various initial underwater repair material mix ratios.

[0034] Please refer to the above description of step S101 for details on this step, which will not be repeated here.

[0035] Step S202: Determine the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to each initial underwater repair material mix ratio.

[0036] Specifically, step S202 above may include the following steps: Step S2021: For each initial underwater repair material mix ratio, obtain the initial input features corresponding to the initial underwater repair material mix ratio.

[0037] The initial input characteristics include water-cement ratio, diatomaceous earth content, HPMC content, water-reducing agent content, yeast extract content, urea concentration, calcium chloride concentration, bacterial concentration, curing age, hydration time, curing method, underwater flow velocity, and applied water pressure.

[0038] Specifically, for each initial underwater repair material mix ratio, the electronic device can receive the initial input features corresponding to the initial underwater repair material mix ratio input by the user. For example, Table 1 below is a schematic table of initial input features.

[0039] Table 1. Schematic diagram of initial input features

[0040] Electronic equipment should remove abnormal data that exceeds the above range (e.g., bacterial concentration of 9.2 Cell / mL, which exceeds 4~8 Cell / mL and needs to be checked and corrected) to ensure logical consistency between features (e.g., when the maintenance method is underwater, the underwater flow rate and the applied water pressure must be non-zero).

[0041] Optionally, if a certain feature data is missing, the electronic device can fill it with the median of the feature data of the same type of formulation (e.g., if the HPMC content of a certain formulation is missing, fill it with the median HPMC content of the other formulations with the same water-binder ratio, which is 0.6%).

[0042] Finally, the electronic device can organize the verified feature data into a standardized initial input feature dataset by combining the matching ratio number, feature name, value, and data source, which supports subsequent model calls and feature processing.

[0043] Step S2022: Input the initial input features into the preset index prediction model.

[0044] Specifically, electronic devices can import the standardized initial input feature dataset into a pre-defined indicator prediction model (such as XGBoost or Random Forest) to provide model support for feature classification and extraction.

[0045] Step S2023: The preset index prediction model performs feature recognition on the initial input features to determine the core features of the mix proportion and the process and operating conditions.

[0046] Specifically, the core characteristics of the mix design (which directly affect the synergy of material components and microstructure) include: water-cement ratio (x1), diatomaceous earth content (x2), HPMC content (x3), water-reducing agent content (x4), yeast extract content (x5), urea concentration (x6), calcium chloride concentration (x7), and bacterial concentration (x8). These core characteristics directly determine the cement hydration process, microbial mineralization efficiency, and paste microstructure, and are the core factors affecting performance.

[0047] Process and operating conditions (affecting material forming quality and performance development) include: curing age (x9), hydration time (x 10 ), maintenance methods (x) 11 ), underwater flow velocity (x) 12 ), Actual water pressure (x) 13 Process and operating conditions indirectly regulate the final properties of materials by affecting slurry setting and hardening, interfacial bonding quality, and underwater stability.

[0048] The preset index prediction model can output feature classification labels for each set of initial underwater repair material mix ratios, clarifying the category to which each feature belongs, and providing a basis for targeted construction of coupled features.

[0049] Step S2024: Extract features from the initial input features and construct at least one coupled feature.

[0050] Specifically, the coupling features include at least one of the following: stable synergistic features, reaction matching features, mineralization efficiency features, interface enhancement features, and hydration regulation features; step S2024 above may include the following steps: Step a1: Based on bacterial concentration and HPMC doping amount, construct stable synergistic features.

[0051] Specifically, the thickening effect of HPMC can reduce the loss of underwater bacteria, and the synergy between bacterial concentration and HPMC dosage directly affects the stability of biomineralization. Electronic devices can construct stable synergistic features by directly multiplying the features, using the formula: Stable Synergistic Feature = Bacterial Concentration (x8) × HPMC Dosage (x3).

[0052] Step a2: Based on urea concentration and hydration time, construct reaction matching features.

[0053] Specifically, urea concentration determines the nutrient release rate, and hydration time determines the cement paste setting process; the degree of matching between the two affects the synergy between the mineralization and hydration reactions. Electronic devices can construct reaction-matching features by directly multiplying the features. The formula is: Reaction-matching feature = Urea concentration (x6) × Hydration time (x... 10 ).

[0054] Step a3: Based on bacterial concentration and calcium chloride concentration, construct mineralization efficiency features.

[0055] Specifically, bacterial activity (bacterial concentration) and calcium source supply (calcium chloride concentration) jointly determine the amount of calcium carbonate precipitate formed, directly affecting the material's strength and density. For electronic devices, mineralization efficiency features can be constructed by directly multiplying these features. The formula is: Mineralization efficiency feature = Bacterial concentration (x8) × Calcium chloride concentration (x7).

[0056] Step a4: Based on the underwater flow velocity and HPMC dosage, construct the working condition adaptation features.

[0057] Specifically, at high flow rates, the HPMC dosage needs to be increased to ensure resistance to dispersion; the coupling of these two factors reflects the degree of adaptation of the operating conditions to the material's stability. Therefore, electronic devices can construct operating condition adaptation features by multiplying the underwater flow rate and HPMC dosage. The formula is: Operating condition adaptation feature = Underwater flow rate (x...) 12 ) × HPMC doping (x3).

[0058] Step a5: Based on the diatomite content and bacterial concentration, construct interface reinforcement features.

[0059] Specifically, diatomaceous earth, as a microbial carrier, synergistically affects the cell immobilization efficiency and the distribution of mineralized products at the interface through its dosage and bacterial concentration, thereby enhancing bonding strength. Therefore, electronic devices can construct interface-strengthening features by multiplying the diatomaceous earth dosage and bacterial concentration. The formula is: Interface-strengthening feature = Diatomaceous earth dosage (x2) × Bacterial concentration (x8).

[0060] Step a6: Based on the water-reducing agent dosage and water-cement ratio, construct hydration regulation features.

[0061] Specifically, the dispersing effect of water-reducing agents and the water-cement ratio together determine the degree of hydration and pore structure of cement paste, and their coupling can reflect the synergistic regulation effect on strength development. Therefore, electronic devices can multiply the water-reducing agent dosage and the water-cement ratio to construct hydration regulation characteristics. The formula is: Hydration regulation characteristic = Water-reducing agent dosage (x4) × Water-cement ratio (x1).

[0062] Step S2025: The coupled features are fused with the initial input features to obtain the initial fused features.

[0063] Specifically, the electronic device can concatenate 6 types of coupled features with 13 initial input features to form an initial fused feature containing 19 features.

[0064] Step S2026: Label the attribute information of each fusion sub-feature in the initial fusion feature.

[0065] Specifically, electronic devices can annotate the attribute information of each fusion sub-feature in the initial fusion features. These include original features, stable cooperative features, and response matching features. The original features represent the initial input features, which facilitates subsequent correlation analysis and screening.

[0066] Step S2027: Based on the attribute information corresponding to each fusion sub-feature, perform a redundancy removal operation on the initial fusion feature to obtain the target fusion feature corresponding to the initial fusion feature.

[0067] Specifically, step S2027 above may include the following steps: Step b1: Based on the attribute information corresponding to each fusion sub-feature, combine each fusion sub-feature in pairs to generate multiple feature pairs.

[0068] Specifically, electronic devices can use a pairing principle based on three categories: "original feature-original feature", "original feature-coupled feature" and "coupled feature-coupled feature". They can use a non-repeating pairing method to combine all the fused sub-features in the initial fused features in pairs to generate feature pairs.

[0069] For example, let the initial fusion features be {F1, F2, ..., F...} n} (n=13~14), generate pairs {(F1,F2),(F1,F3),...,(F1,F2),..., ... n ),(F2,F3),...,(F n-1 ,F n )}.

[0070] Step b2: Calculate the Pearson correlation coefficient and Spearman correlation coefficient for each feature pair.

[0071] Specifically, the electronic device can extract the Fa feature values ​​{x1,x2,...,xb} of all m initial underwater repair material mix proportions for any feature pair (Fa,Fb). m}、Fb feature values ​​{y1,y2,...,y m}

[0072] Electronic devices can calculate the Pearson correlation coefficient of each feature pair based on the following formula to characterize the degree of linear correlation between feature pairs.

[0073] ; in, The mean of the samples corresponding to the Fa feature in the feature pair is... denoted as Fb, where m is the total number of initial underwater repair material mix proportions. Electronic equipment coefficients are retained to two decimal places, and statistical significance is verified (P < 0.05 indicates a valid correlation; otherwise, the coefficient is not statistically significant). Where r... p For r ∈ [-1, 1], the closer the absolute value is to 1, the stronger the linear correlation; p >0 indicates a positive correlation, and rp<0 indicates a negative correlation.

[0074] Electronic devices can detect the values ​​of Fa {x1,x2,...,x} in a feature pair. m The value of Fb in the feature pair is {y1, y2, ..., y}. mAssign ranks (i.e., grades) to each value from smallest to largest, and take the average rank for values ​​with the same value, to obtain the rank set {R} of Fa. x1 ,...,R xm The rank set {R} of Fb y1 ,...,R ym For each initial underwater repair material mix ratio, calculate the rank difference di=R. xi -R yi (i=1,2,...,m), and calculate the square of the rank difference d. i 2 .

[0075] Then, the electronic device calculates the square of the rank difference and substitutes it into the Spearman correlation coefficient formula: Similar to the Pearson coefficient, rounded to two decimal places, the significance was verified as P < 0.05. s For values ​​∈ [-1, 1], the closer the absolute value is to 1, the stronger the nonlinear monotonic correlation; there is no requirement for positive or negative correlation with the Pearson coefficient, only the absolute value matters. For example, ... Figure 3 The diagram shown is a schematic representation of the Spearman correlation coefficient calculated based on the above formula, used to reveal the statistical correlation between different parameters.

[0076] Step b3: Feature pairs whose Pearson correlation coefficient and Spearman correlation coefficient are both greater than the preset correlation coefficient threshold are identified as target feature pairs.

[0077] Specifically, the electronic device can compare the Pearson correlation coefficients of each feature pair with a preset Pearson correlation coefficient threshold, and compare the Spearman correlation coefficients with a preset Spearman correlation coefficient threshold. Then, feature pairs whose Pearson correlation coefficients are greater than the preset Pearson correlation coefficient threshold and whose Spearman correlation coefficients are greater than the preset Spearman correlation coefficient threshold are identified as target feature pairs.

[0078] Step b4: For each target feature pair, remove the first target fusion sub-feature from the target feature pair, and determine the first compressive strength, first splitting tensile strength, and first flowability corresponding to the first remaining feature after removing the first target fusion sub-feature.

[0079] Specifically, for each target feature pair, for each target feature pair Tk(F Tk-1 ,F Tk-2 The electronic device can remove the first target fusion feature F from the preprocessed fusion feature. Tk-1 Retain all other features and construct the first residual feature set S. k-1 That is: Sk-1 =Fusion Sub-feature - First Target Fusion Sub-feature F Tk-1 .

[0080] Then, the electronic device uses the first residual feature set S k-1 Using the input features, and labeling three measured performance metrics (compressive strength, splitting tensile strength, and flowability) from the samples, an XGBoost baseline model is trained. After training, this XGBoost baseline model is used to train the first residual feature set S. k-1 Performance prediction was performed to obtain the first compressive strength f1. 1 First splitting tensile strength f2 1 First flowability f3 1 Simultaneously, the prediction accuracy R1 of the three performance parameters of this XGBoost benchmark model was recorded. 2 ′、R2 2 ′、R3 2 (The closer to 1, the higher the precision).

[0081] The electronic device establishes a contribution assessment ledger for each target feature pair, recording the first remaining feature set, three performance prediction values, and model prediction accuracy to ensure a one-to-one correspondence between the data and the target feature pair.

[0082] Step b5: Remove the second target fusion sub-feature from the target feature pair, and determine the second compressive strength, second splitting tensile strength, and second flowability corresponding to the second remaining feature after removing the second target fusion sub-feature.

[0083] Specifically, symmetrical to step b4 above, for each target feature pair Tk(F Tk-1 ,F Tk-2 The electronic device removes the second target fusion feature F from the preprocessed fusion feature. Tk-2 Retain all other features and construct the second residual feature set S. k-2 S k-2 =Fusion Sub-feature - Second Target Fusion Sub-feature F Tk-2 .

[0084] The operation process is exactly the same as step b4: with S k-2 The input features are fed into the trained XGBoost model to predict the second compressive strength f1. 2 Second splitting tensile strength f2 2 Second fluidity f3 2 Record the model prediction accuracy , , .

[0085] The electronic device supplements the contribution evaluation ledger with the second remaining feature set, three performance prediction values, and model prediction accuracy, completing the dual-set data collection for each pair of target feature pairs.

[0086] Step b6: The first compressive strength, the first splitting tensile strength, and the first flowability are compared with the second compressive strength, the second splitting tensile strength, and the second flowability to determine whether the first target fusion sub-feature and the second target fusion sub-feature in the target feature pair have unique contributions.

[0087] Specifically, for each performance metric, the electronic device calculates the difference between the model accuracy after removing a single feature and the base accuracy, i.e., the change in accuracy after removing the first target fusion sub-feature: ΔR a 2 =R a 2 -R a 2 ′, a=1,2,3, corresponding to three performance parameters, R a 2 Based on the accuracy, R a 2 ′ represents the model accuracy after removing the first target fusion sub-features; the change in accuracy after removing the second feature is: , The accuracy of the model after removing the sub-features for the second objective. ΔR 2 The larger the value, the more significant the decrease in model accuracy after removing the feature, and the more prominent the contribution of the feature to performance prediction.

[0088] For each performance indicator, the electronic device can compare the change in precision after removing the first target fusion sub-feature with a preset change threshold. If the change in precision after removing the first target fusion sub-feature for at least one performance indicator is greater than or equal to the first preset change threshold, then the first target fusion sub-feature is determined to have a unique contribution and is retained. If the change in precision after removing the first target fusion sub-feature for all performance indicators is less than the second preset change threshold, then the first target fusion sub-feature has no unique contribution. The first preset change threshold is greater than the second preset change threshold.

[0089] The electronic device adds the contribution assessment results (with / without contribution, features to be removed) of each pair of target features to the contribution assessment ledger, clarifying the redundancy attributes of each target fusion sub-feature.

[0090] Step b7: Based on the comparison results, perform a redundancy removal operation on the target feature pairs to obtain the target fused features.

[0091] Specifically, the electronic device performs a removal operation on the target fusion sub-features that do not contribute uniquely to each target feature pair, thereby obtaining the target fusion features.

[0092] Step S2028: Perform feature recognition on the target fusion features to determine the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to the initial underwater repair material mix ratio.

[0093] Specifically, the preset index prediction model performs feature identification on the target fusion features to determine the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to the initial underwater repair material mix ratio.

[0094] Step S203: Based on the initial compressive strength, initial splitting tensile strength and initial flowability corresponding to each initial underwater repair material mix ratio, construct a multi-objective fitness function.

[0095] Specifically, step S203 above may include the following steps: Step S2031: Check whether the initial compressive strength, initial splitting tensile strength and initial flowability of each initial underwater repair material mix ratio are within the corresponding preset index range.

[0096] Specifically, the electronic device can receive preset index ranges for the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to each initial underwater repair material mix ratio input by the user. For example, the preset index ranges (hard performance thresholds) can be as follows: compressive strength: f1≥30MPa (minimum load-bearing requirement); splitting tensile strength: f2≥3MPa (minimum bond crack resistance requirement); flowability: 250mm≤f3≤350mm (balancing pumpability and anti-dispersion).

[0097] The electronic device can compare the initial compressive strength, initial splitting tensile strength, and initial flowability of each initial underwater repair material mix with the corresponding preset index ranges to detect whether the initial compressive strength, initial splitting tensile strength, and initial flowability of each initial underwater repair material mix are all within the corresponding preset index ranges.

[0098] Step S2032: The initial underwater repair material mix proportions that are all within the corresponding preset index ranges in terms of initial compressive strength, initial splitting tensile strength, and initial flowability are determined as candidate underwater repair material mix proportions.

[0099] Specifically, based on the comparison results, the electronic device can determine the initial underwater repair material mix proportions that are within the corresponding preset index ranges in terms of initial compressive strength, initial splitting tensile strength, and initial flowability as candidate underwater repair material mix proportions.

[0100] Step S2033: Evaluate the mix proportions of each candidate underwater repair material to determine the mix proportions of each backup underwater repair material.

[0101] Specifically, step S2033 above may include the following steps: Step c1: Calculate the baseline value based on the candidate compressive strength, candidate splitting tensile strength, and candidate flowability corresponding to the mix proportions of each candidate underwater repair material.

[0102] Specifically, the electronic device can calculate the arithmetic mean of the three candidate performance indicators corresponding to the mix proportions of each candidate underwater repair material, and obtain the benchmark value of the corresponding performance. The core formula is: ; in: This is the benchmark value corresponding to the candidate compressive strength. This is the benchmark value corresponding to the candidate splitting tensile strength. This is the baseline value corresponding to the candidate mobility. Let the candidate compressive strength value be the mix proportion of the k-th candidate underwater repair material. Let the candidate splitting tensile strength value be assigned to the mix proportion of the k-th candidate underwater repair material. Let be the candidate flowability value corresponding to the k-th candidate underwater repair material mix proportion.

[0103] Step c2: For each candidate underwater repair material mix ratio, based on the benchmark value, calculate the candidate local SHAP value of each initial input feature in the candidate underwater repair material mix ratio.

[0104] Specifically, electronic devices can select the appropriate SHAP interpreter based on the type of the preset indicator prediction model. For example, if the preset indicator prediction model is a tree model (such as XGBoost or Random Forest), TreeExplainer (high computational efficiency and high accuracy) is preferred; if the preset indicator prediction model is a non-tree model, KernelExplainer (general-purpose, nsamples set to 100~500 to balance accuracy and speed) is used. The electronic device can take all initial input characteristics of the candidate underwater repair material mix proportions as input and three candidate performance indicators as output, and output the benchmark value. Substitute into the core decomposition formula of SHAP value: ; in, is a predicted value (candidate performance index) of a certain performance of the k-th candidate underwater repair material mix proportion; p is the total number of initial input features; Let be the candidate local SHAP value of the i-th feature for the k-th candidate underwater repair material mix ratio.

[0105] Electronic devices can obtain the marginal contribution of a feature to the current matching ratio by comparing the difference in performance predictions when a certain feature is present and absent, combined with a fair weighting allocation; if This indicates that the feature has a positive contribution to performance (improves performance); if This indicates that the feature has a negative contribution to performance (suppresses performance).

[0106] Finally, the electronic device calculates the three performance indicators separately, obtaining three m values. c ×p local SHAP value matrix (m c = Number of samples of candidate underwater repair material mix proportions, p = number of features), each row corresponds to 1 candidate underwater repair material mix proportion, and each column corresponds to 1 feature for a certain performance local SHAP value, forming a candidate local SHAP value ledger.

[0107] Step c3: For each initial input feature, calculate the candidate global SHAP value of the initial input feature for all candidate underwater repair material mix proportions based on the candidate local SHAP value of the initial input feature for the candidate underwater repair material mix proportions.

[0108] Specifically, for each initial input feature and each performance metric, the electronic device calculates the global average contribution. and global contribution intensity I i Two types of core global SHAP values, together constituting candidate global SHAP values.

[0109] Among them, the global average contribution The average marginal contribution of a quantifiable feature to performance reflects the overall performance improvement / inhibition trend of the feature. The formula is: ;like The feature has a positive contribution to overall performance (average performance improvement); if The feature has a negative contribution to the overall performance (average suppression performance).

[0110] Global contribution intensity The significance of a feature's impact on performance (regardless of whether it is positive or negative) is quantified, reflecting whether the feature is a core feature affecting performance. The formula is: ; The larger the value, the more significant the impact of the feature on performance; sorting by Ii in descending order can directly identify the core dominant features and secondary features of a certain performance.

[0111] Optionally, in addition to the core values, the electronic device can calculate the median, 25th percentile, and 75th percentile of the local SHAP values ​​for each feature (to clarify the range of contribution values), as well as the positive / negative percentage (the percentage of samples with positive / negative SHAP values ​​within different feature value ranges), providing complete data support for subsequent construction of quantitative relationships. Finally, the electronic device generates a candidate global SHAP value analysis report, including the global average contribution, global contribution intensity, distribution statistics, and positive / negative percentage for each feature, clarifying the core features and performance impact trends.

[0112] Step c4: Based on the candidate global SHAP values, construct the correspondence between changes in feature values ​​and changes in performance contribution.

[0113] Specifically, for each core feature (the top 5 to 8 features sorted by global contribution intensity Ii), a linear change correspondence (suitable for features with a linear correlation with performance) and a non-linear change correspondence (suitable for features with a non-linear correlation with performance) are constructed based on the correlation between its local SHAP value and the feature value, covering all core performance indicators.

[0114] For constructing linear relationship correspondences (such as water-cement ratio - compressive strength, water-reducing agent dosage - fluidity): electronic equipment for characteristic value x i With the corresponding local SHAP value Perform linear fitting to obtain the contribution coefficient k. i (This reflects the change in performance contribution value for every unit change in the feature value). Then, the electronic device can derive quantification rules based on the contribution coefficient to determine the performance contribution change corresponding to the smallest unit change in the feature value (e.g., 0.01%, 0.01%). The electronic device can then combine the difference between the feature value and the mean to calculate the minimum performance improvement / suppression required, obtaining the lowest predicted performance value for that feature value.

[0115] For example, the contribution coefficient k1 of the water-cement ratio (x1) - compressive strength (f1) is obtained by linear fitting as follows: ① The negative sign indicates that the SHAP contribution value increases as the water-cement ratio decreases; ② Quantification rule: For every 0.01 decrease in the water-cement ratio, the SHAP contribution value of the compressive strength increases by 0.8 MPa; ③ Performance prediction range: If the water-cement ratio x1 of a candidate underwater repair material is 0.42 (lower than the mean φ), then... x1 =0.5, Δ x1 =-0.08), then the compressive strength should be higher than the benchmark value by Δf1≥80×0.08=6.4MPa, that is, f1≥45+6.4=51.4MPa.

[0116] Construction of nonlinear variation correspondences (e.g., bacterial concentration-splitting tensile strength, HPMC dosage-flowability): Electronic equipment for characteristic value x iWith the corresponding local SHAP value Perform nonlinear fitting (such as polynomial fitting or piecewise fitting) to determine the optimal SHAP interval for the feature (the feature value interval where the local SHAP value has the largest positive contribution and the most concentrated distribution). The electronic device calculates the minimum positive SHAP contribution value within the optimal interval (the minimum performance improvement that the feature can bring within this interval). Determine the minimum predicted performance value that should be achieved when the feature is in the optimal interval.

[0117] Example: Bacterial concentration (x8) - splitting tensile strength (f2) ① The optimal range of SHAP obtained by nonlinear fitting is 6~7 Cell / mL; ② The minimum positive SHAP contribution value in this range is 1.2MPa; ③ Quantification rule: When the bacterial concentration is 6~7 Cell / mL, the splitting tensile strength should be ≥1.2MPa higher than the benchmark value (4.5MPa), that is, f2≥5.7MPa.

[0118] Considering model prediction error and experimental error, a tolerance correction is applied to the lowest predicted performance. For core features (such as bacterial concentration and water-gel ratio), a tolerance coefficient ε = 5% is used, and for secondary features, ε = 10%. The correction formula is as follows: Example: The lowest predicted splitting tensile strength in the optimal bacterial concentration range is 5.7 MPa, after correction. .

[0119] Finally, the electronic device generates a quantification table of the correspondence between features and performance changes, which includes feature name, associated performance, type of change relationship, quantification rules, optimal SHAP range, and lowest predicted value after correction, providing a clear judgment standard for subsequent verification.

[0120] Step c5: Based on the corresponding changes, check whether the mix proportions of each candidate underwater repair material conform to the global rules.

[0121] Specifically, the electronic device performs the following steps for each candidate underwater repair material mix ratio and each core performance indicator: Calculate the difference in feature values: ,in Let the i-th feature value be taken for the mix proportion of the k-th candidate underwater repair material. This is the mean of the characteristic across all candidate underwater repair material mix proportions. Electronic equipment is determined according to the characteristic-performance quantification rule by... Calculate the minimum performance improvement Δf that should be achieved compared to the baseline value for this feature. i-min .

[0122] In linear scenarios, ,in, This is the contribution coefficient. In nonlinear scenarios: if the feature is within the optimal interval, It equals the minimum positive SHAP value of the optimal interval; otherwise, it is 0.

[0123] Electronic devices can calculate the actual performance difference: ,in Let this be the value of the performance index for the mix proportion of the k-th candidate underwater repair material. This is the baseline value for this performance.

[0124] like If the global pattern verification for the feature-performance dimension is passed, then the feature-performance dimension is considered to have passed. Otherwise, it is considered to have a global pattern anomaly (the feature value should bring performance improvement, but the actual performance does not meet the standard, which is a logical contradiction). Among them, a candidate underwater repair material mix ratio must satisfy the global pattern verification of all core feature-performance dimensions to be considered to have passed the overall global pattern verification; if any dimension is abnormal, it is considered to have a global pattern anomaly.

[0125] Example: Water-cement ratio - compressive strength dimension: For candidate underwater repair materials, the water-cement ratio x1 = 0.42, Δx1 = 0.42 - 0.5 = -0.08; quantification rules yield Δf. 1-min =6.4MPa, tolerance ε=5%, then Δf 1-min ×(1-ε)=6.1MPa; Actual compressive strength f1=32MPa, reference value φ 0(f1) =45MPa, Δf 1-actual =32-45=-13MPa. Judgment: -13MPa < 6.1MPa, the global pattern in this dimension is abnormal, and the overall global pattern verification of this mix ratio fails.

[0126] The electronic device marks the global regularity verification results (pass / abnormal) for each candidate underwater repair material mix ratio, and records the abnormality dimension and cause, forming a global regularity verification ledger.

[0127] Step c6: Based on the candidate local SHAP values, check whether the mix proportions of each candidate underwater repair material meet the optimal layout interval verification.

[0128] Specifically, the electronic device can check each candidate underwater repair material mix ratio one by one and determine whether the values ​​of its core characteristics are within the optimal range of SHAP determined in step c4 (such as whether the bacterial concentration is 6~7 Cell / mL and the water-gel ratio is 0.45~0.50).

[0129] The electronic equipment performs local optimal interval verification only on candidate underwater repair material mix proportions that have at least one core feature value in the SHAP optimal interval (mix proportions whose feature values ​​are not in the optimal interval do not need to perform this verification).

[0130] For the selected candidate underwater repair material mix proportions, for each feature that falls within the optimal range, the following steps are performed: extract the corrected lowest predicted performance value corresponding to the optimal range of that feature using SHAP. (The value after tolerance correction in step c4).

[0131] If the relevant performance index corresponding to the mix proportion of the candidate underwater repair material is greater than or equal to If a feature dimension passes the local optimal interval check, then the feature dimension passes the check; otherwise, it is considered an anomaly in the local optimal interval (the feature takes the optimal value, but the performance does not reach the minimum expected level). If a candidate underwater repair material mix ratio has multiple features in the optimal interval, the overall local optimal interval check must pass the check for all optimal interval feature dimensions to pass; if any dimension is abnormal, it is considered an anomaly in the local optimal interval.

[0132] For example, in the bacterial concentration-split tensile strength dimension: the bacterial concentration in the candidate underwater repair material mix ratio x8 = 6.5 Cell / mL, which is within the SHAP optimal range [6,7] Cell / mL; the lowest predicted value of the split tensile strength after correction for this range is... The actual splitting tensile strength of this mix proportion The local optimum interval for this dimension is deemed abnormal, and the overall local optimum interval verification for this mix ratio fails.

[0133] The electronic device marks the local optimal interval verification results (pass / not applicable / abnormal) for each candidate underwater repair material mix ratio, and records the abnormal dimension and cause, forming a local optimal interval verification log.

[0134] Step c7: Based on the verification results, the candidate underwater repair material mix proportions that conform to both the global rule verification and the optimal layout interval verification are determined as the backup underwater repair material mix proportions.

[0135] Specifically, the electronic device integrates the global regularity verification results and the local optimal interval verification results of each candidate underwater repair material mix ratio to form four types of verification results: Type A: Global regularity verification passed + local optimal interval verification passed / not applicable (no features in the optimal interval); Type B: Global regularity verification only abnormal; Type C: Local optimal interval verification only abnormal; Type D: Global regularity verification abnormal + local optimal interval verification abnormal.

[0136] The electronic equipment only selects the Class A candidate underwater repair material mix proportions as backup underwater repair material mix proportions, and directly eliminates the Class B / C / D categories (Class B / C / D categories have logical contradictions in their characteristic contributions, and even if their performance meets the standards, they have no engineering optimization value).

[0137] Step S2034: Based on the backup compressive strength, backup splitting tensile strength and backup flowability corresponding to each backup underwater repair material mix ratio, construct a multi-objective fitness function.

[0138] Specifically, step S2034 above may include the following steps: Step d1: Obtain the application scenarios corresponding to the mix proportions of each spare underwater repair material.

[0139] Specifically, electronic devices can classify application scenarios into three typical scenarios based on the core influencing factors of underwater engineering repair (crack width, leakage degree, and water flow velocity). For example, Scenario 1: Cracks ≥ 2mm + severe leakage, such as deep cracks in dam bodies and leakage cracks in bridge pier foundations; Scenario 2: Cracks 0.5~2mm + medium flow velocity, such as cracks at water pipeline interfaces and shallow cracks in riverbank protection; Scenario 3: Fine pores + low disturbance, such as micropore repair on concrete surfaces and fine cracks in tunnel linings.

[0140] Step d2: Determine the scenario weight corresponding to each spare underwater repair material mix ratio based on the application scenario corresponding to each spare underwater repair material mix ratio.

[0141] Specifically, the electronic device can receive the correspondence between application scenarios and scenario weights input by the user. Then, based on the correspondence between application scenarios and scenario weights, the electronic device can determine the scenario weight corresponding to each backup underwater repair material mix ratio according to the application scenario corresponding to each backup underwater repair material mix ratio.

[0142] Scenario weights are represented as a vector form of "weight value + corresponding backup performance", denoted as (w1, f1) b (w2, f2) b (w3, f3) b And it satisfies the weight constraint: w1 + w2 + w3 = 1 (ensuring the mathematical rationality of priority allocation). Where f1 b As a reserve compressive strength, f2 b For backup splitting tensile strength, f3 b This is for backup flowability.

[0143] For example, the electronic device can use the preset three scenario priority weights to directly match the scenario type of the backup underwater repair material mix ratio: Scenario 1 (cracks ≥ 2mm + severe leakage): w1 = 0.45 (compressive strength), w2 = 0.35 (tensile strength), w3 = 0.20 (flowability); f1 b ≥50MPa, f2 b≥5.5MPa; Scenario 2 (crack 0.5~2mm + medium flow velocity): w1=0.30 (compressive strength), w2=0.25 (tensile strength), w3=0.45 (flowability); f3 b ≥280mm, f1 b ≥37MPa; Scenario 3 (fine pores + low disturbance): w1=0.25 (compressive strength), w2=0.40 (tensile strength), w3=0.35 (flowability); f2 b ≥5.8MPa.

[0144] Step d3 involves constructing a dynamic operating condition constraint function based on the underwater flow velocity and applied water pressure corresponding to the mix proportions of each spare underwater repair material.

[0145] Specifically, the dynamic operating condition constraint function includes a backup flowability constraint term, a backup compressive strength constraint term, and dynamic constraint on characteristic values.

[0146] Among them: the standby flowability constraint term is: f3 b ≥280-30×max(0,v-0.5); logically, when v≤0.5m / s, f3 b ≥280mm; When v>0.5m / s, for every 0.1m / s increase in flow velocity, the minimum required standby flowability is reduced by 3mm to avoid loss at high flow velocities; The alternative compressive strength constraint term is: f1 b ≥35+10×max(0,P-0.3); logically, when P≤0.3MPa, f1 b ≥35MPa; When P>0.3MPa, for every 0.1MPa increase in water pressure, the minimum requirement for standby compressive strength is increased by 1MPa to meet the high pressure bearing requirements.

[0147] Dynamic constraints on feature values ​​include the following: Water-to-binder ratio (x1): 0.45≤x1≤0.55 (adjusted to 0.47≤x1≤0.52 when v>1.0m / s); Diatomaceous earth (x2): 0 ≤ x2 ≤ 10% (3%~7% recommended when biological components are present); HPMC (x3): 0.5 ≤ x3 ≤ 0.7% (adjusted to 0.55 ≤ x3 ≤ 0.7% when 0.5 < v ≤ 1.0 m / s); Water-reducing agent (x4): 0.05 ≤ x4 ≤ 0.15%; Yeast extract (x5): 1 ≤ x5 ≤ 20 g / L (5~15 g / L recommended); Urea (x6): 0.0 6 ≤ x6 ≤ 0.18 mol / L; Calcium chloride (x7): 0.06 ≤ x7 ≤ 0.18 mol / L; Bacterial concentration (x8): 4 ≤ x8 ≤ 8 Cell / mL (adjusted to 6 ≤ x8 ≤ 8 Cell / mL when v > 1.0 m / s or P > 0.3 MPa); Curing period (x9): 0 ≤ x9 ≤ 28 days (≥ 28 days for intensity-priority scheme, ≥ 14 days for flowability-priority scheme); Underwater flow velocity (x 12 ): 0≤x 12 ≤1.5m / s; applied water pressure (x 13 ): 0.1≤x 13 ≤0.5MPa.

[0148] Step d4: Obtain the backup global SHAP value of each initial input feature in each backup underwater repair material mix ratio for all backup underwater repair material mix ratios.

[0149] Specifically, the electronic device can calculate the backup global SHAP value of each initial input feature in each backup underwater repair material mix ratio for all backup underwater repair material mix ratios, based on the calculation step in step c3. This will not be elaborated upon here.

[0150] Step d5: Construct a feature contribution constraint function based on the backup global SHAP value.

[0151] Specifically, the electronic device can construct the following constraint functions based on the backup global SHAP value, targeting the core features (bacterial concentration, water-gel ratio) that contribute significantly: When the bacterial concentration (x8) contributes ≥20% to the splitting tensile strength; The constraint function corresponding to the bacterial concentration (x8) is: f2 b ≥4+0.5×(x8-5) (only applicable to x8≥5Cell / mL). When the bacterial concentration is ≥5Cell / mL, for every 1Cell / mL increase, the minimum required tensile strength increases by 0.5MPa (matching the positive contribution law of bacterial concentration).

[0152] When the absolute value of the contribution of the water-cement ratio (x1) to the fluidity is ≥0.4; The constraint function corresponding to the water-to-glue ratio (x1) is: f3 b≤320-20×(0.55-x1) (only applicable to x1≤0.55). When the water-cement ratio is ≤0.55, for every 0.01 decrease, the maximum flowability limit decreases by 0.2mm (to balance flowability and anti-dispersion properties, and avoid insufficient flowability due to an excessively low water-cement ratio).

[0153] The electronic equipment can check each spare underwater repair material mix ratio one by one to determine whether the core features meet the "triggering conditions". Only the mix ratios that meet the conditions are checked by the corresponding constraint function. The mix ratios that "meet the triggering conditions but fail to meet the constraints" are recorded and the corresponding feature values ​​are adjusted first in subsequent optimization.

[0154] Step d6: Based on the mandatory requirements of the engineering specifications and the backup global SHAP value, set the performance hard constraint function.

[0155] Specifically, electronic devices can construct layered hard constraints targeting three core performance aspects. For example, regarding the backup compressive strength (f1)... b The basic hard constraint is f1. b ≥30MPa; when the contribution of curing age is ≥30%, the strength priority scheme f1 ≥50MPa (corresponding to a curing age ≥28d); for the spare splitting tensile strength (f2) b The basic hard constraint is f2. b ≥3MPa; when the bacterial concentration contributes ≥20%, f2 b ≥5.4MPa (corresponding to bacterial concentration ≥5Cell / mL); for standby flowability (f3) b The basic hard constraint is 250mm≤f3 b When the thickness is ≤350mm and the absolute value of the water-cement ratio contribution is ≥0.4, f3 is calculated in the scenario where v>1.0m / s. b ≤310mm (corresponding to a water-to-glue ratio ≤0.52).

[0156] Electronic devices can first verify the basic hard constraints; if they are not met (e.g., f1), b If the basic constraint is between 28MPa and 30MPa, it is directly judged as a failure of the hard constraint. After the basic constraint is passed, the additional constraints are checked (only when the triggering condition is met). If the additional constraints are not met, it is marked as "needs optimization". For example, a certain alternative underwater repair material mix ratio f2 b =4MPa≥3MPa (basic constraints passed), but bacterial concentration contribution ≥20% and f2 b =4MPa < 5.4MPa (additional constraint failed), the bacterial concentration needs to be optimized to improve tensile strength.

[0157] Step d7: Construct a basic objective function based on the scenario weights corresponding to the mix proportions of each backup underwater repair material.

[0158] Specifically, the electronic device can standardize each backup performance value according to the maximum value of the "effective backup dataset," converting the performance value into a dimensionless value between 0 and 1. The effective backup dataset consists of the mix proportions of all backup underwater repair materials that satisfy the constraints of steps d3-d6.

[0159] Standardized formula: ,in This represents the maximum value of this performance in the valid dataset.

[0160] Weighted reserve compressive strength target: ; Weighted standby splitting tensile strength target: ; Weighted reserve liquidity target: ; Basic objective function: (Value range: 0~1, with a higher value indicating a higher degree of scene adaptability).

[0161] Step d8: Construct a constraint penalty function based on the dynamic operating condition constraint function, the feature contribution constraint function, and the performance hard constraint function.

[0162] Specifically, the constraint penalty function includes dynamic operating condition constraints, characteristic contribution constraints, and hard performance constraints. For example, the dynamic operating condition constraint C1 is: the proportion of performance deviating from the constraint baseline × 0.2 (maximum penalty 0.2), such as the flowability f3. b =260mm < 271mm (constraint baseline), deviation ratio = (271-260) / 271 ≈ 0.04, C1 = 0.04 × 0.2 = 0.008; the characteristic contribution constraint C2 is: the absolute value of the performance deviation constraint value × 0.1 (maximum penalty 0.1), such as f2 b =4MPa<4.5MPa (bacterial concentration constraint value), C2=(4.5-4)×0.1=0.05; the performance hard constraint C3 is: if the basic constraint is not met: penalty 0.5 (directly and significantly reduce the fit); if the additional constraint is not met: penalty 0.1 (maximum penalty 0.1).

[0163] The electronic equipment calculates the total penalty value C for each spare underwater repair material mix ratio. total =C1+C2+C3, and calculate the penalized objective function F. penalty =F base -C total Ensure that the value after penalty is still ≥0 (if <0, it is directly judged as unqualified for optimization).

[0164] Step d9: Construct a multi-objective fitness function based on the basic objective function and the constraint penalty function.

[0165] Specifically, electronic devices can construct a multi-objective fitness function by subtracting the constraint penalty function from the basic objective function.

[0166] For example, a multi-objective fitness function can be shown below: ; Optionally, the multi-objective fitness function can be as follows: ; Among them, To maximize the basic objective function in the effective dataset, ensure that the fitness value is between 0 and 1.

[0167] Step S204: Based on the multi-objective fitness function, optimize the mix proportions of each initial underwater repair material to obtain the target underwater repair material mix proportion.

[0168] Specifically, step S204 above may include the following steps: Step S2041: Determine the mix proportions of each spare underwater repair material as the initial individuals in the initial population.

[0169] Specifically, the electronic equipment can determine the mix proportions of each spare underwater repair material as the initial individuals in the initial population.

[0170] Step S2042: Determine the target parent population from the initial population.

[0171] Specifically, step S2042 above may include the following steps: Step e1: Traverse each initial individual in the initial population and determine the dominance relationship between each initial individual.

[0172] Specifically, for any two initial individuals i and j in the initial population, the electronic device can compare three performance indicators one by one: if f1 b (i)≥f1 b (j) and f2 b (i)≥f2 b (j) and f3 b (i)≥f3 b (j), and at least one of the performance characteristics is strictly greater than (e.g., f1). b (i)>f1 b If i dominates j, then i is considered to have a superior property (e.g., i has better compressibility and j has better flowability). If i and j each have a superior property (e.g., i has better compressibility and j has better flowability), then i is considered to have no dominant relationship.

[0173] The electronic device constructs an N×N dominance relationship matrix (matrix element M) based on the dominance relationships between the initial individuals. ij =1 indicates that i dominates j, M ij=0 indicates no dominance relationship.

[0174] Step e2: Based on the dominance relationship, identify the first-level initial individuals that are not dominated by any other initial individuals.

[0175] Specifically, the electronic device can traverse the dominance matrix to find the initial individuals with "domination count = 0" (i.e., no other individual dominates this initial individual), which are denoted as first-level individuals (the Pareto front with optimal performance). Example: The initial population has 100 initial individuals, of which 15 initial individuals are not dominated by any other individual; these 15 initial individuals are first-level individuals.

[0176] Step e3: Remove the first-level initial individuals and find the second-level initial individuals in the remaining population that are not dominated by any other initial individuals.

[0177] Specifically, the electronic device can remove first-level individuals from the initial population, repeat steps e1 to e2 on the remaining population, and find the initial individuals in the remaining population whose "domination count = 0" are denoted as second-level individuals. The higher the level (first > second > ...), the better the performance of the initial individual.

[0178] Step e4, and so on, completes the ranking of all initial individuals.

[0179] Specifically, the electronic device repeats the operation of "removing the current level → filtering the next level" until all initial individuals are assigned to the corresponding level (e.g., levels one through five). Finally, the electronic device generates an individual number-level-number of times it has been dominated, with smaller level numbers indicating better individual performance.

[0180] Step e5: For each initial individual of the same level, sort them from smallest to largest according to each performance indicator.

[0181] Specifically, for each initial individual within each level, according to f1... b (Compression resistance), f2 b (Tensile strength), f3 b The three dimensions of (flowability) are sorted from smallest to largest, resulting in three sets of sorting results.

[0182] Step e6: For each sorted initial individual, calculate its crowding distance with its neighboring initial individuals for each metric.

[0183] Specifically, the electronic device can define the crowding distance between the two outermost individuals as infinite (∞) for the sorted initial individuals, ensuring that high-quality individuals at the boundaries are not eliminated. For the middle initial individual k, the crowding distance is calculated using the following formula: Where: m represents the performance dimension (1=compressive strength, 2=tensile strength, 3=flowability). This represents the maximum performance value within this performance level. This is the minimum value for this performance within the grade. These are the performance values ​​of the two adjacent individuals of the individual k involved in the incident.

[0184] For each initial individual, the electronic device calculates the crowding distance d in three performance dimensions. k 1 d k 2 d k 3 .

[0185] Step e7: For each initial individual, calculate the crowding degree corresponding to the initial individual based on the crowding distance corresponding to the initial individual.

[0186] Specifically, the crowding degree of the initial individual is the sum of the crowding distances across all performance dimensions of the initial individual, as shown in the formula: CD k =d k 1 +d k 2 +d k 3 .

[0187] Step e8: For each initial individual in each level, remove the initial individuals whose crowding degree is less than the first preset crowding degree threshold.

[0188] Specifically, the electronic device can receive a first preset congestion threshold input by the user, or it can set the first preset congestion threshold based on engineering experience. Optionally, the electronic device can also take the average congestion of all individuals within that level × 0.5 as the first preset congestion threshold (e.g., if the average congestion of the first level is 10, then the first preset congestion threshold is 5).

[0189] Electronic devices can eliminate initial individuals (i.e., homogeneous individuals that "cluster") within a certain level whose crowding level is less than the first preset crowding level threshold, and retain scattered high-quality individuals, thereby reducing redundant individuals within the same level and improving population efficiency.

[0190] Step e9: Mark the initial individuals whose crowding degree is greater than the second preset crowding degree threshold as scarce high-quality individuals.

[0191] Specifically, the second preset crowding threshold can be taken as the average crowding level of all individuals within that level × 1.5 (e.g., if the average is 10, then the second preset crowding threshold is 15). Electronic devices can mark initial individuals with a crowding level greater than a second preset crowding threshold within this level as rare and high-quality individuals (with excellent performance and good diversity, and are the core resources of the population).

[0192] Step e10: Retain the rare and high-quality individuals from the initial individuals of the first level for the next iteration.

[0193] Specifically, electronic devices can directly include rare and high-quality individuals from the first tier into the parent candidate pool, without them participating in subsequent tournament competitions (to avoid eliminating core high-quality individuals). Rare and high-quality individuals are the "seed resources" of the population and are given priority for retention.

[0194] Step e11: For the first remaining initial individuals excluding the rare and high-quality individuals, pair them into multiple pairs.

[0195] Specifically, the first remaining initial individuals are the other initial individuals in the first tier excluding rare and high-quality individuals (e.g., the first tier includes 15 initial individuals, 3 rare and high-quality individuals, and 12 remaining initial individuals). The electronic device can randomly pair the first remaining initial individuals into 6 pairs (if there is an odd number, the last individual is paired with a random individual).

[0196] Step e12: For each pair of individuals, calculate the initial fitness of each remaining initial individual based on the multi-objective fitness function.

[0197] Specifically, the electronic device can calculate the F-value of each remaining individual based on the previously constructed multi-objective fitness function. fitness The value is the initial fitness of each remaining initial individual.

[0198] Step e13: Compare the initial fitness values ​​of the two remaining initial individuals in the individual pair, and mark the one with the higher initial fitness value as the tournament winner.

[0199] Specifically, for each pair of individuals, the electronic device compares the initial fitness values ​​of the two remaining initial individuals, and the one with the higher value is the "tournament winner".

[0200] Step e14: If the initial fitness values ​​of the two remaining initial individuals are equal, the one with the higher crowding is marked as the tournament winner.

[0201] Specifically, if the initial fitness values ​​of the two remaining initial individuals are exactly the same (e.g., both are 0.8), the electronic device can compare the crowding of the two remaining initial individuals, and the one with higher crowding wins (prioritizing the retention of individuals with better diversity). This avoids selecting homogeneous individuals when fitness is the same.

[0202] Step e15, repeat this process a preset number of times to obtain the winning individuals for each competition.

[0203] Specifically, the electronic device can cycle a preset number of times, randomly pairing individuals again in each cycle, and ultimately retaining individuals that have won multiple times (e.g., individuals that have won 2 or more times in 3 cycles). From the first remaining individuals, core high-quality individuals are selected (e.g., 6 winning individuals are selected from 12 remaining individuals).

[0204] The preset number of attempts can be determined based on engineering experience or input by the user. The preset number of attempts can be 3 to 5 (the more attempts, the stricter the filtering).

[0205] Step e16: Combine the rare and high-quality individuals and the winners of the competition to form the initial parent population.

[0206] Specifically, the electronic device can merge "three rare and high-quality individuals" + "six individuals selected from the tournament" to form an initial parent population (a total of nine individuals). This allows the initial parent population to possess both the characteristics of "rare and high-quality" and "high-quality selected from the tournament," but diversity needs to be verified.

[0207] Step e17: If the variance of the target feature in the initial parent population is less than a preset variance threshold, then find a preset number of diverse individuals whose target feature values ​​differ from the preset difference threshold from the second remaining initial individuals, excluding rare high-quality individuals and individuals that won the competition.

[0208] Specifically, electronic devices can select the core features that have the greatest impact on performance (water-gel ratio x1, bacterial concentration x8, HPMC x3), and then calculate the variance of the target feature values ​​in the initial parent population. If the variance of the values ​​is less than the preset variance threshold (e.g., x1 variance < 0.001), it indicates that the feature values ​​are highly homogenized.

[0209] The electronic device can search for initial individuals whose target feature value difference is greater than a preset difference threshold (e.g., the difference between x1 and the mean of the parent population is greater than 0.05) from the second remaining initial individuals (all initial individuals except for rare and high-quality individuals and tournament winners), and record them as diverse individuals.

[0210] Step e18: Add diverse individuals to the initial parent population to obtain the target parent population.

[0211] Specifically, the electronic device can supplement the initial parent population with a predetermined number of diverse individuals selected from the screening process, ultimately obtaining the target parent population. This ensures the quality of the parent population while avoiding homogeneity in feature values, providing sufficient diversity for subsequent crossover and mutation operations.

[0212] The preset number of replacements can be 20% to 30% of the parent population size (e.g., if there are 9 parents initially, 2 to 3 can be added).

[0213] Step S2043: Perform crossover and mutation operations on the target parent population to obtain updated individuals.

[0214] Specifically, the electronic device can set the crossover probability based on engineering experience values, or it can receive the crossover probability input by the user. For example, the crossover probability is set to p. c =0.7 (to ensure population iteration efficiency).

[0215] The electronic device can randomly select pairs of initial individuals (e.g., initial individual A [x1=0.5, x8=6], initial individual B [x1=0.48, x8=7]) from the target parent population based on the crossover probability. Then, it generates offspring eigenvalues ​​according to the SBX crossover formula: ; Where β is the distribution factor (taken as 1 to 2, controlling the degree of crossover). and These are the i-th feature values ​​of parent individuals A and B, respectively. The electronic device can verify whether the feature values ​​after crossover meet the dynamic working condition constraints of step d3 (e.g., after crossover, x1=0.56, if v>1.0m / s then adjust to 0.52).

[0216] For example, if the initial individual A has a water-to-gel ratio x1 = 0.5, the initial individual B has a water-to-gel ratio x1 = 0.48, and β = 1.2, then the offspring x after crossover will be... new,1 =0.5×[(1+1.2)×0.5+(1-1.2)×0.48]=0.502.

[0217] Electronic devices can set mutation probabilities based on engineering experience values, or they can receive mutation probabilities input by the user. For example, the mutation probability can be set to p. m =0.1 (far lower than the crossover probability, to avoid disrupting high-quality feature combinations). The electronic device can randomly select mutation features from the target parent population (prioritizing core features with significant performance impact: water-gel ratio x1, bacterial concentration x8, HPMC x3). Then, the electronic device can fine-tune the feature values ​​according to the polynomial mutation formula: ;in, The asynchronous length is variable (ranging from -0.1 to 0.1). and Let be the upper and lower limits of the value of this feature, respectively.

[0218] Electronic devices can detect the value of the verification feature after mutation (e.g., if x8=9 after mutation, it is adjusted to 8, which meets the constraint that x8≤8Cell / mL).

[0219] After crossover and mutation operations, a new generation of updated individuals (denoted as Q0) is generated, with the same size as the target parent population, forming an updated individual ID-feature vector-performance vector data table.

[0220] Step S2044: Calculate the update fitness of each update individual based on the multi-objective fitness function.

[0221] Specifically, based on the multi-objective fitness function, the electronic device calculates the update fitness corresponding to each update individual.

[0222] Step S2045: Based on the update fitness of each updated individual, perform crossover and mutation operations on each updated individual until the preset stopping condition is met, and output the candidate individual set.

[0223] Specifically, the electronic device performs crossover and mutation operations on each updated individual based on their corresponding fitness. Round 1: Target parent P0 → crossover and mutation → update population Q0 → calculate fitness → select high-quality individuals to form P1; Round 2: P1 → crossover and mutation → Q1 → calculate fitness → select high-quality individuals to form P2; this process iterates until a preset stopping condition is met. The preset stopping condition is (any one of the following is sufficient): ≥50 iterations (the upper limit of engineering experience); the fitness value of the best individual increases by <0.01 after 10 consecutive iterations (convergence); the fitness of the best individual is ≥0.95 (reaching the engineering optimal standard). The electronic device summarizes the high-quality individuals with fitness ≥0.8 from all iterations, removing duplicate individuals (individuals with completely identical feature vectors); finally forming a candidate individual set (typically 10-20 individuals).

[0224] Step S2046: Determine the target underwater repair material mix ratio from the candidate individual set.

[0225] Specifically, step S2046 above may include the following steps: Step f1: Detect whether each candidate individual in the candidate individual set meets the preset biocompatibility core index and the preset hydration synergy index.

[0226] Specifically, the preset core biocompatibility indicators can include the following: bacterial concentration-nutrient compatibility ≥ 0.8 (0~1), the detection logic is: if the molar ratio of bacterial concentration x8 to urea x6 / calcium chloride x7 is within the range of 1:(1~1.2), the compatibility = 1; if it deviates, it decreases linearly; diatomaceous earth-biological component synergy ≥ 0.7 (0~1), the detection logic is: if diatomaceous earth x2 is 3%~7% (when biological components are present), the synergy = 1; if it deviates, it decreases linearly.

[0227] The preset hydration synergy index test items may include the following: water-cement ratio-water-reducing agent compatibility ≥ 0.85 (0~1), the test logic is: if the ratio of water-cement ratio x1 to water-reducing agent x4 is within the range of 500:(0.5~1.5), the compatibility = 1; if it deviates, it decreases linearly; curing age-strength synergy ≥ 0.9 (0~1), the test logic is: for strength priority schemes, the curing age is ≥ 28 days, and for flowability priority schemes, it is ≥ 14 days, the synergy = 1; if it is insufficient, it decreases linearly.

[0228] Electronic devices can input preset threshold formulas to calculate the indicator values ​​for each candidate individual and determine whether they meet the standards.

[0229] Step f2: Delete all candidate individuals that do not meet the preset biocompatibility core indicators and / or preset hydration synergy indicators, and obtain the remaining candidate individuals.

[0230] Specifically, the electronic device can delete candidate individuals that do not meet the preset biocompatibility core indicators and / or preset hydration synergy indicators, resulting in the remaining candidate individuals. This ensures that the remaining candidate individuals meet the core engineering characteristics of underwater repair, avoiding the failure of a mix design that is "high-performing but poorly biocompatible / hydration-compatible".

[0231] Step f3: Calculate the core performance score, biocompatibility score, and constraint satisfaction score for each remaining candidate individual.

[0232] Specifically, the electronic device can calculate the core performance score S1 for each remaining candidate individual, using the following formula: (Weight is adapted to scenario 1, and other scenarios are adjusted according to scenario weight); Value range: 0~1, the larger the value, the better the core performance.

[0233] The electronic device can calculate the biocompatibility score S2 for each remaining candidate individual, using the following formula: Value range: 0~1, the larger the value, the better the biocompatibility; The electronic device can calculate the constraint satisfaction score S3 for each remaining candidate individual: Formula: Value range: 0~1, the larger the value, the higher the constraint satisfaction (the smaller the penalty value).

[0234] Step f4: Based on the core performance score, biocompatibility score, and constraint satisfaction score of each remaining candidate individual, obtain the comprehensive score for each remaining candidate individual.

[0235] Specifically, the electronic device can perform a weighted summation of the core performance score, biocompatibility score, and constraint satisfaction score corresponding to each of the remaining candidate individuals to obtain a comprehensive score for each of the remaining candidate individuals.

[0236] The electronic device can receive the weight information corresponding to the core performance score, biocompatibility score and constraint satisfaction score input by the user, and can also determine the weight information of the core performance score, biocompatibility score and constraint satisfaction score corresponding to each remaining candidate individual based on the scenario corresponding to each remaining candidate individual.

[0237] Step f5: Select the remaining candidate individual with the highest overall score as the target individual.

[0238] Specifically, the electronic device can compare the comprehensive scores of each remaining candidate individual and identify the remaining candidate individual with the highest comprehensive score as the target individual.

[0239] Step f6: Determine the target underwater repair material mix ratio based on the target individual.

[0240] Specifically, electronic devices can extract the feature vector [x1, x2, ..., x] of the target individual. 13 (All characteristic values ​​such as water-cement ratio, diatomaceous earth, HPMC, water-reducing agent, bacterial concentration, etc.) are selected, and finally the target underwater repair material mix ratio is output.

[0241] The underwater repair material mix proportion determination method provided in this embodiment obtains the initial input features corresponding to the initial underwater repair material mix proportion. It comprehensively collects three key input features of the underwater repair material mix proportion: core components, process parameters, and service conditions. This covers all dimensions of influencing factors, including the material itself, the construction process, and the application environment, providing complete and accurate basic data support for subsequent performance index prediction, ensuring that the prediction results closely match engineering realities. The initial input features are input into a preset index prediction model. The model's quantitative analysis capabilities replace traditional manual experience-based judgment, achieving efficient and objective prediction of material performance indicators, avoiding human error, and significantly reducing the workload of offline testing, thus improving the efficiency and scientific rigor of performance index determination. The preset index prediction model identifies the core features of the mix proportion and process and service condition features, accurately classifying the initial input features and clarifying the influence dimensions of different features on material performance. This provides a clear direction for subsequent feature processing and fusion, avoiding interference from irrelevant features, making subsequent feature operations more targeted, and improving the accuracy of model predictions. At least one coupled feature is extracted and constructed from the initial input features. The nonlinear coupling relationships between different initial input features are explored, transforming single features into coupled features that better reflect the material performance characteristics. This compensates for the insufficient explanatory power of single features and improves the model's predictive ability for material performance under complex operating conditions. The coupled features are then fused with the initial input features to obtain initial fused features. This integrates single basic features and coupled features to form a multi-dimensional, multi-level feature system. This system retains the basic information of the original features while incorporating the synergistic / balancing relationships between features, providing the model with more comprehensive feature input and further improving the accuracy of performance indicator predictions. Attribute information is labeled for each fused sub-feature of the initial fused features, assigning clear attribute labels (such as component features, process features, operating condition features, and coupling features). This clarifies the attribute type and functional dimension of each feature, providing a clear basis for subsequent redundancy removal and ensuring the logic and standardization of feature processing. Redundancy is eliminated from the initial fusion features to obtain the target fusion features. Redundant features, such as those that are repetitive, inefficient, or overly correlated, are removed, simplifying the feature system, reducing model computation, and improving operational efficiency. This also avoids overfitting caused by feature redundancy, making the target fusion features more representative and further improving the accuracy and stability of performance indicator predictions. Based on the target fusion features, initial compressive strength, initial splitting tensile strength, and initial flowability are determined. With these simplified and efficient target fusion features as a foundation, the core material performance indicators output by the model better align with actual engineering needs. The key performance of each initial mix proportion is accurately quantified, providing a unified and comparable quantitative evaluation basis for subsequent mix proportion selection and optimization.

[0242] Then, the electronic equipment checks whether the core performance indicators are within the preset range. By setting a hard screening threshold based on the preset indicator range, initial mix proportions that do not meet the performance standards are quickly eliminated, reducing the amount of ineffective work in subsequent optimization and ensuring that all mix proportions entering the next stage meet the basic performance requirements of underwater repair materials, thus improving overall optimization efficiency. Candidate underwater repair material mix proportions that meet the performance standards are identified, and mix proportion samples that meet the basic performance thresholds are screened to form a high-quality candidate sample library. This lays the foundation for the selection of subsequent alternative mix proportions, preventing performance-deficient mix proportions from entering the subsequent optimization stage and ensuring the effectiveness and relevance of the optimization work. Candidate mix proportions are evaluated, and alternative underwater repair material mix proportions are determined. Under the premise of meeting the basic performance standards, through multi-dimensional comprehensive evaluation, alternative mix proportions with better overall performance are selected, further streamlining the optimization samples and improving the efficiency and accuracy of subsequent genetic algorithm optimization, while ensuring the engineering practicality of the alternative mix proportions. Based on backup performance indicators, a multi-objective fitness function is constructed to transform the multi-objective performance requirements of underwater repair materials, such as compressive strength, tensile strength, and flowability, into a calculable and optimizable mathematical function. This quantifies the weight and constraint relationship of each performance, overcomes the limitations of traditional single-index evaluation, provides a unified objective guide for subsequent genetic algorithm optimization, and ensures that the optimization results can take into account the multi-dimensional performance requirements of underwater repair.

[0243] Next, the electronic device determines the backup matching ratio as the initial individuals in the initial population, transforms the backup matching ratio at the engineering level into digital individuals that can be processed by the genetic algorithm, completes the population initialization, lays the foundation for subsequent parent population screening, crossover and mutation iteration, ensures that the samples optimized by the algorithm are matching ratios with better overall performance, and improves the starting point and efficiency of the algorithm optimization.

[0244] The target parent population is determined from the initial population. Through operations such as non-dominated sorting, crowding screening, and tournament selection, a target parent population with both high quality and diversity is selected from the initial population. This avoids optimization getting stuck in local optima due to population homogeneity, while ensuring the high performance foundation of the parent population, providing high-quality "seed" samples for subsequent crossover and mutation iterations. Crossover and mutation are performed on the target parent population to obtain updated individuals. Crossover operations achieve combination optimization of high-quality mixing ratio features, and mutation operations achieve local fine-tuning of feature values, exploring better combinations of mixing ratio features, uncovering the performance potential of the mixing ratio, avoiding algorithm convergence to local optima, and expanding the exploration range of optimal mixing ratios. The update fitness of the updated individuals is calculated based on a multi-objective fitness function. Using the multi-objective fitness function as an evaluation criterion, the comprehensive performance and operational adaptability of each updated individual are quantified, providing a clear basis for judging the quality of individuals in subsequent iterative optimizations. This ensures that iterative optimization always revolves around the multi-objective performance requirements of underwater repair, improving the directionality and effectiveness of the optimization. The process iterates through crossover and mutation until a preset stopping condition is met, outputting a set of candidate individuals. Through multiple rounds of crossover and mutation-fitness evaluation, the overall performance of the population is continuously optimized until a convergence criterion is reached. Finally, the set of candidate individuals with the best overall performance after multiple iterations is output, ensuring that the set includes mix proportions with excellent adaptability and performance under underwater conditions. This provides sufficient high-quality samples for determining the final target underwater repair material mix proportion. The target underwater repair material mix proportion is then determined from the set of candidate individuals. From the high-quality candidate individuals optimized through multiple iterations, the target underwater repair material mix proportion with the best overall performance and strongest adaptability to operating conditions is selected. This approach considers the material's multi-objective performance requirements (compressive strength, tensile strength, flowability) while conforming to the actual engineering conditions of underwater repair. Ultimately, the optimal mix proportion that can be directly applied to engineering practice is obtained, realizing the transformation from theoretical optimization to engineering implementation.

[0245] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for determining the mix proportion of underwater repair materials, characterized in that, The method includes: Obtain various initial underwater repair material mix proportions; Determine the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to the respective initial underwater repair material mix proportions; Based on the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to each of the initial underwater repair material mix proportions, a multi-objective fitness function is constructed. Based on the multi-objective fitness function, the initial underwater repair material mix ratios are optimized to obtain the target underwater repair material mix ratio.

2. The method according to claim 1, characterized in that, Determining the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to each of the initial underwater repair material mix proportions includes: For each of the initial underwater repair material mix proportions, the initial input features corresponding to the initial underwater repair material mix proportions are obtained; the initial input features include water-cement ratio, diatomaceous earth content, HPMC content, water-reducing agent content, yeast extract content, urea concentration, calcium chloride concentration, bacterial concentration, curing age, hydration time, curing method, underwater flow velocity, and applied water pressure. The initial input features are input into a preset index prediction model; The preset index prediction model performs feature recognition on the initial input features to determine the core features of the mix proportion and the process and operating condition features. Feature extraction is performed on the initial input features to construct at least one coupled feature; The coupled features are fused with the initial input features to obtain the initial fused features; Attribute information is labeled for each fusion sub-feature in the initial fusion feature; Based on the attribute information corresponding to each of the fusion sub-features, a redundancy removal operation is performed on the initial fusion feature to obtain the target fusion feature corresponding to the initial fusion feature; Feature recognition is performed on the target fusion features to determine the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to the initial underwater repair material mix ratio.

3. The method according to claim 2, characterized in that, The coupling features include at least one of the following: stable synergistic features, reaction matching features, mineralization efficiency features, interface enhancement features, and hydration regulation features; the step of extracting features from the initial input features to construct at least one coupling feature includes: Based on the bacterial concentration and the HPMC dosage, a stable synergistic class feature is constructed; Based on the urea concentration and the hydration time, a reaction matching feature is constructed. Based on the bacterial concentration and the calcium chloride concentration, a mineralization efficiency feature is constructed. Based on the underwater flow velocity and the HPMC dosage, construct a working condition adaptation feature; Based on the diatomite content and the bacterial concentration, interface enhancement features are constructed. Based on the water-reducing agent dosage and the water-cement ratio, a hydration regulation feature is constructed.

4. The method according to claim 2, characterized in that, The step of performing a redundancy removal operation on the initial fusion feature based on the attribute information corresponding to each of the fusion sub-features to obtain the target fusion feature corresponding to the initial fusion feature includes: Based on the attribute information corresponding to each of the fusion sub-features, the fusion sub-features are combined in pairs to generate multiple feature pairs; Calculate the Pearson correlation coefficient and Spearman correlation coefficient for each of the feature pairs. The feature pairs whose Pearson correlation coefficient and Spearman correlation coefficient are both greater than a preset correlation coefficient threshold are identified as target feature pairs. For each of the target feature pairs, the first target fusion sub-feature in the target feature pair is removed, and the first compressive strength, first splitting tensile strength and first flowability corresponding to the first remaining feature after removing the first target fusion sub-feature are determined. Remove the second target fusion sub-feature from the target feature pair, and determine the second compressive strength, second splitting tensile strength, and second flowability corresponding to the second remaining feature after removing the second target fusion sub-feature; The first compressive strength, the first splitting tensile strength, and the first flowability are compared with the second compressive strength, the second splitting tensile strength, and the second flowability to determine whether the first target fusion sub-feature and the second target fusion feature in the target feature pair have unique contributions. Based on the comparison results, a redundancy removal operation is performed on the target feature pairs to obtain the target fused features.

5. The method according to claim 1, characterized in that, The multi-objective fitness function is constructed based on the initial compressive strength, initial splitting tensile strength, and initial flowability corresponding to each of the initial underwater repair material mix proportions, including: The initial compressive strength, initial splitting tensile strength, and initial flowability of each initial underwater repair material mix ratio are tested to see if they are all within the corresponding preset index ranges. The initial underwater repair material mix proportions whose initial compressive strength, initial splitting tensile strength, and initial flowability are all within the corresponding preset index ranges are determined as candidate underwater repair material mix proportions; The mix proportions of each of the candidate underwater repair materials are evaluated to determine the mix proportions of each backup underwater repair material. Based on the alternative compressive strength, alternative splitting tensile strength, and alternative flowability corresponding to the respective alternative underwater repair material mix proportions, a multi-objective fitness function is constructed.

6. The method according to claim 5, characterized in that, The evaluation of the mix proportions of each of the candidate underwater repair materials, and the determination of each standby underwater repair material mix proportion, includes: Based on the candidate compressive strength, candidate splitting tensile strength and candidate flowability corresponding to the mix proportions of each candidate underwater repair material, the baseline value is calculated. For each of the candidate underwater repair material mix proportions, based on the benchmark value, the candidate local SHAP value of each initial input feature in the candidate underwater repair material mix proportion is calculated for the candidate underwater repair material mix proportion. For each of the initial input features, based on the candidate local SHAP value of the initial input feature for the candidate underwater repair material mix ratio, the candidate global SHAP value of the initial input feature for all the candidate underwater repair material mix ratios is calculated; Based on the candidate global SHAP values, a correlation between changes in feature values ​​and changes in performance contribution is constructed. Based on the aforementioned change correspondence, check whether the mixing ratio of each candidate underwater repair material conforms to the global rule verification. Based on the candidate local SHAP values, it is determined whether the mix ratio of each candidate underwater repair material conforms to the optimal layout interval verification. Based on the verification results, the candidate underwater repair material mix proportions that conform to both the global rule verification and the optimal layout interval verification are determined as the backup underwater repair material mix proportions.

7. The method according to claim 5, characterized in that, The multi-objective fitness function is constructed based on the backup compressive strength, backup splitting tensile strength, and backup flowability corresponding to the mix proportions of each of the backup underwater repair materials, including: Obtain the application scenarios corresponding to the mix proportions of each of the aforementioned backup underwater repair materials; Based on the application scenarios corresponding to each of the aforementioned spare underwater repair material mix proportions, determine the scenario weight corresponding to each of the aforementioned spare underwater repair material mix proportions; Based on the underwater flow velocity and applied water pressure corresponding to the mix proportions of each of the aforementioned backup underwater repair materials, a dynamic working condition constraint function is constructed. Obtain the backup global SHAP value of each initial input feature in each of the backup underwater repair material mix proportions for all the backup underwater repair material mix proportions; Based on the aforementioned backup global SHAP value, a feature contribution constraint function is constructed; Based on the mandatory requirements of engineering specifications and the aforementioned backup global SHAP value, a performance hard constraint function is set; Based on the scenario weights corresponding to the respective mix proportions of the backup underwater repair materials, a basic objective function is constructed. Based on the dynamic operating condition constraint function, the feature contribution constraint function, and the performance hard constraint function, a constraint penalty function is constructed. Based on the basic objective function and the constraint penalty function, the multi-objective fitness function is constructed.

8. The method according to claim 7, characterized in that, The optimization of each initial underwater repair material mix ratio based on the multi-objective fitness function to obtain the target underwater repair material mix ratio includes: The mixing ratio of each of the aforementioned spare underwater repair materials is determined as the initial individual in the initial population; Determine the target parent population from the initial population; The target parent population is subjected to crossover and mutation operations to obtain updated individuals. Based on the multi-objective fitness function, the update fitness corresponding to each updated individual is calculated; Based on the update fitness of each updated individual, crossover and mutation operations are performed on each updated individual until a preset stopping condition is met, and a set of candidate individuals is output. The mix proportion of the target underwater repair material is determined from the set of candidate individuals.

9. The method according to claim 8, characterized in that, The step of determining the target parent population from the initial population includes: Traverse each of the initial individuals in the initial population to determine the dominance relationships between the initial individuals; Based on the dominance relationship, identify the first-level initial individuals that are not dominated by any other initial individuals; Remove the first-level initial individuals and find the second-level initial individuals in the remaining population that are not dominated by any other initial individuals; This process is repeated to complete the ranking of all initial individuals; For each initial individual of the same level, sort them from smallest to largest according to each performance indicator; For each of the sorted initial individuals, calculate its crowding distance from its neighboring initial individuals for each metric; For each initial individual, the crowding degree corresponding to the initial individual is calculated based on the crowding distance corresponding to the initial individual; For each initial individual in each level, the initial individuals with a crowding degree less than a first preset crowding degree threshold are removed; The initial individuals whose crowding level is greater than a second preset crowding threshold are marked as scarce high-quality individuals; Each of the rare and high-quality individuals in the initial individuals of the first level will be retained for the next iteration; For the first remaining initial individuals other than the aforementioned scarce high-quality individuals, pair them into multiple pairs to obtain multiple pairs of individuals; For each of the aforementioned individual pairs, the initial fitness corresponding to each of the remaining initial individuals is calculated based on the multi-objective fitness function; Compare the initial fitness values ​​of the two remaining initial individuals in the pair of individuals, and mark the one with the higher initial fitness value as the tournament winner; If the initial fitness values ​​of the two remaining initial individuals are equal, the one with the higher crowding degree is marked as the tournament winner. This process is repeated a predetermined number of times to obtain the winning individuals for each of the aforementioned competitions; The rare and high-quality individuals and the winning individuals of the competition are combined to form the initial parent population. If the variance of the target feature in the initial parent population is less than a preset variance threshold, then a preset number of diverse individuals with a target feature value difference greater than a preset difference threshold are searched from the second remaining initial individuals other than the rare high-quality individuals and the winning individuals in the competition. The diverse individuals are added to the initial parent population to obtain the target parent population.

10. The method according to claim 8, characterized in that, Determining the mix proportion of the target underwater repair material from the set of candidate individuals includes: Detect whether each candidate individual in the candidate individual set meets the preset biocompatibility core index and the preset hydration synergy index; Delete all candidate individuals that do not meet the preset biocompatibility core indicators and / or the preset hydration synergy indicators to obtain the remaining candidate individuals; Calculate the core performance score, biocompatibility score, and constraint satisfaction score for each of the remaining candidate individuals. Based on the core performance score, biocompatibility score, and constraint satisfaction score corresponding to each of the remaining candidate individuals, a comprehensive score is obtained for each of the remaining candidate individuals. The remaining candidate individuals with the highest overall score are selected as the target individuals. Based on the target individual, the mix proportion of the target underwater repair material is determined.

Images