A method, system and device for determining roadbed construction parameters

Abstract

The application belongs to the field of roadbed construction, and specifically discloses a method, system and device for determining roadbed compaction construction parameters, obtaining a sample data set of roadbed compaction; the sample data set comprises a plurality of sample data, each sample data comprising construction parameters of single-roller compaction and corresponding compaction degree increase value and construction time; taking the construction parameters as an input vector of KAN, taking the compaction degree increase value and the construction time as an output vector of KAN, training the KAN based on DBA, and taking the trained KAN as a prediction model; taking maximization of final compaction degree as a first optimization target, taking minimization of total construction time as a second optimization target, taking the prediction model as an agent model, and constructing a multi-objective optimization model; solving the multi-objective optimization model by using NSGA-II to obtain a Pareto optimal solution set. Through the application, high-precision prediction of roadbed compaction quality and efficiency is realized, and the decision level of intelligent compaction is significantly improved.

Description

Technical Field

[0001] This application belongs to the field of roadbed construction, and more specifically, relates to a method, system and equipment for determining roadbed compaction construction parameters. Background Technology

[0002] The compaction quality of highway subgrade directly affects the service life of the road and driving safety. Intelligent compaction technology, by integrating sensing, positioning, and data analysis systems and equipment, enables real-time continuous monitoring of the compaction process, significantly improving construction quality and efficiency. However, existing intelligent compaction technologies still face many challenges.

[0003] First, establishing a high-precision, highly interpretable compaction state prediction model from massive amounts of multi-source heterogeneous construction data is crucial for achieving process control. Traditional neural network models, such as multilayer perceptrons (MLPs), are often considered "black boxes," making it difficult to reveal the intrinsic mechanisms between input parameters and output results. Kolmogorov-Arnold Networks (KANs), proposed in recent years, are based on the Kolmogorov-Arnold representation theorem, decomposing multivariate functions into a superposition of univariate functions, possessing inherent interpretability and symbolic representation potential. However, the performance of KAN networks is highly dependent on the selection of their structural parameters (such as the number of network layers and the order of B-splines). Traditional manual parameter tuning or grid search methods are inefficient and cannot guarantee optimality. The second-generation non-dominated sorting genetic algorithm NSGA-II, by introducing fast non-dominated sorting, crowding distance, and an elitist strategy, exhibits good convergence and diversity in solving two-objective optimization problems. However, the classic NSGA-II lacks a mechanism for verifying the validity and repeatability of individuals during population initialization, crossover, and mutation operations. This leads to invalid individuals consuming computational resources, affecting convergence speed and solution quality. To address this issue, previous studies have improved NSGA-II by introducing validity and repeatability verification, achieving good results in engineering problems such as facility layout optimization.

[0004] Secondly, compaction construction needs to strike an optimal balance between two conflicting objectives: ensuring quality (maximizing final compaction) and improving efficiency (minimizing total construction time). Currently, the selection of construction parameters relies heavily on engineering experience and lacks scientific multi-objective optimization decision support.

[0005] Therefore, there is an urgent need for an intelligent compaction method, system, and equipment that can deeply integrate data-driven algorithms to achieve high-precision and highly interpretable prediction of compaction status and provide multi-objective optimization decision-making. Summary of the Invention

[0006] In view of the shortcomings of the prior art, the purpose of this application is to provide a method, system and equipment for determining roadbed compaction construction parameters, which aims to solve the problems of insufficient interpretability of compaction state prediction models, low efficiency of parameter optimization and lack of scientific decision support for construction parameter optimization in the prior art.

[0007] To achieve the above objectives, firstly, this application provides a method for determining roadbed compaction construction parameters, comprising the following steps: Obtain a sample dataset; the sample dataset includes multiple sample data, each of which includes single-pass construction parameters and the corresponding compaction increase value and construction time; The construction parameters in each sample data are used as the input vector of KAN, and the compaction increase and construction time are used as the output vector of KAN. KAN is optimized and trained based on the Detective Behavior Algorithm (DBA), and the trained KAN is used as the prediction model. The prediction model is used to predict the final compaction and total construction time based on the construction parameters of each pass. With maximizing the final compaction degree as the first optimization objective and minimizing the total construction time as the second optimization objective, a multi-objective optimization model is constructed using the prediction model as a surrogate model. The multi-objective optimization model is solved using NSGA-II to obtain the Pareto optimal solution set; the Pareto optimal solution set includes the optimal combination of construction parameters under different trade-offs between the first and second optimization objectives.

[0008] In one possible implementation, the above method further includes the following steps: The optimal compromise solution is selected from the Pareto optimal solution set using the fuzzy decision method, and is used as the final recommended subgrade compaction construction parameters.

[0009] In one possible implementation, the DBA-based optimization training of KAN includes: Based on the dimensions of the input vector and the output vector, the initial number of network layers and the number of neurons in each layer of the KAN are set; and the structural parameters to be optimized in the KAN are determined, including: the number of network layers, the number of neurons in each layer, the order of the B-spline function, and the number of grids. The structural parameters to be optimized are encoded into individual detectives, and the detective population is initialized. Each individual detective performs a preliminary search in the solution space, records the structural parameter combinations and fitness values ​​they attempt, and forms a clue database; the fitness value refers to the negative root mean square error (RMSE) obtained by cross-validation of KAN on the validation set under the corresponding structural parameters; Based on the fitness value of each detective in the clue database, the candidate detectives are sorted, and the top few detectives with the highest fitness values ​​are selected and marked as the suspect area. A refined search is conducted within the suspect area, new detective individuals are generated through a local perturbation mechanism, and dense sampling is performed to approximate the local optimum. By exchanging optimal clues through information sharing mechanisms among different detectives, updating their respective key search areas, and coordinating to adjust search strategies, individuals can avoid getting trapped in local optima. When the maximum number of iterations is reached or the fitness value converges, the optimal detective individual is output; The prediction model is obtained by taking the structural parameters corresponding to the optimal detective individual as the optimal KAN structural parameters.

[0010] In one possible implementation, the first optimization objective for: in, Initial compaction degree, The number of rolling passes, For the first The increase in compaction degree per pass is calculated and predicted by the prediction model. For the first Construction parameters for roadbed compaction; Second optimization objective for: in, The length of a single compaction pass. For the first The rolling speed of the whole process To supplement the work time.

[0011] In one possible implementation, the multi-objective optimization model is solved using NSGA-II to obtain the Pareto optimal solution set, including: Step S1: A two-layer coding mechanism is used to generate an initial population; the first layer of the two-layer coding mechanism is the construction parameter sequence coding, and the second layer is the vibration mode coding; each set of codes in the initial population corresponds to a chromosome. Step S2: Using a hybrid crossover strategy, for each pair of parent chromosomes, if the two chromosomes are completely identical, a mutation operation is introduced into the pair of chromosomes; otherwise, a single-point matching crossover is performed. Step S3: Using a multi-point mutation strategy, approximately one-third of the genes are randomly selected for position exchange, and the corresponding vibrational pattern genes are regenerated to obtain the parent population. Step S4: Select an individual population from the parent population through a tournament, and generate the offspring population through simulated binary crossover and polynomial mutation. Step S5: Merge the parent population and the offspring population, perform non-dominated sorting on the merged population, and select the best multiple individuals to form the next generation parent population according to the principles of hierarchy priority and crowding priority. Step S6: Determine whether the maximum number of iterations has been reached or the convergence condition has been met. If the maximum number of iterations has not been reached or the convergence condition has not been met, return to step S2. If the maximum number of iterations has been reached or the convergence condition has been met, output the non-dominated solution set in the final population, i.e., the Pareto optimal solution set.

[0012] In one possible implementation, in steps S1-S3, after generating the initial population, after mixing and crossover, and / or after multi-point mutation, each individual is subjected to validity and repeatability verification to check whether the constraints of the preset construction parameters are met and to ensure that there are no duplicate individuals in the population.

[0013] In one possible implementation, the step of selecting the optimal compromise solution from the Pareto optimal solution set using a fuzzy decision-making method includes: For each solution in the Pareto solution set, calculate the corresponding first membership degree and second membership degree for the first and second optimization objectives; wherein, the larger the compaction degree corresponding to each solution, the larger the first membership degree, and the smaller the corresponding total construction time, the larger the second membership degree. The sum of the first and second membership degrees of each solution is taken as its total membership degree; The ratio of the total membership degree of each solution to the sum of the total membership degrees of all solutions is taken as its overall dominance value; The solution with the largest overall dominance value is selected as the optimal compromise solution.

[0014] In one possible implementation, the first membership degree for: in, Let J be the compaction degree corresponding to the j-th solution. The preset lower limit of compaction. This is the preset upper limit of compaction. Second membership degree for: in, Let j be the total construction time corresponding to the j-th solution. This is the preset lower limit of the total construction time. This is the preset maximum total construction time. The overall dominance value of solution j for: Where n is the total number of solutions in the Pareto optimal solution set.

[0015] In one possible implementation, the construction parameters include: Initial compaction degree of subgrade, subgrade temperature, roller excitation force, roller amplitude, moisture content of subgrade fill material, gradation parameters of subgrade fill material, roller compaction speed, and number of roller compaction passes.

[0016] Secondly, this application provides a system for determining roadbed compaction construction parameters, comprising: The sample data acquisition module is used to acquire a sample dataset of roadbed compaction; the sample dataset includes multiple sample data, each of which includes construction parameters for a single pass of compaction and the corresponding compaction degree increase value and construction time; The prediction model training module is used to take the construction parameters in each sample data as the input vector of KAN, the compaction increase value and construction time as the output vector of KAN, optimize and train KAN based on DBA, and use the trained KAN as the prediction model; the prediction model is used to predict the final compaction degree and total construction time based on the construction parameters of each rolling pass. A multi-objective optimization model construction module is used to construct a multi-objective optimization model with the prediction model as a surrogate model, taking maximizing the final compaction degree as the first optimization objective and minimizing the total construction time as the second optimization objective. The construction parameter solution module is used to solve the multi-objective optimization model using the second-generation non-dominated sorting genetic algorithm NSGA-II to obtain the Pareto optimal solution set; the Pareto optimal solution set includes the optimal combination of construction parameters under different trade-offs between the first optimization objective and the second optimization objective.

[0017] Thirdly, this application provides an electronic device, including a memory and one or more processors; The memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions; The one or more processors invoke the computer instructions to cause the electronic device to perform the method as described in the first aspect above.

[0018] In summary, compared with the prior art, the technical solutions conceived in this application have the following main technical advantages: This application provides a method, system, and equipment for determining roadbed compaction construction parameters. It adopts a KAN network as the core prediction model. Its structure, based on the Kolmogorov-Arnold representation theorem, is inherently interpretable and can reveal the intrinsic functional relationship between input construction parameters and compaction quality indicators, thus overcoming the shortcomings of traditional neural network "black box" models.

[0019] This application provides a method, system, and equipment for determining roadbed compaction construction parameters. It utilizes the Detective Behavior Algorithm (DBA) to globally optimize the structural parameters of the KAN network. By simulating a multi-stage collaborative mechanism of detective investigation, DBA can effectively balance global exploration and local development, solving the problems of large parameter space and difficult parameter tuning in KAN.

[0020] This application provides a method, system, and equipment for determining roadbed compaction construction parameters. It combines a high-precision, highly interpretable DBA-KAN prediction model as a surrogate model with an improved NSGA-II multi-objective optimization algorithm. The improved NSGA-II effectively avoids wasting computational resources by using mechanisms such as validity verification, repeatability verification, double-layer coding, hybrid crossover, and multi-point mutation, thereby improving convergence speed and solution set quality. It can efficiently handle the conflict between compaction quality and construction efficiency, generating a set of Pareto-optimal construction parameter combinations, providing engineers with a scientific and quantitative basis for decision-making.

[0021] This application provides a method, system, and equipment for determining roadbed compaction construction parameters. In the modeling and optimization process, various construction parameters and material parameters are comprehensively considered. The convergence and diversity of the optimization solution are ensured by improving the elite strategy of NSGA-II, which has good engineering application value.

[0022] This application provides a method, system, and equipment for determining roadbed compaction construction parameters. Targeting a two-objective optimization problem, this application introduces validity verification and repeatability verification mechanisms based on the classic NSGA-II. Through double-layer encoding, hybrid crossover strategy, and multi-point mutation strategy, it ensures that each individual in the population satisfies engineering constraints while maintaining population diversity. This effectively avoids the waste of computational resources by invalid individuals and improves convergence speed and solution set quality. Attached Figure Description

[0023] Figure 1 This is a flowchart of the method for determining roadbed compaction construction parameters provided in the embodiments of this application; Figure 2 This is a flowchart of the method for predicting and optimizing subgrade compaction parameters based on the DBA-KAN-improved NSGA-II hybrid algorithm provided in the embodiments of this application; Figure 3This is a graph showing the predicted results of the final compaction test set provided in the embodiments of this application; Figure 4 This is a graph showing the final compaction training set fitting results provided in the embodiments of this application; Figure 5 This is a graph showing the predicted compaction time test set results provided in the embodiments of this application; Figure 6 This is a graph showing the fitting results of the compaction time training set provided in the embodiments of this application; Figure 7 This is an example diagram of the Pareto front diagram provided in the embodiments of this application; Figure 8 This is a comparison chart of the final compaction degree before and after optimization provided in the embodiments of this application; Figure 9 This is a comparison chart of the results before and after the optimization of the compaction time provided in the embodiments of this application; Figure 10 This is a system architecture diagram of the roadbed compaction construction parameter determination system provided in the embodiments of this application; Figure 11 This is an electronic device architecture diagram provided in the embodiments of this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0025] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0026] Furthermore, throughout this specification, references to "an embodiment"; "an embodiment," "an example," or similar language indicate that a particular feature, structure, or characteristic described in connection with that embodiment is included in at least one embodiment of this application. Therefore, the appearance of the phrase "in one embodiment;" throughout this specification, and similar language, may, but not necessarily, refer to the same embodiment.

[0027] Figure 1 This application provides a method for determining roadbed compaction construction parameters, such as... Figure 1 As shown, it includes the following steps: Step S101: Obtain a sample dataset of roadbed compaction; the sample dataset includes multiple sample data, each sample data includes construction parameters for a single pass of compaction and the corresponding compaction degree increase value and construction time; Step S102: The construction parameters in each sample data are used as the input vector of the Kolmogorov-Arnold network (KAN), and the compaction increase and construction time are used as the output vector of the KAN. The KAN is trained based on DBA optimization, and the trained KAN is used as the prediction model. The prediction model is used to predict the final compaction and total construction time based on the construction parameters of each rolling pass. Step S103: With maximizing the final compaction degree as the first optimization objective and minimizing the total construction time as the second optimization objective, a multi-objective optimization model is constructed using the prediction model as a surrogate model. Step S104: Solve the multi-objective optimization model using NSGA-II to obtain the Pareto optimal solution set; the Pareto optimal solution set includes the optimal combination of construction parameters under different trade-offs between the first and second optimization objectives.

[0028] The NSGA-II provided in this application, based on the classic NSGA-II framework, introduces validity verification and repeatability verification mechanisms, specifically including the following improvements: Improved population initialization: A two-layer coding mechanism is adopted to generate an initial population, in which each set of codes in the initial population corresponds to a chromosome; the first layer of the two-layer coding mechanism is the construction parameter sequence coding, and the second layer is the vibration mode coding; when generating the initial population, each individual is validated for validity and repeatability to check whether all constraints are met and to ensure that there are no duplicate individuals in the population.

[0029] Improved crossover operation: A hybrid crossover strategy is adopted. For each pair of parent chromosomes, if the two chromosomes are completely identical, a mutation operation is introduced into the pair of chromosomes; otherwise, a single-point matching crossover is performed. After the crossover is completed, validity and repeatability are verified. Improved mutation operation: A multi-point mutation strategy is adopted, randomly selecting about one-third of the genes for position exchange and regenerating the corresponding vibrational mode genes; after mutation, validity and repeatability are verified.

[0030] Understandably, this application, taking into account the technological characteristics of roadbed compaction construction and the need for multi-objective optimization decision-making, designs a hybrid intelligent method that integrates Kolmogorov-Arnold Network (DBA-KAN) optimized by detective behavior algorithm and an improved non-dominated sorting genetic algorithm (improved NSGA-II) to solve the following three technical problems existing in the current intelligent compaction technology for highway roadbeds: (1) The multi-objective optimization of construction parameters lacks scientific decision support: Compaction quality (final compaction degree) and construction efficiency (total construction time) are two conflicting objectives. Existing methods mostly rely on engineering experience or single-objective optimization to select construction parameters, which cannot quantify the trade-off between the two objectives and lack scientific Pareto optimal decision basis.

[0031] (2) Low efficiency in optimizing the structural parameters of the prediction model: Although KAN has good interpretability potential, its structural parameters (number of network layers, number of neurons, B-spline order, number of grids, etc.) have a significant impact on prediction accuracy. Traditional manual parameter tuning or grid search methods are inefficient and cannot guarantee obtaining the global optimal structure.

[0032] (3) Poor interpretability of compaction state prediction model: Existing compaction degree prediction models based on neural networks such as BP and MLP lack clear physical meaning in their internal weights and biases, and cannot explain the independent influence law between input construction parameters and output compaction quality. This makes it difficult for engineers to trust the model prediction results, which limits the engineering promotion and application of intelligent compaction system.

[0033] As a further preferred option, the construction parameters for single-pass compaction include: material state parameters (initial compaction degree, subgrade temperature), equipment input parameters (excitation force, vibration amplitude), soil properties (moisture content, gradation parameters), and process control parameters (compacting speed, number of compaction passes).

[0034] The above construction parameters are divided into four categories, totaling eight parameters. The independent influence paths of each parameter on the output parameters are as follows: (1) Material state parameters Initial compaction degree ( The impact on the final compaction degree is reflected in the baseline relationship. The higher the initial compaction degree, the greater the incremental compaction in each subsequent compaction pass. The smaller the value, the lower the trend. It has no direct impact on the total construction time.

[0035] Roadbed temperature ( The effect of temperature on the final compaction degree manifests as the influence of water viscosity and evaporation rate. Below 5℃, water freezes, hindering particle movement; above 30℃, water evaporates rapidly, deviating from the optimum moisture content. The effect is minimal only within the suitable range (5~30℃). It has no direct impact on the total construction time.

[0036] (2) Equipment input parameters Excitation force ( ) and amplitude ( The combination of these two factors determines the vibration energy input. The greater the energy, the easier it is for soil particles to rearrange, resulting in a greater compaction increment; however, exceeding a threshold will cause "over-vibration," leading to surface loosening. Typically, this is... Characterizes energy intensity. Has no direct impact on total construction time.

[0037] (3) Soil parameters Moisture content ( The moisture content has a parabolic effect on the final compaction degree. Too low a moisture content (<6%) results in high interparticle friction, making particle movement difficult; too high a moisture content (>14%) leads to increased pore water pressure, pushing particles apart. There is an optimal moisture content range (typically 8%~12%). It has no direct impact on the total construction time.

[0038] Grading parameters (uniformity coefficient) curvature coefficient The effect on the final compaction degree is manifested in good gradation ( When fine particles fill the pores of coarse particles, it is easier to achieve high density. This has no direct impact on the total construction time.

[0039] (4) Process control parameters Rolling speed ( The vibration velocity has an inverse relationship with the final compaction degree; the lower the velocity, the longer the vibration energy acts, and the greater the compaction increment. The larger the layer, the higher the speed, and the less energy can be transferred to deeper layers. This is inversely proportional to the total construction time. .

[0040] Number of rolling passes ( The effect on the final compaction degree exhibits diminishing marginal returns; the compaction increment decreases with increasing number of passes, eventually approaching the ultimate compaction degree. It is directly proportional to the total construction time. .

[0041] The prediction model described above predicts the final compaction degree and total construction time as defined below: (1) Final compaction degree (%): Characterizes the compaction degree of the roadbed after compaction and is the core indicator for measuring compaction quality.

[0042] (2) Total construction time (seconds): The total time taken to complete all compaction passes is a core indicator for measuring construction efficiency.

[0043] It should be noted that the final compaction degree Total construction time These are a pair of typical contradictory goals, and the root causes of their conflict can be analyzed from the following three aspects.

[0044] (1) The two-way effect of compaction speed: The influence of compaction speed on two objectives is completely opposite. From the perspective of compaction quality, the lower the speed, the longer the vibration energy acts on the soil per unit length, the more complete the particle rearrangement, and the greater the increase in compaction degree. The larger the speed, the more inversely proportional the compaction time; from the perspective of construction efficiency, the lower the speed, the shorter the time per compaction cycle. The longer the speed, the longer the total construction time. Therefore, if too low a speed is used in pursuit of compaction quality, the construction time will be significantly extended; conversely, if too high a speed is used in pursuit of construction efficiency, the compaction degree will be difficult to achieve. This constitutes a fundamental conflict in the speed dimension.

[0045] (2) The two-way effect of the number of compaction passes: The number of compaction passes also has opposite effects on the two objectives. Increasing the number of compaction passes can accumulate the compaction increment, but it follows the law of diminishing marginal returns, that is, the compaction degree is significantly improved in the first few passes, and the increment of subsequent passes gradually approaches zero; at the same time, the total construction time increases linearly in proportion to the number of compaction passes. Therefore, there is an optimal range of passes, rather than the more passes the better. Excessively increasing the number of passes will seriously sacrifice construction efficiency, while the improvement in compaction degree is negligible; but too few passes may lead to the compaction degree not meeting the requirements of the engineering specifications.

[0046] (3) Trade-offs in vibration modes: The selection of vibration modes (the combination of excitation force and amplitude) also reflects the conflict between quality and efficiency. High-energy modes can increase the compaction increment per pass, thus potentially achieving the target compaction degree in fewer passes, indirectly saving construction time; however, if the energy is too high, it can lead to surface soil breakage or "over-vibration," which in turn reduces the compaction degree. Low-energy modes, while safe, require more compaction passes to achieve the same compaction degree, thus increasing construction time. Therefore, the selection of vibration modes needs to strike a balance between "rapidly achieving the target" and "avoiding over-vibration."

[0047] As a further preferred option, the training optimization of the above-mentioned KAN network includes the following steps: 1. Preprocess the data in the sample dataset, including data cleaning, outlier removal, and normalization; 2. The Detective Behavior Algorithm (DBA) is introduced, using the root mean square error (RMSE) of cross-validation as the fitness function to globally optimize the structural parameters of the KAN network. The structural parameters include the number of network layers, the number of neurons in each layer, the type of activation function, and the order and number of grids of the B-spline function. 3. Utilize the optimal combination of structural parameters to establish a DBA-KAN prediction model based on the training sample set, and use the test sample set to verify the prediction accuracy and interpretability of the model.

[0048] As a further preferred embodiment, the core structure of the KAN is based on the Kolmogorov-Arnold representation theorem, which represents a multivariable function as a superposition of single-variable functions. For an input vector... The KAN network achieves function fitting through multiple layers, and the transformation relationship from the l-th layer to the (l+1)-th layer can be expressed as: in, and All parameters are learnable univariate functions, parameterized using B-spline curves. This structure makes KAN networks more interpretable and parameter-efficient compared to traditional MLPs.

[0049] As a further preferred embodiment, the Detective Behavior Algorithm (DBA) simulates the clue collection, suspect screening, evidence analysis, and collaborative case-solving mechanisms in a detective investigation process to achieve a global search for optimization problems. Its core operations include: (1) Clue collection stage: Each detective conducts an initial search in the solution space, collects location information of potential optimal solutions, and records fitness values ​​along the search path; (2) Suspect screening stage: Candidate solutions are ranked according to the quality of clues (i.e., fitness value), and the most promising "suspect" areas are selected as key search targets. (3) Evidence analysis stage: A refined search is conducted in the suspect's area, and high-quality evidence is obtained through localized excavation, i.e., intensive application in this area to approach the local optimum: (4) Collaborative crime-solving stage: Different detectives exchange clues through an information-sharing mechanism, coordinate to adjust search strategies, avoid individuals getting stuck in local optima, and improve the global convergence ability of the population.

[0050] As a further optimization, maximizing the final compaction degree The objective function is: in, Initial compaction degree, The number of rolling passes, For the first The increase in compaction degree per pass was calculated using the DBA-KAN prediction model. For the first Construction parameters Minimize total construction time The objective function is: in, The length of a single compaction pass. For the first The rolling speed of the whole process To supplement the work time.

[0051] As a further preferred option, the above-mentioned construction parameters are decision variables, including: the compaction speed (v) of each compaction pass. i ), vibration mode (M) i The vibration mode is a combination of excitation force and amplitude, and its value is a discrete level.

[0052] As a further optimization, the improved NSGA-II algorithm maintains population diversity in each generation of evolution through fast non-dominated sorting, crowding distance calculation, and elitist selection operators, ultimately converging to the Pareto front. The algorithm flow is as follows: (41) Improved population initialization: A two-layer coding mechanism is adopted. The first layer is the construction parameter sequence coding, and the second layer is the vibration mode coding. When generating the initial population, each individual is validated for validity and repeatability to check whether all constraints are met and to ensure that there are no duplicate individuals in the population.

[0053] (42) Improved crossover operation: A hybrid crossover strategy is adopted. For each pair of parent chromosomes, if the two chromosomes are completely identical, a mutation operation is introduced into the pair of chromosomes; otherwise, a single-point matching crossover is performed; after the crossover is completed, validity verification and repeatability verification are performed. (43) Improved mutation operation: adopt a multi-point mutation strategy, randomly select about one-third of the genes for position exchange, and regenerate the corresponding vibrational mode genes; after the mutation is completed, perform validity verification and repeatability verification.

[0054] Selection, crossover, and mutation: Individuals are selected from the parents through tournament selection, and the offspring population is generated through simulated binary crossover and polynomial mutation. .

[0055] (44) Elite strategy: merging the parent and offspring populations into ,right Perform a non-dominated sort, selecting the optimal one based on the principles of hierarchy priority and crowding priority. Individuals constitute the next generation of the parent population. .

[0056] (45) Iteration Termination: When the maximum number of iterations is reached or the convergence condition is met, output the non-dominated solution set in the final population, i.e., the Pareto optimal solution set; otherwise, return to step (42).

[0057] The above step S104 is followed by: Step S105: The optimal compromise solution is selected from the Pareto optimal solution set using the fuzzy decision method, and is used as the final recommended combination of subgrade compaction construction parameters. Specifically: Step 1: Define a membership function for each solution in the Pareto solution set. Calculate the membership degree of each solution to each optimization objective; Step 2: Calculate the overall dominance value of each Pareto optimal solution based on the membership degree of each solution to each optimization objective. ; Step 3: Based on the principle of maximizing the overall dominance value, select the Pareto solution corresponding to the maximum dominance value as the optimal compromise solution.

[0058] The following specific embodiments illustrate the solution provided in this application.

[0059] Example 1 This embodiment provides a method for predicting and optimizing subgrade compaction parameters based on a hybrid algorithm of DBA-KAN-improved NSGA-II. For example... Figure 2 As shown, the method includes the following steps: Step 1: Constructing the sample dataset.

[0060] Based on engineering experience and theoretical analysis, the key sensitive factors affecting the compaction quality of the roadbed are identified. In this embodiment, the selected input parameters include: material state parameters (initial compaction degree, roadbed temperature), equipment input parameters (excitation force, vibration amplitude), soil property parameters (moisture content, gradation parameters), and process control parameters (rolling speed, number of rolling passes). The output parameter is the final compaction degree. ) and total construction time ( ).

[0061] Taking a test section of a highway project in Jilin Province as an example, vibration acceleration, GPS location, travel speed, and equipment status data were collected during the operation of the road roller. The compaction degree was measured using the sand cone method after each pass of compaction, resulting in 60 sets of effective compaction degree samples. The collected raw data were cleaned to remove outliers caused by starting, stopping, and turning, and the data was normalized to construct the final sample dataset. Table 1 shows the distribution of sample data in this application example. The dataset was randomly divided into a training set and a test set in an 8:2 ratio.

[0062] Table 1. Sample Data Distribution Table Step 2: Construct the DBA-KAN prediction model.

[0063] This step aims to establish a highly accurate and interpretable prediction model for final compaction degree (K) and compaction time (t).

[0064] (1) KAN network structure initialization: First, determine the structure of the KAN network. Based on the input feature dimension (8 features) and output dimension (2 features), set the initial number of network layers (e.g., 3 layers: input layer, hidden layer, output layer) and the number of neurons in each layer. A core characteristic of the KAN network is that its edge activation functions are learnable univariate functions, typically represented by B-spline curves. The structural parameters that need optimization include: the number of network layers L and the number of neurons in each layer. The order of the B-spline function and the number of grids The input-output relationship of its network layers is as follows: The transformation relationship from the l-th layer to the (l+1)-th layer can be expressed as: (2) Optimization of KAN structural parameters based on DBA: The Detective Behavior Algorithm (DBA) is introduced to optimize the structural parameters of the KAN network. DBA achieves optimization by simulating the detective's investigation process. Encoding and Initialization: The structural parameters to be optimized are encoded into individual detectives, such as... Initialize the detective population, setting parameters such as the clue collection radius and suspect screening threshold.

[0065] Clue collection phase: Each detective conducts an initial search in the solution space, records its access location and fitness value (i.e., the negative value of the cross-validation RMSE of the KAN model on the validation set), forming a clue database.

[0066] Suspect screening phase: Candidate solutions are ranked based on their fitness values ​​in the clue database, and the solutions with the highest fitness values ​​are selected. The area was designated as a "suspect" area and marked as a key search target.

[0067] Evidence analysis phase: A refined search is conducted within the suspect's area, new solutions are generated through local perturbation mechanisms, and dense sampling is performed to approximate the local optimum.

[0068] Collaborative crime-solving phase: Different detectives exchange optimal clues through information sharing mechanisms, update their respective key search areas, and coordinate to adjust search strategies to avoid individuals getting stuck in local optima.

[0069] Iteration Termination: When the maximum number of iterations is reached or the fitness value converges, the optimal detective individual is output, which is the optimal combination of KAN structure parameters.

[0070] (3) Establish the DBA-KAN prediction model: Using the optimal structural parameters obtained through optimization, the final KAN network is constructed and fully trained on the entire training set to obtain the DBA-KAN prediction model. This model can output the final compaction degree prediction value given the input parameters. And the predicted compaction time .

[0071] (4) Model validation: Introducing root mean square error (RMSE), mean absolute error (MAE), and goodness of fit (R²). 2 Three metrics are used to evaluate the model's predictive performance. RMSE is the square root of the deviation between the predicted and observed values. The smaller the value, the higher the model's prediction accuracy, and the more consistent the physical dimensions of the result with the predicted target, making it easier to interpret. MAE calculates the average absolute error between the predicted and observed values; a smaller value also indicates a smaller prediction error. 2 R² measures the proportion of the variance in the target variable explained by the model, ranging from 0 to 1. A value closer to 1 indicates a better model fit. RMSE and MAE measure model accuracy from the perspective of the absolute magnitude of the prediction error, while R² evaluates the relative goodness of fit. Using all three together provides a more comprehensive assessment of model performance. The formulas for calculating the three evaluation metrics are as follows: in, For sample size, For the true value, For predicted values, This represents the average true value.

[0072] The trained DBA-KAN model was validated using a test set. Figure 3 This is a graph showing the predicted results for the final compaction test set. Figure 4 This is a graph showing the final compaction degree training set fitting results. Figure 5 This is a graph showing the predicted results for the compaction time test set. Figure 6 The image shows the fitting results for the compaction time training set. The results indicate that the final compaction degree prediction model has a high coefficient of determination on the validation set. Achieving an accuracy of over 0.92 and keeping the root mean square error (RMSE) below 0.4% indicates that the model has good predictive accuracy. Furthermore, by visualizing the shape of the univariate function learned by the KAN network, the influence of each input parameter on the output can be intuitively analyzed, verifying the model's interpretability.

[0073] Step 3: Define the multi-objective optimization problem.

[0074] This step formalizes the optimization problem of compaction construction parameters into a multi-objective optimization problem.

[0075] The decision variable vector is: ,in Let be the compaction speed of the i-th pass (a continuous variable). For the first Vibration patterns (discrete variables, such as "low vibration and low amplitude", "high vibration and high amplitude", etc.). The total number of rolling passes (an integer variable).

[0076] There are two optimization objectives: Quality objective: Maximize final compaction degree .

[0077] in, It is calculated by the DBA-KAN prediction model trained in step two.

[0078] Efficiency objective: Minimize total construction time .

[0079] in, The length of the compaction section (can be set to a fixed value of 200m). This is for auxiliary time such as turning (30 seconds can be taken as an empirical value).

[0080] The constraints on the above construction parameters include the range of values ​​for each decision variable, such as speed. km / h, vibration mode (corresponding to four different combinations), number of passes wait.

[0081] Step 4: Solve the multi-objective optimization problem based on the improved NSGA-II.

[0082] This step employs an improved second-generation non-dominated sorting genetic algorithm (improved NSGA-II) to solve the multi-objective optimization problem defined in step three. The algorithm flow is as follows: (1) Improved population initialization: A two-level encoding mechanism is used to represent each individual. The first layer of sequence encoding: the length is the total number of compaction passes, and the encoding is an index sequence of the combination of compaction speed and vibration mode for each pass, using real number encoding; The second layer of pattern encoding records the type of vibration mode for each cycle. The vibration mode value is {10, 1, 2, 3}, which corresponds to four different combinations (such as low vibration and low amplitude, low vibration and high amplitude, high vibration and low amplitude, and high vibration and high amplitude).

[0083] Finiteness verification: Check whether the combination of construction parameters corresponding to this individual test box meets all constraints (such as speed range, vibration mode validity, total number of passes limit, etc.). If not, regenerate. Repeat the above mechanism to ensure that all individuals in the initial population are feasible solutions and have diversity.

[0084] (2) Improved crossover operation: A hybrid crossover strategy is adopted. For each pair of parent chromosomes, if the two chromosomes are completely identical, a mutation operation is introduced into the pair of chromosomes; otherwise, a single-point matching crossover is performed. The specific steps are as follows: Region division: Randomly select the crossover point to divide the two parent chromosomes into a matching region (before the crossover point) and a non-matching region (after the crossover point); Matching region determination: Compare whether the genes in the matching regions of the two parent chromosomes are completely identical. If they are not completely identical, perform single-point matching crossover, exchanging the sequence genes and pattern genes in the non-matching regions; if they are completely identical, introduce mutation operations and process them according to subsequent mutation rules to increase population diversity.

[0085] Repetitive gene adjustment: After crossover, if a gene that is repeated in the matching region appears in the non-matching region, a matching relationship is established and the gene is exchanged to ensure the integrity of the sequence. Validation and repeatability verification: After crossover, the offspring individuals are validated for validity and repeatability. If they fail, a new parent is selected for crossover.

[0086] (3) Improved mutation operation: A multi-point mutation strategy is adopted, and the specific steps are as follows: Selecting mutation sites: Randomly select approximately one-third of the gene sites as flattened locations; Sequence exchange: Randomly swaps genes at selected mutation sites to generate a new sequence order; Pattern regeneration: Randomly regenerating the vibrational model genes corresponding to the mutation sites; Constraint correction: For operating conditions with special constraints (such as prohibition of certain vibration modes), the mode genes are corrected according to the constraint rules; Validation and repeatability verification: After mutation is completed, the offspring individuals are subjected to validity and repeatability verification. If they fail, new individuals are selected for mutation.

[0087] 4) Classic NSGA-II core framework After completing population initialization, crossover, and mutation, the core evolutionary mechanism of the classic NSGA-II is used: Fast non-dominated sorting: Non-dominated stratification of the population, assigning individuals to different frontiers; Crowding distance calculation: Calculate the crowding distance for individuals within the same level to maintain the uniformity of solution distribution; Selection operator: A tournament selection strategy is adopted to select parent individuals based on non-dominated hierarchy and crowding degree; Elite strategy: Merge parent and offspring populations, and select the best individuals to form the next generation population through non-dominated ranking and crowding comparison.

[0088] 5) Iteration Termination and Output When the maximum number of iterations is reached or the population converges, output the non-dominated solution set in the final population, i.e., the Pareto optimal solution set. Otherwise, repeat steps (2)-(4). Figure 7 An example of Pareto frontier diagramming is shown.

[0089] Step 5: Selection and recommendation of the optimal compromise solution.

[0090] To select the unique optimal construction scheme from the Pareto solution set, a fuzzy decision-making method is adopted.

[0091] For each solution in the Pareto solution set For each target ( ), calculate membership degree based on its optimization direction .

[0092] For the target that needs to be minimized (such as...) The membership function is: For the goal that needs to be maximized (such as...) The membership function is: Then, calculate the overall dominance value for each solution, using the following formula: The solution with the largest overall dominance value is selected as the optimal compromise solution. The decision variable values ​​corresponding to this solution are the final recommended combination of roadbed compaction construction parameters for the site. Figure 8 This is a comparison chart showing the results before and after the final compaction optimization. Figure 9 The figure shows a comparison of the results before and after the optimization of the compaction time. The results indicate that the optimized combination of construction parameters significantly shortened the construction time while ensuring compaction quality, thus verifying the effectiveness of this application.

[0093] This application belongs to the field of intelligent compaction technology for highway subgrades, specifically disclosing a method, system, and equipment for predicting and optimizing subgrade compaction parameters based on a hybrid DBA-KAN-improved NSGA-II algorithm. The method includes: analyzing sensitive factors affecting subgrade compaction quality and constructing a sample dataset containing construction parameters and compaction quality indicators; based on the sample dataset, establishing a prediction model based on the Detective Behavior (DBA) algorithm to optimize the Kolmogorov-Arnold network (KAN) for predicting the final compaction degree and total construction time; using the trained DBA-KAN model as a surrogate model for the objective function to construct a multi-objective optimization model with the objectives of maximizing the final compaction degree and minimizing the total construction time; and using an improved second-generation non-dominated sorting genetic algorithm to improve the NSGA-II algorithm to solve the multi-objective optimization model, obtaining the Pareto optimal solution set, and recommending the optimal combination of construction parameters to the user. This invention improves model interpretability by introducing a KAN network, and combines the global optimization capability of DBA with the efficient multi-objective optimization capability of improved NSGA-II to achieve high-precision prediction of subgrade compaction quality and efficiency, and can provide scientific suggestions for optimizing construction parameters, thus significantly improving the decision-making level of intelligent compaction.

[0094] Figure 10 The system architecture diagram for determining roadbed compaction construction parameters provided in the embodiments of this application is as follows: Figure 10 As shown, it includes: The sample data acquisition module 1010 is used to acquire a sample dataset of roadbed compaction; the sample dataset includes multiple sample data, each sample data includes construction parameters for a single pass of compaction and the corresponding compaction degree increase value and construction time; The prediction model training module 1020 is used to take the construction parameters in each sample data as the input vector of KAN, the compaction increase value and construction time as the output vector of KAN, optimize and train KAN based on DBA, and use the trained KAN as the prediction model; the prediction model is used to predict the final compaction degree and total construction time based on the construction parameters of each rolling pass. The multi-objective optimization model construction module 1030 is used to construct a multi-objective optimization model with the prediction model as a proxy model, taking maximizing the final compaction degree as the first optimization objective and minimizing the total construction time as the second optimization objective. The construction parameter solution module 1040 is used to solve the multi-objective optimization model using the second-generation non-dominated sorting genetic algorithm NSGA-II to obtain the Pareto optimal solution set; the Pareto optimal solution set includes the optimal combination of construction parameters under different trade-offs between the first optimization objective and the second optimization objective.

[0095] It should be understood that the above system is used to execute the methods in the above embodiments. The corresponding program modules in the system are similar in implementation principle and technical effect to those described in the above methods. The working process of the system can be referred to the corresponding process in the above methods, and will not be repeated here.

[0096] Figure 10 The roadbed compaction construction parameter determination system shown is only illustrated by the division of the above functional modules when processing data. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the roadbed compaction construction parameter determination system can be divided into different functional modules to complete all or part of the functions described above.

[0097] Based on the methods in the above embodiments, this application provides an electronic device, such as... Figure 11 As shown, the electronic device may include a processor 1110, a communication interface 1120, a memory 1130, and a communication bus 1140, wherein the processor 1110, the communication interface 1120, and the memory 1130 communicate with each other through the communication bus 1140. The processor 1110 can call logical instructions in the memory 1130 to execute the methods in the above embodiments.

[0098] Furthermore, the memory 1130 is coupled to the one or more processors 1110, the memory 1130 being used to store computer program code, the computer program code including computer instructions; the one or more processors 1110 invoke the computer instructions to cause the electronic device to perform the method described in the above method embodiments.

[0099] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0100] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0101] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0102] It is understood that the processor in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0103] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0104] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0105] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for determining roadbed compaction construction parameters, characterized in that, Includes the following steps: Obtain a sample dataset of roadbed compaction; the sample dataset includes multiple sample data, each of which includes construction parameters for a single pass of compaction and the corresponding increase in compaction degree and construction time; The construction parameters in each sample data are used as the input vector of the Kolmogorov-Arnold network (KAN), and the compaction increase and construction time are used as the output vector of the KAN. The KAN is trained based on the Detective Behavior Algorithm (DBA), and the trained KAN is used as the prediction model. The prediction model is used to predict the final compaction and total construction time based on the construction parameters of each rolling pass. With maximizing the final compaction degree as the first optimization objective and minimizing the total construction time as the second optimization objective, a multi-objective optimization model is constructed using the prediction model as a surrogate model. The second-generation non-dominated sorting genetic algorithm NSGA-II is used to solve the multi-objective optimization model to obtain the Pareto optimal solution set; the Pareto optimal solution set includes the optimal combination of construction parameters under different trade-offs between the first optimization objective and the second optimization objective.

2. The method for determining roadbed compaction construction parameters according to claim 1, characterized in that, Also includes: The optimal compromise solution is selected from the Pareto optimal solution set using the fuzzy decision method, and is used as the final recommended subgrade compaction construction parameters.

3. The method for determining roadbed compaction construction parameters according to claim 1, characterized in that, The DBA-based optimized training of KAN includes: Based on the dimensions of the input vector and the output vector, set the initial number of network layers and the number of neurons in each layer of the KAN; and determine the structural parameters of the KAN to be optimized. The structural parameters to be optimized are encoded into individual detectives, and the detective population is initialized. Each individual detective performs a preliminary search in the solution space, records the structural parameter combinations and fitness values ​​they attempt, and forms a clue database; the fitness value refers to the negative root mean square error (RMSE) obtained by cross-validation of KAN on the validation set under the corresponding structural parameters; Based on the fitness value of each detective in the clue database, the candidate detectives are sorted, and the top few detectives with the highest fitness values ​​are selected and marked as the suspect area. A refined search is conducted within the suspect area, new detective individuals are generated through a local perturbation mechanism, and dense sampling is performed to approximate the local optimum. By exchanging optimal clues through information sharing mechanisms among different detectives, updating their respective key search areas, and coordinating to adjust search strategies, individuals can avoid getting trapped in local optima. When the maximum number of iterations is reached or the fitness value converges, the optimal detective individual is output; The prediction model is obtained by taking the structural parameters corresponding to the optimal detective individual as the optimal KAN structural parameters.

4. The method for determining roadbed compaction construction parameters according to claim 1, characterized in that, First optimization objective for: in, Initial compaction degree, The number of rolling passes, For the first The increase in compaction degree per pass is calculated and predicted by the prediction model. For the first Construction parameters for roadbed compaction; Second optimization objective for: in, The length of a single compaction pass. For the first The rolling speed of the whole process To supplement the work time.

5. The method for determining roadbed compaction construction parameters according to claim 1, characterized in that, The multi-objective optimization model was solved using NSGA-II to obtain the Pareto optimal solution set, including: Step S1: A two-layer coding mechanism is used to generate an initial population; the first layer of the two-layer coding mechanism is the construction parameter sequence coding, and the second layer is the vibration mode coding; each set of codes in the initial population corresponds to a chromosome. Step S2: Using a hybrid crossover strategy, for each pair of parent chromosomes, if the two chromosomes are completely identical, a mutation operation is introduced into the pair of chromosomes; otherwise, a single-point matching crossover is performed. Step S3: Using a multi-point mutation strategy, approximately one-third of the genes are randomly selected for position exchange, and the corresponding vibrational pattern genes are regenerated to obtain the parent population. Step S4: Select an individual population from the parent population through a tournament, and generate the offspring population through simulated binary crossover and polynomial mutation. Step S5: Merge the parent population and the offspring population, perform non-dominated sorting on the merged population, and select the best multiple individuals to form the next generation parent population according to the principles of hierarchy priority and crowding priority. Step S6: Determine whether the maximum number of iterations has been reached or the convergence condition has been met. If the maximum number of iterations has not been reached or the convergence condition has not been met, return to step S2. If the maximum number of iterations has been reached or the convergence condition has been met, output the non-dominated solution set in the final population, i.e., the Pareto optimal solution set.

6. The method for determining roadbed compaction construction parameters according to claim 5, characterized in that, In steps S1-S3, after generating the initial population, after mixing and crossover, and / or after multi-point mutation, each individual is validated for validity and repeatability to check whether the constraints of the preset construction parameters are met and to ensure that there are no duplicate individuals in the population.

7. The method for determining roadbed compaction construction parameters according to claim 2, characterized in that, The step of selecting the optimal compromise solution from the Pareto optimal solution set using the fuzzy decision-making method includes: For each solution in the Pareto solution set, calculate the corresponding first membership degree and second membership degree for the first and second optimization objectives; wherein, the larger the compaction degree corresponding to each solution, the larger the first membership degree, and the smaller the corresponding total construction time, the larger the second membership degree. The sum of the first and second membership degrees of each solution is taken as its total membership degree; The ratio of the total membership degree of each solution to the sum of the total membership degrees of all solutions is taken as its overall dominance value; The solution with the largest overall dominance value is selected as the optimal compromise solution.

8. The method for determining roadbed compaction construction parameters according to claim 7, characterized in that, First membership degree for: in, Let J be the compaction degree corresponding to the j-th solution. The preset lower limit of compaction. This is the preset upper limit of compaction. Second membership degree for: in, Let j be the total construction time corresponding to the j-th solution. This is the preset lower limit of the total construction time. This is the preset maximum total construction time. The overall dominance value of solution j for: Where n is the total number of solutions in the Pareto optimal solution set.

9. A system for determining roadbed compaction construction parameters, characterized in that, include: The sample data acquisition module is used to acquire a sample dataset of roadbed compaction; the sample dataset includes multiple sample data, each of which includes construction parameters for a single pass of compaction and the corresponding compaction degree increase value and construction time; The prediction model training module is used to take the construction parameters in each sample data as the input vector of the Kolmogorov-Arnold network (KAN), the compaction increase value and the construction time as the output vector of the KAN, optimize and train the KAN based on the Detective Behavior Algorithm (DBA), and use the trained KAN as the prediction model; the prediction model is used to predict the final compaction degree and total construction time based on the construction parameters of each rolling pass. A multi-objective optimization model construction module is used to construct a multi-objective optimization model with the prediction model as a surrogate model, taking maximizing the final compaction degree as the first optimization objective and minimizing the total construction time as the second optimization objective. The construction parameter solution module is used to solve the multi-objective optimization model using the second-generation non-dominated sorting genetic algorithm NSGA-II to obtain the Pareto optimal solution set; the Pareto optimal solution set includes the optimal combination of construction parameters under different trade-offs between the first optimization objective and the second optimization objective.

10. An electronic device, characterized in that, Includes memory and one or more processors; The memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions; The one or more processors invoke the computer instructions to cause the electronic device to perform the method as described in any one of claims 1-8.

Images