A method and device for intelligent compaction optimization of highway subgrade

Abstract

The application belongs to the technical field of intelligent design of highway construction, and particularly discloses a method and equipment for intelligent compaction optimization of highway subgrade. The method comprises the following steps: taking the compaction degree increase value of the highway subgrade and the compaction time of the highway subgrade as the optimization target, collecting a multi-source data set in real time by using the Internet of Things, extracting frequency domain features, and identifying key parameters; using IBKA to capture the nonlinear mapping relationship between the optimization target and the key parameters to obtain the fitness function of the optimization target; using the EMOEAD combined with engineering data clustering and Logic chaotic mapping enhancement and TOPSIS to perform intelligent compaction optimization control of the highway subgrade, and realizing the natural language interaction support of compaction degree cloud map visualization, parameter dynamic adjustment suggestion and intelligent compaction quality problem diagnosis. The application realizes the intelligent compaction prediction and optimization of the highway subgrade based on a large language model, and provides a systematic solution for the intelligent and fine compaction of the highway subgrade.

Description

Technical Field

[0001] This invention belongs to the field of intelligent design technology for highway construction, and more specifically, relates to a method and equipment for intelligent compaction optimization of highway subgrade. Background Technology

[0002] The compaction quality of highway subgrade directly affects the bearing capacity and long-term service performance of roads. Traditional quality control mainly relies on point-based, destructive sampling inspections such as sand cone and ring cutter methods, which have inherent defects such as limited coverage, delayed results, and inability to guide real-time construction. Intelligent compaction technology, by integrating vibration sensing, high-precision positioning, and the Internet of Things, achieves continuous monitoring of compaction process parameters and indirectly assesses the compaction state using indicators such as harmonic ratio and compaction meter readings. However, current technology still faces significant bottlenecks: First, the correlation between core evaluation indicators and the actual degree of compaction is affected by multiple factors such as filler type and moisture content, resulting in insufficient universality and stability under complex working conditions; second, although the system can collect multi-source data, it lacks deep integration and deep mining capabilities, making it difficult to build high-precision adaptive prediction models; finally, existing intelligent compaction platforms are mostly in the "monitoring-display" stage, relying on human experience for decision-making and adjustments, with limited intelligence levels, and unable to achieve autonomous real-time optimization of compaction process parameters.

[0003] In recent years, machine learning methods have been introduced to improve the accuracy of compaction quality prediction. However, traditional machine learning models suffer from problems such as complex feature engineering, weak model interpretability, and poor adaptability to new working conditions and small sample scenarios when dealing with high-dimensional and heterogeneous construction time-series data. In addition, these models usually exist as independent "black box" modules, making it difficult to deeply integrate with domain knowledge, engineering specifications, and natural language interaction, thus limiting their application in complex decision support.

[0004] With breakthroughs in artificial intelligence technology, large language models have demonstrated powerful capabilities in natural language understanding, context learning, complex reasoning, and unified multi-task modeling. In the engineering field, combined with retrieval-enhanced generative techniques, they can construct domain-specific knowledge bases, enabling intelligent question answering and decision interpretation. This presents a revolutionary opportunity for intelligent compaction technology: by integrating large language models, it is expected to deeply fuse compaction mechanism knowledge, multi-source sensor information, and historical engineering data to construct a next-generation intelligent compaction system capable of interpretable reasoning, few-sample adaptive learning, and providing compaction parameter optimization schemes and decision support through natural language interaction. Therefore, exploring intelligent compaction optimization methods based on large language models is a key direction for breaking through current technological bottlenecks and promoting the advancement of roadbed construction towards cognitive intelligence. Summary of the Invention

[0005] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a method and equipment for intelligent compaction optimization of highway subgrades. Combining the characteristics of intelligent compaction construction of highway subgrades and its multivariable, complex, and nonlinear process features, a corresponding intelligent compaction optimization method for highway subgrades is designed. Key aspects such as feature extraction from multi-source IoT data, the prediction model based on IBKA-NGBoost, and the algorithm architecture and specific optimization settings of the multi-objective optimization model based on IBKA-NGBoost-EMOEAD are studied and designed. This effectively solves the problems of difficulty in accurately balancing compaction quality and construction time in traditional compaction construction, and the large errors caused by relying on experience-based parameter settings. Furthermore, it has the advantages of scientifically guiding and proposing intelligent compaction parameter setting ranges, and achieving effective control of compaction efficiency and cost. Therefore, it is particularly suitable for applications requiring high-quality, high-efficiency intelligent compaction construction of highway subgrades.

[0006] To achieve the above objectives, according to one aspect of the present invention, a method for intelligent compaction optimization of highway subgrade is proposed, comprising the following steps: Step 1: Using the increase in roadbed compaction degree and roadbed compaction time as optimization targets, the Internet of Things is used to collect multi-source datasets in real time, extract frequency domain features, identify key parameters, and construct a sample dataset based on the key parameters and the corresponding increase in roadbed compaction degree and roadbed compaction time. Step 2: Based on the sample dataset, establish a prediction model for the increase in roadbed compaction degree and roadbed compaction time based on IBKA-NGBoost, in order to obtain the nonlinear mapping relationship between the optimization objective and key parameters; Step 3: Using the nonlinear mapping relationship as the objective optimization fitness function and the constraint range of the key parameters as the constraint function, construct a multi-objective optimization model based on IBKA-NGBoost-EMOEAD to obtain the optimal solution of the objective function, and propose a suggested range of values ​​for intelligent compaction parameters of highway subgrade.

[0007] As a further preferred embodiment, in step one, the key parameters include: initial compaction degree of the roadbed. Intelligent compaction vibration force for highway subgrade Intelligent compaction vibration amplitude of highway subgrade Highway subgrade temperature Highway subgrade intelligent compaction pass count Final compaction degree of highway subgrade Intelligent compaction speed of highway subgrade .

[0008] As a further preferred option, step two specifically includes the following steps: (21) The sample dataset is preprocessed and then divided into training dataset and test dataset; (22) Based on the training dataset, IBKA is used to optimize the key hyperparameters of NGBoost, the optimal hyperparameter combination is selected, an IBKA-NGBoost prediction model is built based on the optimal hyperparameter combination, and the IBKA-NGBoost prediction model is tested using the test dataset. (23) Based on the IBKA-NGBoost prediction model, the predicted values ​​and the true values ​​are fitted to obtain the nonlinear mapping relationship between the optimization target and the key parameters.

[0009] As a further preferred embodiment, the optimization of key hyperparameters of NGBoost using IBKA in step (22) includes the following steps: (221) Population initialization based on the optimal point set strategy: The optimal point set strategy is introduced to solve the problems of uneven initial population distribution and insufficient diversity caused by random initialization. The population initialization can be expressed as: in, This represents the position of the i-th individual in the j-th dimension. Indicates the upper bound. Indicates the lower bound. This represents the optimal point set value in the j-th dimension, and k represents the index number, i.e., the current position when the point set was generated; (222) Based on the improved predation stage of the golden sine algorithm, a golden sine optimization strategy is introduced to overcome the defect of the original black kite algorithm being prone to getting trapped in local optima. Its predation stage can be expressed as: in, and These represent the positions of the i-th individual in the j-th dimension at time t+1 and time t, respectively. This represents the perturbation factor that decreases over time. This represents a random number that is uniformly distributed within the interval [0,1]. and This represents the random sinusoidal perturbation factor. Indicates the current optimal global position. This indicates the adjustment factor related to the golden ratio. Represents a fixed value; (223) When population evolution stagnates, a differential evolution mechanism is introduced to perform guided population updates, including mutation, crossover and selection operations.

[0010] As a further preferred embodiment, the mutation operation includes: Three distinct individuals are randomly selected from the current population, and a mutant individual is generated through mutation, represented as: in, Indicates a variant individual. This represents three distinct individuals randomly selected from the current population. Indicates the scaling factor; Preferably, the crossover operation includes: The mutant individuals are cross-mixed with the original individuals to generate experimental individuals, represented as follows: in, This represents the new experimental individual after the crossover operation. Indicates the crossover probability. This represents a randomly generated integer index within the range [1, pop], where pop is the population size; Preferably, the selection operation includes: A greedy strategy is adopted, comparing the fitness of experimental individuals with that of the original individuals, and selecting individuals with better fitness for the next generation of the population, as follows: in, This refers to the next generation of individuals selected from among the previous generation. This represents the fitness value.

[0011] As a further preferred embodiment, in step three, the objective function is determined by predicting the nonlinear mapping relationship function using IBKA-NGBoost. , They are respectively: in, These are key parameters, i = 1 to 7, specifically... This indicates the initial compaction degree of the highway subgrade. This indicates the intelligent compaction excitation force of the highway subgrade. Indicates the amplitude of intelligent compaction vibration of highway subgrade, Indicates the temperature of the roadbed, Indicates the number of intelligent compaction passes for highway subgrade. Indicates the final compaction degree of the highway subgrade, Indicates the intelligent compaction speed of highway subgrade; Preferably, the constraint function includes: in, The objective function for the increase in roadbed compaction degree based on IBKA-NGBoost is: The objective function for highway subgrade compaction time based on IBKA-NGBoost.

[0012] As a further optimization, in step three, the EMOEAD algorithm is used to obtain the Pareto optimal solution, as follows: (31) Set the population size, neighborhood size, crossover mutation probability, maximum iteration time and frontier solution set, initialize the weight vector, and group them into neighborhoods by calculating the Euclidean distance between each weight vector; (32) Perform cluster analysis on the sample dataset and select areas with large increases in roadbed compaction degree and short roadbed compaction time; (33) Calculate the average value of the decision variables in the selection area, generate the initial solution using the chaotic mapping, calculate the objective function value of all individuals, generate the initial ideal point, and select non-dominated individuals to update the frontier solution set; (34) Select two individuals from the current population, perform crossover and mutation strategies, generate new offspring individuals after crossover and mutation, and update the neighborhood by calculating the aggregation function value between the new individuals and their neighbors; (35) Calculate the aggregation function value of the neighborhood of the individual grouped according to the current neighborhood. If the aggregation function value of the new descendant individual is better, replace the domain individual and update the frontier solution set and ideal point. (36) Repeat steps (31)-(35) above until the maximum number of iterations is reached, and output the front solution set as the Pareto front set; Preferably, in step (33), in order to replace the random initialization method, an initialization sequence generated by chaotic mapping is used, and the following equations are followed: In the formula, X represents the final initial population position generated after the mapping, lb represents the lower bound of the search space, ub represents the upper bound of the search space, chaos(N, dim) represents the chaotic sequence matrix generated by the chaotic mapping, N represents the initial population size, and dim represents the spatial dimension of the optimization problem. This represents the value of the chaotic variable at the nth iteration. Represents the result calculated iteratively. A chaotic sequence value, Represents the control parameters for chaotic mapping.

[0013] According to another aspect of the present invention, a smart compaction optimization system for highway subgrade is also provided, comprising: The first main control module takes the increase in roadbed compaction degree and roadbed compaction time as optimization targets, uses the Internet of Things to collect multi-source datasets in real time, extracts frequency domain features, identifies key parameters, and constructs a sample dataset based on the key parameters and the corresponding increase in roadbed compaction degree and roadbed compaction time. The second main control module, based on the sample dataset, establishes a prediction model for the increase in roadbed compaction degree and roadbed compaction time based on IBKA-NGBoost, in order to obtain the nonlinear mapping relationship between the optimization objective and key parameters; The third main control module uses the nonlinear mapping relationship as the objective optimization fitness function and the constraint range of key parameters as the constraint function to construct a multi-objective optimization model based on IBKA-NGBoost-EMOEAD to obtain the optimal solution of the objective function, and proposes a suggested range of values ​​for intelligent compaction parameters of highway subgrade.

[0014] According to another aspect of the invention, an electronic device is also provided, comprising: At least one processor, at least one memory, and a communication interface; wherein, The processor, memory, and communication interface communicate with each other; The memory stores program instructions that can be executed by the processor. The processor calls the program instructions to execute any of the above embodiments or a combination of the above embodiments for a method of intelligent compaction optimization of highway subgrade.

[0015] According to another aspect of the present invention, a non-transitory computer-readable storage medium is also provided, the non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute any of the above embodiments or a combination of the above embodiments for a method for intelligent compaction optimization of highway subgrade.

[0016] In summary, compared with existing technologies, the technical solutions conceived in this invention mainly possess the following technical advantages: This invention, combining the characteristics of intelligent compaction of highway subgrades and its optimized process features, designs a hybrid machine learning method based on a large language model and incorporating the IBKA-NGBoost-EMOEAD algorithm to solve the optimization problem of intelligent compaction of highway subgrades considering the actual construction conditions. The IBKA-NGBoost model is used to simulate and train the monitored sample data, providing a high-precision prediction function for EMOEAD optimization. Then, EMOEAD is used for multi-objective optimization to obtain the optimal range of intelligent compaction parameters for highway subgrades, achieving the goal of optimizing intelligent compaction parameters for highway subgrades. Specifically: 1. This invention fully utilizes the advantages of NGBoost in providing probabilistic prediction and handling data heteroscedasticity. Based on construction time series data, a regression function is constructed to enhance the robustness and interpretability of the model under complex working conditions. It achieves accurate prediction of the relationship between the increase in roadbed compaction degree and the time of roadbed compaction, providing a reliable foundation for the intelligent optimization of roadbed compaction parameters.

[0017] 2. This invention utilizes the EMOEAD algorithm, which combines engineering data clustering and logic chaotic mapping enhancement, to establish a multi-objective optimization model. Through improved initialization and search mechanisms, this algorithm can more quickly and stably approach the Pareto front, thereby improving the convergence speed and optimization accuracy in multi-objective optimization problems of compaction parameters.

[0018] 3. This invention replaces the traditional mathematical relationship with the NGBoost prediction model optimized by IBKA, using it as the fitness function for multi-objective optimization, and deeply integrates it with the EMOEAD algorithm. It accurately reflects the complex nonlinear mapping relationship between parameters such as the initial compaction degree of highway subgrade, the intelligent compaction excitation force of highway subgrade, and the intelligent compaction rolling speed of highway subgrade, and the compaction degree and compaction efficiency of highway subgrade, providing a core driving force for the precise optimization and control of compaction technology.

[0019] 4. This invention uses both the increase in roadbed compaction degree and the roadbed compaction time as optimization objectives to perform multi-objective collaborative optimization. This effectively balances the contradictory relationship between compaction quality and construction efficiency, so that the final optimized solution significantly improves construction efficiency while ensuring project quality, and is more in line with the economic and timeliness requirements of actual engineering projects.

[0020] 5. This invention innovatively develops a large language model platform that integrates prediction models, optimization algorithms, and engineering specification knowledge bases. It not only realizes the visualization of compaction cloud maps and dynamic parameter adjustment suggestions, but also provides intelligent compaction quality problem diagnosis and rectification support through natural language interaction, providing a systematic solution for intelligent, refined, and knowledge-based decision-making in intelligent compaction operations of highway subgrades. Attached Figure Description

[0021] Figure 1 This is a flowchart of an intelligent compaction optimization method for highway subgrade according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a smart compaction optimization system for highway subgrade according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the physical structure of the electronic device provided in the embodiment of the present invention; Figure 4This is a graph showing the fitting results of the training set for predicting the increase in roadbed compaction using the IBKA-NGBoost method in this embodiment of the invention. Figure 5 This is a test set prediction result diagram of the IBKA-NGBoost prediction of the increase in roadbed compaction in an embodiment of the present invention; Figure 6 This is a figure showing the fitting results of the training set for predicting highway subgrade compaction time using IBKA-NGBoost in an embodiment of the present invention; Figure 7 This is a graph showing the prediction results of the IBKA-NGBoost test set for predicting highway subgrade compaction time in an embodiment of the present invention. Figure 8 This is a diagram showing the optimization results of the IBKA-NGBoost-EMOEAD optimization of the increase in roadbed compaction in an embodiment of the present invention. Figure 9 This is a diagram showing the optimization results of highway subgrade compaction time using the IBKA-NGBoost-EMOEAD method in an embodiment of the present invention. Figure 10 This is a schematic diagram of the large language model platform in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0023] This invention discloses an intelligent compaction optimization method for highway subgrades. It uses the increase in subgrade compaction degree and compaction time as optimization objectives. The method utilizes the Internet of Things (IoT) to collect multi-source datasets in real time, extracts frequency domain features, and identifies key parameters. It employs IBKA (Improved Black Kite Algorithm)-NGBoost (Natural Gradient Boosting) to capture the nonlinear mapping relationship between the optimization objectives and key parameters, deriving the fitness function of the optimization objectives. It utilizes EMOEAD (Enhanced Multi-objective Evolutionary Algorithm Based on Decomposition) and TOPSIS (Technique for Order Preference by Similarity to an Ideal Solution), which combines engineering data clustering and logic chaotic mapping, for intelligent compaction optimization control of highway subgrades. A large language model platform integrating prediction, optimization, and specification knowledge bases is developed to achieve natural language interactive support for compaction degree cloud map visualization, dynamic parameter adjustment suggestions, and intelligent compaction quality problem diagnosis. This invention realizes intelligent compaction prediction and optimization of highway subgrades based on a large language model, providing a systematic solution for intelligent and refined compaction of highway subgrades.

[0024] Example 1 like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for intelligent compaction optimization of highway subgrade, comprising the following steps: Step 1: Using the increase in roadbed compaction degree and roadbed compaction time as optimization objectives, the system utilizes the Internet of Things (IoT) to collect multi-source datasets in real time, extract frequency domain features, and identify key parameters. More specifically, based on the overall optimization concept of intelligent roadbed compaction, the system uses the increase in roadbed compaction degree and roadbed compaction time as optimization objectives, utilizes the IoT to collect multi-source datasets in real time, extract frequency domain features, and identify key parameters. Under the construction disturbance of intelligent roadbed compaction, the sensitive factors causing the increase in roadbed compaction degree and the time between roadbed compaction times include the initial compaction degree of the roadbed. Intelligent compaction vibration force for highway subgrade Intelligent compaction vibration amplitude of highway subgrade Highway subgrade temperature Highway subgrade intelligent compaction pass count Final compaction degree of highway subgrade Intelligent compaction speed of highway subgrade By selecting the above seven main influencing factors as input parameters, the optimal values ​​for the increase in roadbed compaction degree and the comprehensive optimal values ​​for roadbed compaction time under different parameter influences were obtained.

[0025] Step 2: Utilize IBKA-NGBoost to capture the nonlinear mapping relationship between the optimization objective and key parameters, derive the fitness function of the optimization objective, and use EMOEAD and TOPSIS for intelligent compaction optimization control of highway subgrade. More specifically, based on the sample dataset, perform IBKA optimization on the model and use NGBoost to learn and simulate on the training set, establishing a prediction model for the increase in highway subgrade compaction degree and highway subgrade compaction time based on IBKA-NGBoost. Test the model's prediction results using the test set.

[0026] Step two involves several steps. First, the data collected from engineering monitoring is aligned and preprocessed for missing and outlier values ​​to improve data quality. Then, IBKA is used to optimize the key hyperparameters of NGBoost, selecting the optimal hyperparameter combination and establishing an IBKA-NGBoost prediction model based on this combination. Finally, the predicted values ​​of the IBKA-NGBoost model are fitted to the actual values, and the prediction accuracy is evaluated using root mean square error, mean absolute error, and goodness of fit. The specific steps include: (1) Data preprocessing During data preprocessing, it is crucial to eliminate abnormal data from non-normal intelligent compaction operation states, such as abnormal signals generated during start-up, stopping, turning, non-vibration operation, and conditions like bouncing and voiding. By denoising, filtering, and spatiotemporally aligning the raw data, the interference of invalid data on compaction indicators can be reduced, improving the reliability of the results. After calculating the indicators, the CMV value should be integrated with the corresponding time, location, velocity, vibration state, and pass count information to form a data record that can be directly uploaded to the platform.

[0027] (2) Determination of hyperparameters of the IBKA-NGBoost prediction model In building the NGBoost prediction model, the key hyperparameters—learning rate, row sampling ratio, column sampling ratio, and number of iterations—affect the model's accuracy. The learning rate controls the magnitude of the correction each new decision tree makes to the model's predictions. A lower learning rate leads to more stable and accurate convergence but requires more iterations; an excessively high learning rate may cause convergence oscillations or overfitting. The row sampling ratio determines the proportion of random samples used to train a single tree in each iteration. Introducing this mechanism enhances the diversity among base learners, reduces model variance, and is an important regularization technique to prevent overfitting. The column sampling ratio sets the proportion of features randomly considered when splitting nodes in each tree. It further improves model robustness by limiting feature usage, avoids dependence on a few strong features, and accelerates training. The number of iterations specifies the total number of base learners. Insufficient iterations lead to underfitting and an inability to capture complex patterns, while excessive iterations cause overfitting and increase computational costs.

[0028] This invention employs the computationally efficient and easily implemented IBKA optimization algorithm to optimize the hyperparameters of the NGBoost model. The IBKA optimization process mainly consists of three parts: population initialization based on the optimal point set strategy, predation phase based on the golden sine algorithm, and migration phase based on the differential evolution algorithm.

[0029] The optimal point set strategy is introduced primarily to address the problems of uneven initial population distribution and insufficient diversity caused by random initialization. The optimal point set strategy achieves a more uniform and representative distribution of individuals within the search space, providing richer global information and effectively enhancing the initial diversity of the population. Population initialization based on the optimal point set strategy can be expressed as: in, This represents the position of the i-th individual in the j-th dimension. Indicates the upper bound. Indicates the lower bound. represents the optimal point set value in the j-th dimension, and k represents the index number, i.e., the current position when the point set was generated.

[0030] The introduction of the golden sine optimization strategy is primarily aimed at overcoming the shortcomings of the original Black Kite algorithm, such as excessive randomness and susceptibility to local optima during the predation phase. This strategy incorporates a guiding perturbation to the current global optimum into the position update process using a sine function and the golden ratio. This allows the algorithm to more intelligently balance global search and local exploitation during exploration, effectively alleviating local convergence problems and improving convergence speed and optimization accuracy. The predation phase based on the improved golden sine algorithm can be represented as: in, and These represent the positions of the i-th individual in the j-th dimension at time t+1 and time t, respectively. This represents the perturbation factor that decreases over time. This represents a random number that is uniformly distributed within the interval [0,1]. and This represents the random sinusoidal perturbation factor. Indicates the current optimal global position. This indicates the adjustment factor related to the golden ratio. It represents a fixed value.

[0031] When population evolution stalls during algorithm iteration, the random migration strategy of the original Black Kite algorithm is inefficient. Introducing differential evolution aims to perform a guided population update, systematically generating high-quality new individuals by simulating the "mutation," "crossover," and "selection" operations in biological evolution, thereby helping the population escape local optima. The migration stage improved based on differential evolution can be represented as: 1) Mutation operation: From the current population, three distinct individuals are randomly selected, and a mutated individual is generated through mutation. This can be represented as: in, Indicates a variant individual. This represents three distinct individuals randomly selected from the current population. This represents the scaling factor.

[0032] 2) Crossover operation: The mutant individuals are cross-mixed with the original individuals to generate experimental individuals, which can be represented as follows.

[0033] in, This represents the new experimental individual after the crossover operation. Indicates the crossover probability. This represents an integer index randomly generated in the range [1, pop] (where pop is the population size).

[0034] 3) Selection operation: A greedy strategy is adopted, comparing the fitness of the experimental individuals with that of the original individuals, and selecting the individuals with better fitness for the next generation of the population, which can be represented as: in, This refers to the next generation of individuals selected from among the previous generation. This represents the fitness value.

[0035] (3) Establish a training model The sample data is randomly divided into training and testing sets. The hyperparameter optimization results are then input into the model, which is built using Python.

[0036] (4) Evaluate the model performance. 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.

[0037] Step 3: Using the nonlinear mapping function obtained from the prediction model as the objective optimization fitness function, construct a multi-objective optimization model based on IBKA-NGBoost-EMOEAD, and propose a suggested range for intelligent compaction parameters of highway subgrade.

[0038] Step 3 replaces the traditional mathematical function with the IBKA-NGBoost prediction model as the fitness function for optimization. Multi-objective optimization is performed based on EMOEAD to obtain the optimal solution for the objective function, and a suggested range of values ​​for intelligent compaction parameters for highway subgrade is proposed. This includes the following specific steps: (1) Determine the objective function The IBKA-NGBoost prediction algorithm is introduced to replace the traditional mathematical relation as the fitness function in the multi-objective genetic algorithm. This clearly reflects the complex nonlinear relationship between the increase in roadbed compaction degree and roadbed compaction time under intelligent compaction construction disturbance. The objective function is determined by the nonlinear mapping relationship function predicted by IBKA-NGBoost. , They are respectively: in (i=1~7) represents the initial compaction degree of the highway subgrade. Intelligent compaction vibration force for highway subgrade Intelligent compaction vibration amplitude of highway subgrade Highway subgrade temperature Highway subgrade intelligent compaction pass count Final compaction degree of highway subgrade Intelligent compaction speed of highway subgrade .

[0039] (2) Determine the range of constraints To ensure the practical significance of optimizing intelligent compaction of highway subgrades and to guarantee that the resulting compaction parameter settings are more reasonable, feasible, and free from safety hazards, it is necessary to set constraints on the values ​​of the target decision parameters based on engineering requirements and existing project data. These constraints are generally expressed as follows: in, and These represent the lower and upper limits of the input parameter values, respectively.

[0040] (3) EMOEAD multi-objective optimization and obtaining the optimal solution The EMOEAD algorithm is applied to achieve multi-objective optimization of intelligent compaction parameters for highway subgrade, so as to determine the Pareto optimal solution set of intelligent compaction parameters for highway subgrade, and to ensure effective control of efficiency and cost between the increase in highway subgrade compaction degree and the time of highway subgrade compaction. Step (33) The optimization process of the EMOEAD algorithm is divided into six main optimization stages and the specific steps of obtaining the Pareto optimal solution using the EMOEAD algorithm are as follows: 1) Set the population size, neighborhood size, crossover / mutation probability, maximum iteration time, and frontier solution set. Initialize the weight vectors and group them into neighborhoods by calculating the Euclidean distance between each weight vector; 2) Perform cluster analysis on the dataset and select areas with large increases in roadbed compaction degree and short roadbed compaction time; 3) From the identified regions, focus on areas with large increases in roadbed compaction degree and short roadbed compaction time. Calculate the average values ​​of decision variables for these regions and generate initial solutions using chaotic mapping. Calculate the objective function values ​​for all individuals and generate initial ideal points. Select non-dominated individuals to update the frontier solution set. To replace the random initialization method, this paper uses an initialization sequence generated by chaotic mapping, following the settings of the following equations: In the formula, X represents the final initial population position generated after the mapping, lb represents the lower bound of the search space, ub represents the upper bound of the search space, chaos(N, dim) represents the chaotic sequence matrix generated by the chaotic mapping, N represents the initial population size, and dim represents the spatial dimension of the optimization problem. This represents the value of the chaotic variable at the nth iteration. Represents the result calculated iteratively. A chaotic sequence value, Represents the control parameters for chaotic mapping.

[0041] 4) Select two individuals from the current population and perform crossover and mutation. After crossover and mutation, new offspring individuals are generated, and the neighborhood is updated by calculating the aggregation function value between the new individuals and their neighbors. 5) Calculate the aggregation function value of the neighborhood of the individual grouped according to the current neighborhood. If the aggregation function value of the new descendant individual is better, replace the domain individual and update the frontier solution set and ideal point; 6) Repeat the above steps until the maximum number of iterations is reached, and output the front solution set as the Pareto front set.

[0042] Based on the above model predictions, to obtain the optimal roadbed compaction degree, two output parameters—the increase in roadbed compaction degree and the roadbed compaction time—are selected for multi-objective optimization. Due to the large number of parameters involved, this invention considers using an improved intelligent optimization algorithm to solve the multi-objective optimization problem. Among numerous algorithms, EMOEAD, which combines engineering data clustering and logic chaotic mapping enhancement, is one of the latest and best-performing multi-objective optimization methods. The EMOEAD algorithm exhibits good convergence and robustness, reduces computational complexity, and has good adaptability to multi-objective optimization problems. It can find more optimal Pareto solutions in multi-objective problems and better solve high-dimensional multi-objective problems. This algorithm has been widely applied to multi-objective optimization problems in various fields in recent years. Therefore, this invention selects the EMOEAD algorithm to optimize and control the two objective parameters—the increase in roadbed compaction degree and the roadbed compaction time—to achieve the optimization of roadbed compaction degree.

[0043] Example 2 In one embodiment of the present invention, such as Figure 1 As shown, the intelligent compaction optimization method and equipment system for highway subgrade mainly includes the following steps: (1) Construction of the influencing factor index system and collection of sample data Based on a combination of extensive engineering practice and literature research, the initial compaction degree of highway subgrade was selected. Intelligent compaction vibration force for highway subgrade Intelligent compaction vibration amplitude of highway subgrade Highway subgrade temperature Highway subgrade intelligent compaction pass count Final compaction degree of highway subgrade Intelligent compaction speed of highway subgrade A total of seven influencing factors were used as input variables. The increase in roadbed compaction degree and roadbed compaction time were used as output variables. 143 sets of field monitoring data were selected as the sample dataset, and the sample data scheme is shown in Table 1 below. 114 sets were randomly selected as the training set, and the remaining 29 sets were used as the test set.

[0044] Table 1 (2) Hyperparameter optimization of NGBoost model To improve the prediction performance of the NGBoost prediction model, this paper employs IBKA optimization to optimize four hyperparameters: learning rate, row sampling ratio, column sampling ratio, and number of iterations. Before optimizing the hyperparameters, the initial ranges for the learning rate, row sampling ratio, column sampling ratio, and number of iterations are given as (0, 0.1), (0.4, 0.8), (0.4, 0.8), and (200, 2000), respectively. The hyperparameter search results between the increase in roadbed compaction and the time of roadbed compaction are shown in Table 2 below.

[0045] Table 2 (3) Establishment and evaluation of NGBoost prediction model Based on the hyperparameter optimization results of NGBoost, training sets were used for learning simulations to establish prediction models for the increase in roadbed compaction degree and roadbed compaction time based on NGBoost. The prediction models on the training sets were then validated using a test set. The prediction results are as follows: Figure 4-7 As shown.

[0046] from Figure 4 and Figure 5 It can be seen that the NGBoost model can effectively predict the increase in roadbed compaction. Figure 4 The results show the prediction results of NGBoost for the increase in road subgrade compaction on the training set. The results indicate that the NGBoost model has fully learned the relationship between the input variables and output indices of the training sample set. The prediction results for the increase in road subgrade compaction obtained through the training set are basically consistent with the actual observed values, with an RMSE of 88.651, a MAE of 6.600, and a R² value of [missing value]. 2 It is 0.956. Figure 5Comparing the prediction results with the test set, it can be seen that the predicted values ​​of the increase in roadbed compaction in the test set are in good agreement with the actual observed values, with an RMSE of 30.232, a MAE of 4.168, and a R... 2 The value is 0.912. The results obtained from both the training and test sets indicate that the NGBoost prediction model has high accuracy and good generalization performance, and its fitted nonlinear prediction function has high accuracy.

[0047] Similarly, from Figure 6 and Figure 7 As can be seen, NGBoost also has good accuracy in predicting the time of roadbed compaction.

[0048] (4) Establish the objective function The nonlinear mapping function of IBKA-NGBoost is introduced as the objective function of EMOEAD.

[0049] Objective function for increasing the compaction degree of highway subgrade based on IBKA-NGBoost : Objective function for roadbed compaction time based on IBKA-NGBoost : in These represent the initial compaction degree of the highway subgrade. Intelligent compaction vibration force for highway subgrade Intelligent compaction vibration amplitude of highway subgrade Highway subgrade temperature Highway subgrade intelligent compaction pass count Final compaction degree of highway subgrade Intelligent compaction speed of highway subgrade .

[0050] (5) Determine the range of constraints The constraint ranges of each influencing factor parameter are set according to the actual engineering situation. When using the EMOEAD algorithm to find the optimal solution for intelligent compaction parameters of highway subgrade, the decision range of the initial population must first be set to ensure that the initial population has practical significance. To avoid the optimized intelligent compaction parameter results of highway subgrade differing too much from the actual engineering parameters, resulting in large parameter adjustment spans and potential safety hazards, this invention uses the fluctuation range of each input parameter in the selected sample as the main reference to set the range of the initial decision variables.

[0051] (6) Multi-objective optimization and decision-making based on EMOEAD Before performing EMOEAD multi-objective optimization, the values ​​for the number of objectives, population size, and stopping criteria of the optimization algorithm were first determined. Considering that an appropriate population size and number of iterations can promote the convergence of multi-objective optimization, this paper sets the crossover operator to 0.8, the mutation operator to 0.005, the population size to 100, the neighborhood size to 20, and the maximum and stopping generations to 100. After setting the parameters, the EMOEAD algorithm was run, and 100 pairs of Pareto front solutions were obtained. The corresponding parameters and optimization results are shown in [the table below]. Figure 8 and Figure 9 .

[0052] To select the optimal solution from multiple options, this paper employs the TOPSIS decision-making method. First, the membership degree of each solution in the Pareto set to each optimization objective is calculated. Then, the comprehensive dominance value of each solution group is calculated. The overall performance of the solutions is judged based on the magnitude of the dominance value. The calculated solution group 68 is significantly higher than the others and is therefore the optimal solution. The input parameters corresponding to this solution are taken as the optimal control scheme. After optimization, the increase in roadbed compaction degree at most sample points is higher than before optimization, with an average increase of approximately 0.5 percentage points, and the volatility is reduced, with the standard deviation decreasing from 0.42% to 0.28%. Simultaneously, the average construction time for intelligent compaction of the roadbed is reduced from 580 seconds before optimization to 540 seconds, a reduction of approximately 6.9%. The optimized scheme significantly shortens the construction time while ensuring improved final compaction quality, effectively achieving a synergistic improvement in construction quality and efficiency. This optimization result provides a practical reference for the intelligent setting and dynamic adjustment of vibration compaction parameters in similar roadbed engineering projects.

[0053] (7) Construction and functional implementation of the large language model platform like Figure 10 As shown, this invention develops an intelligent compaction cloud platform management system that integrates prediction models, optimization algorithms, and engineering specification knowledge bases. The platform, built on Python and PYQT5 libraries, features a visual interactive interface and aims to provide full-chain support for highway subgrade compaction construction and operation, from intelligent prediction and optimization decision-making to knowledge-based Q&A. Its core lies in integrating proactive quality control during construction with long-term performance management during operation, achieving integrated construction and operation.

[0054] 1) Platform Overview and Core Components Rapid prediction and recommendations for the increase in roadbed compaction degree and the relationship between roadbed compaction time: Based on the trained IBKA-NGBoost model, rapid inference is performed on the input real-time or preset parameters to predict compaction quality and efficiency, and a preliminary assessment is given.

[0055] Multi-objective optimization guidance for parameters: Integrate multi-objective optimization algorithms such as EMOEAD, take the increase in roadbed compaction degree and roadbed compaction time as objectives, solve for the Pareto optimal solution set, and provide a quantitative basis for dynamic adjustment of on-site process parameters.

[0056] Intelligent query and interactive Q&A for roadbed compaction specifications: Based on retrieval enhancement generation technology, a localized engineering knowledge base is built, supporting users to query professional specifications, diagnose quality problems, and consult on rectification solutions in natural language.

[0057] 2) Principles and Integration Design of Large Language Model Technology Traditional intelligent compaction methods have limitations in handling complex working conditions, multi-source information fusion, and interpretable decision-making. This platform introduces a large language model as the "intelligent hub," aiming to achieve the following technological breakthroughs: Contextual learning and few-shot adaptation enable the system to quickly adjust prediction and decision-making models based on a small amount of new working condition data, enhancing its engineering applicability.

[0058] Knowledge reasoning and decision interpretability: Large language models can not only output decision suggestions, but also generate reasoning chains, explain the basis for decisions, and improve decision transparency.

[0059] Multi-objective collaborative optimization under complex constraints: Within a unified cognitive framework, predictive models and optimization algorithms are integrated to simulate the long-term effects of different parameter combinations, generating and providing various trade-off schemes and their impact assessments.

[0060] Human-computer natural interaction and dynamic knowledge base: Through a natural language interface, construction and supervision personnel can query real-time quality, request fault diagnosis, and retrieve historical cases through dialogue, and can continuously learn from the interaction to update the internal knowledge base.

[0061] 3) Core Function Implementation Process Compaction quality and efficiency prediction: Users import datasets through the interface, select target variables and input features, and the platform calls the optimized IBKA-NGBoost model to make predictions and visualizes the results. A large language model can further interpret the results, pointing out key influencing parameters and engineering considerations.

[0062] Multi-objective optimization and parameter recommendation: Users select an optimization algorithm and surrogate model, and set the optimization objective. The platform executes the optimization algorithm and outputs a Pareto optimal solution set and corresponding recommended parameter values. A large language model can analyze the optimization results, clarify the contradictory effects of different parameters on the objective, and generate directly applicable operational suggestions.

[0063] Intelligent Interaction and Diagnosis of Procedures: The platform has a built-in local knowledge base management system based on the RAGFlow framework. Users can create and upload relevant engineering specifications, construction manuals, and other documents to build the knowledge base. In the Q&A interface, users ask questions in natural language, and the system retrieves relevant information from the knowledge base, organizes it through a large language model to generate a clearly structured and standard-compliant answer, and can provide rectification steps and long-term quality management suggestions.

[0064] Example 3 like Figure 2 As shown, according to one aspect of the present invention, a smart compaction optimization system for highway subgrade is also provided, which implements the above-mentioned method, comprising: a first main control module, which takes the increase in highway subgrade compaction degree and highway subgrade compaction time as optimization objectives, uses the Internet of Things to collect multi-source datasets in real time, extracts frequency domain features, identifies key parameters, and constructs a sample dataset based on the key parameters and the corresponding increase in highway subgrade compaction degree and highway subgrade compaction time; a second main control module, which, based on the sample dataset, establishes a prediction model for the increase in highway subgrade compaction degree and highway subgrade compaction time based on IBKA-NGBoost to obtain a nonlinear mapping relationship between the optimization objectives and key parameters; and a third main control module, which uses the nonlinear mapping relationship as the objective optimization fitness function and the constraint range of the key parameters as the constraint function, constructs a multi-objective optimization model based on IBKA-NGBoost-EMOEAD to obtain the optimal solution of the objective function and proposes a suggested range of values ​​for setting smart compaction parameters for highway subgrade.

[0065] In a preferred embodiment, the third main module utilizes EMOEAD and TOPSIS, which combine engineering data clustering and Logic chaotic mapping enhancements, to perform intelligent compaction optimization control of highway subgrade.

[0066] Preferably, in this embodiment, the system further includes a fourth main module: a large language model platform for developing and integrating prediction, optimization and procedure knowledge bases, to achieve natural language interaction support for compaction cloud map visualization, dynamic parameter adjustment suggestions and intelligent compaction quality problem diagnosis and rectification.

[0067] like Figure 3 As shown, according to another aspect of the present invention, an electronic device is also provided, characterized in that it includes: at least one processor, at least one memory, and a communication interface; wherein the processor, memory, and communication interface communicate with each other; the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the above-described method.

[0068] In a preferred embodiment, according to another aspect of the present invention, a non-transitory computer-readable storage medium is also provided, the non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute a method for intelligent compaction optimization of highway subgrade according to any of the above embodiments or a combination of embodiments.

[0069] In this invention, machine learning (ML) algorithms possess powerful data analysis capabilities, enabling rapid search for optimal solutions. They are suitable for solving complex nonlinear problems involving multiple factors and are widely used in engineering research related to prediction, analysis, and optimization. Commonly used predictive machine learning algorithms in engineering include NGBoost, RF, BNPP, and GBDT. NGBoost can solve predictive analysis problems involving nonlinearity, data noise, and complex dependencies, exhibiting strong generalization and high prediction accuracy. It also performs well on small datasets and is widely used in engineering prediction. In this invention, the two output targets—the increase in roadbed compaction degree and the roadbed compaction time—have a complex nonlinear relationship with their parameters, and their objective function cannot be directly expressed using a specific mathematical function. Therefore, the NGBoost algorithm can be chosen to predict the increase in roadbed compaction degree and the time between roadbed compaction and compaction.

[0070] The method in this embodiment of the invention is implemented using an electronic device; therefore, it is necessary to introduce the relevant electronic device. For this purpose, this embodiment of the invention provides an electronic device, such as... Figure 3 As shown, the electronic device includes at least one processor, a communications interface, at least one memory, and a communications bus, wherein the at least one processor, the communications interface, and the at least one memory communicate with each other via the communications bus. The at least one processor can invoke logical instructions stored in the at least one memory to execute all or part of the steps of the methods provided in the foregoing method embodiments.

[0071] Furthermore, when the logical instructions in at least one of the aforementioned memories can be implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part 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 method embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0072] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0073] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0074] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Based on this understanding, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, or sometimes in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0075] In this invention, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0076] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for optimizing intelligent compaction of highway subgrade, characterized in that, Includes the following steps: Step 1: Using the increase in roadbed compaction degree and roadbed compaction time as optimization targets, the Internet of Things is used to collect multi-source datasets in real time, extract frequency domain features, identify key parameters, and construct a sample dataset based on the key parameters and the corresponding increase in roadbed compaction degree and roadbed compaction time. Step 2: Based on the sample dataset, establish a prediction model for the increase in roadbed compaction degree and roadbed compaction time based on IBKA-NGBoost, in order to obtain the nonlinear mapping relationship between the optimization objective and key parameters; Step 3: Using the nonlinear mapping relationship as the objective optimization fitness function and the constraint range of the key parameters as the constraint function, construct a multi-objective optimization model based on IBKA-NGBoost-EMOEAD to obtain the optimal solution of the objective function, and propose a suggested range of values ​​for intelligent compaction parameters of highway subgrade.

2. The method for intelligent compaction optimization of highway subgrade according to claim 1, characterized in that, In step one, the key parameters include: initial compaction degree of the highway subgrade. Intelligent compaction vibration force for highway subgrade Intelligent compaction vibration amplitude of highway subgrade Highway subgrade temperature Highway subgrade intelligent compaction pass count Final compaction degree of highway subgrade Intelligent compaction speed of highway subgrade .

3. The method for intelligent compaction optimization of highway subgrade according to claim 1, characterized in that, Step two specifically includes the following steps: (21) The sample dataset is preprocessed and then divided into training dataset and test dataset; (22) Based on the training dataset, IBKA is used to optimize the key hyperparameters of NGBoost, the optimal hyperparameter combination is selected, an IBKA-NGBoost prediction model is built based on the optimal hyperparameter combination, and the IBKA-NGBoost prediction model is tested using the test dataset. (23) Based on the IBKA-NGBoost prediction model, the predicted values ​​and the true values ​​are fitted to obtain the nonlinear mapping relationship between the optimization target and the key parameters.

4. The method for intelligent compaction optimization of highway subgrade according to claim 2, characterized in that, The optimization of key hyperparameters of NGBoost using IBKA in step (22) includes the following steps: (221) Population initialization based on the optimal point set strategy: The optimal point set strategy is introduced to solve the problems of uneven initial population distribution and insufficient diversity caused by random initialization. The population initialization can be expressed as: ， in, This represents the position of the i-th individual in the j-th dimension. Indicates the upper bound. Indicates the lower bound. This represents the optimal point set value in the j-th dimension, and k represents the index number, i.e., the current position when the point set was generated; (222) Based on the improved predation stage of the golden sine algorithm, a golden sine optimization strategy is introduced to overcome the defect of the original black kite algorithm being prone to getting trapped in local optima. Its predation stage can be expressed as: ， in, and These represent the positions of the i-th individual in the j-th dimension at time t+1 and time t, respectively. This represents the perturbation factor that decreases over time. This represents a random number that is uniformly distributed within the interval [0,1]. and This represents the random sinusoidal perturbation factor. Indicates the current optimal global position. This indicates the adjustment factor related to the golden ratio. Represents a fixed value; (223) When population evolution stagnates, a differential evolution mechanism is introduced to perform guided population updates, including mutation, crossover and selection operations.

5. The intelligent compaction optimization method for highway subgrade according to claim 4, characterized in that, The mutation operation includes: Three distinct individuals are randomly selected from the current population, and a mutant individual is generated through mutation, represented as: ， in, Indicates a variant individual. This represents three distinct individuals randomly selected from the current population. Indicates the scaling factor; Preferably, the crossover operation includes: The mutant individuals are cross-mixed with the original individuals to generate experimental individuals, represented as follows: ， in, This represents the new experimental individual after the crossover operation. Indicates the crossover probability. This represents a randomly generated integer index within the range [1, pop], where pop is the population size; Preferably, the selection operation includes: A greedy strategy is adopted, comparing the fitness of experimental individuals with that of the original individuals, and selecting individuals with better fitness for the next generation of the population, as follows: ， in, This refers to the next generation of individuals selected from among the previous generation. This represents the fitness value.

6. The method for intelligent compaction optimization of highway subgrade according to claim 1, characterized in that, In step three, the objective function is determined by predicting the nonlinear mapping relationship function using IBKA-NGBoost. , They are respectively: ， in, These are key parameters, i = 1 to 7, specifically... This indicates the initial compaction degree of the highway subgrade. This indicates the intelligent compaction excitation force of the highway subgrade. Indicates the amplitude of intelligent compaction vibration of highway subgrade, Indicates the temperature of the roadbed, Indicates the number of intelligent compaction passes for highway subgrade. Indicates the final compaction degree of the highway subgrade, Indicates the intelligent compaction speed of highway subgrade; Preferably, the constraint function includes: ， in, The objective function for the increase in roadbed compaction degree based on IBKA-NGBoost is: The objective function for highway subgrade compaction time based on IBKA-NGBoost.

7. The method for intelligent compaction optimization of highway subgrade according to claim 1, characterized in that, In step three, the EMOEAD algorithm is used to obtain the Pareto optimal solution, as detailed below: (31) Set the population size, neighborhood size, crossover mutation probability, maximum iteration time and frontier solution set, initialize the weight vector, and group them into neighborhoods by calculating the Euclidean distance between each weight vector; (32) Perform cluster analysis on the sample dataset and select areas with large increases in roadbed compaction degree and short roadbed compaction time; (33) Calculate the average value of the decision variables in the selection area, generate the initial solution using the chaotic mapping, calculate the objective function value of all individuals, generate the initial ideal point, and select non-dominated individuals to update the frontier solution set; (34) Select two individuals from the current population, perform crossover and mutation strategies, generate new offspring individuals after crossover and mutation, and update the neighborhood by calculating the aggregation function value between the new individuals and their neighbors; (35) Calculate the aggregation function value of the neighborhood of the individual grouped according to the current neighborhood. If the aggregation function value of the new descendant individual is better, replace the domain individual and update the frontier solution set and ideal point. (36) Repeat steps (31)-(35) above until the maximum number of iterations is reached, and output the front solution set as the Pareto front set; Preferably, in step (33), in order to replace the random initialization method, an initialization sequence generated by chaotic mapping is used, and the following equations are followed: ， ， In the formula, X represents the final initial population position generated after the mapping, lb represents the lower bound of the search space, ub represents the upper bound of the search space, chaos(N, dim) represents the chaotic sequence matrix generated by the chaotic mapping, N represents the initial population size, and dim represents the spatial dimension of the optimization problem. This represents the value of the chaotic variable at the nth iteration. Represents the result calculated iteratively. A chaotic sequence value, Represents the control parameters for chaotic mapping.

8. A smart compaction optimization system for highway subgrade, characterized in that, include: The first main control module takes the increase in roadbed compaction degree and roadbed compaction time as optimization targets, uses the Internet of Things to collect multi-source datasets in real time, extracts frequency domain features, identifies key parameters, and constructs a sample dataset based on the key parameters and the corresponding increase in roadbed compaction degree and roadbed compaction time. The second main control module, based on the sample dataset, establishes a prediction model for the increase in roadbed compaction degree and roadbed compaction time based on IBKA-NGBoost, in order to obtain the nonlinear mapping relationship between the optimization objective and key parameters; The third main control module uses the nonlinear mapping relationship as the objective optimization fitness function and the constraint range of key parameters as the constraint function to construct a multi-objective optimization model based on IBKA-NGBoost-EMOEAD to obtain the optimal solution of the objective function, and proposes a suggested range of values ​​for intelligent compaction parameters of highway subgrade.

9. An electronic device, characterized in that, include: At least one processor, at least one memory, and a communication interface; wherein, The processor, memory, and communication interface communicate with each other; The memory stores program instructions that can be executed by the processor, which calls the program instructions to execute the intelligent compaction optimization method for highway subgrade as described in any one of claims 1 to 8.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to execute the intelligent compaction optimization method for highway subgrade as described in any one of claims 1 to 4.

Images