Parameter optimization method and system for vibration compaction of high-speed rail filler

Abstract

The present invention relates to a parameter optimization method for vibration compaction of a high-speed rail filler, comprising the following steps: (1) fitting dry density in a compaction process b

Description

This invention relates to the field of civil engineering technology, and specifically to a parameter optimization method and system for high-speed railway filling vibration compaction. Background Art Subgrade performance is closely related to the composition of the filling material and the control of the vibration compaction process. To ensure the safety of high-speed railway subgrades and improve the comfort of train operation, there are strict standards for the gradation, maximum particle size, and shape of the subgrade filling material. Traditional high-speed railway subgrade vibration compaction quality inspection generally uses the test pit sampling method, which ignores the problem of filling material discreteness, resulting in low compaction efficiency and uneven compaction. The compaction state of the filling material changes continuously during the compaction process. Adjusting the vibration parameters continuously based on the compaction state of the filling materiaI, rather than using fixed compaction parameters, can improve compaction efficiency. To achieve this goal, Intelligent Compaction (IC), consisting of Continuous Compaction Control (CCC) and Intelligent Compaction, has gradually become a research hotspot in recent years. Current research results on vibration parameters (amplitude, frequency, time) still focus on the dynamic response under different vibration parameters, mainly revealing the influence of single vibration parameters on the compaction effect. There is a lack of research results on the improvement of vibration parameters during the compaction process, which cannot meet the requirements of real-time adjustment of vibration parameters for intelligent compaction. With the emergence of continuously variable amplitude and frequency rollers, the frequency and amplitude of the roller can be continuously adjusted during the compaction process. However, adjusting the vibration parameters to maximize the compaction degree of the next compaction is a local 2 improvement. During the compaction process, the vibration parameters need to be improved in real-time so that the filling material can reach the optimal compaction state better and faster. At the same time, excessive excitation energy can easily lead to the "jumping vibration" phenomenon of the vibration compaction equipment, reducing the service life of the equipment. It can also cause increased particle breakage, resulting in uneven settlement diseases of the subgrade. Summary of the Invention The main objective of this invention is to provide a parameter optimization method and system for high-speed railway subgrade filling vibration compaction, in order to address the problems in the prior art where the compaction vibration parameters cannot be adjusted in real-time during traditional vibration compaction, the excitation energy cannot be well controlled, and the vibration compaction equipment is prone to "jumping vibration" phenomena, resulting in increased particle breakage and uneven settlement diseases of the subgrade. The present invention provides a parameter optimization method for high-speed railway filling vibration compaction, comprising the following steps: (1) Using a hyperbolic model to fit the dry density during the compaction process, calculating and constructing the original data; (2) Based on the acquired training data, using a BP neural network to establish a dry density increment prediction model; (3) Adopting the compaction degree index to evaluate the compaction quality and establishing compaction degree constraints; (4) Solving the vibration parameter optimization process based on the GA algorithm and establishing a GA-based dynamic optimization model; (5) Determining the dynamic optimization result after parameter optimization processing as the optimal scheme for vibration compaction. Further, the formula of the hyperbolic model in step (1) is: a(pmax - Po)n p“ =——b—+p° Where pd is the current state density; pmax is the dry density value at the stable compaction state under the current working condition; p0 is the initial dry density value; n is the number of vibrations; and a and b are model parameters. Further, the establishment steps of the dry density increment prediction model in step (2) include: a. Constructing a high-speed railway filling vibration compaction parameter dataset, including the steps: original data input, data preprocessing, and dividing the dataset into training, validation, and test sets; b. Building a BP neural network model architecture, inputting the training and validation sets to perform model training, and determining whether the validation set error is less than the set error; If it is greater, modify the model architecture and retrain; If it is less, proceed to the next step; c. Using the test set to verify the prediction ability of the BP neural network model, and determining whether the test set error is less than the set error; If it is greater, return to step b to retrain the model; If it is less, save the final dry density increment prediction model. Further, the dry density increment prediction model establishes a nonlinear functional relationship between the model input and output. The function is expressed as follows: Aft, = / (4, ,f„p^.,) Where f represents the mapping relationship, i40Z and ft represent the amplitude and frequency at the current time, respectively, pdi represents the dry density at the previous time, and &Pd. represents the increment of dry density at the current time. Further, the data preprocessing adopts the data normalization method, and the formula is as follows: _ % ^■mm(axis=0) PC cl-fl ~~ ' •’LU -y. __ y Am.in(axis=0) JimaxCaxis=O) X scaled = xstd * OflX “ + mln Where x is the data to be normalized, xmin(axis=o) is the row vector composed of the minimum values in each column, %max(aXiS=o) is the row vector composed of the maximum values in each column, max is the maximum value of the interval to be mapped to, the default is 1, min is the minimum value of the interval to be mapped to, the default is 0, xstd is the standardization result, and xscaied is the normalization result. Further, the BP neural network is an optimized BP neural network, and the optimized BP neural network is obtained by optimizing the traditional BP neural network by introducing the AdamOptimizer learning rate optimizer; The AdamOptimizer improvement algorithm is a conventional optimization algorithm that dynamically adjusts the learning rate of each parameter using the first-order moment estimation and second-order moment estimation of the gradient; The improvement formula of the AdamOptimizer improvement algorithm is: lrt - _ pt * U Where lr0 is the initial learning rate, lrt is the learning rate at time t, is the exponential decay rate of the first-order moment estimation, and is the exponential decay rate of the second-order moment estimation. Further, In step (3), the formula for the compaction degree constraint is as follows: ?dn >g Pd r v-max Where pdn is the dry density value after n vibrations, and Pdmax is the maximum dry density. Further, the GA-based dynamic optimization model in step (4) is obtained by the following steps: Set the initial parameters of the Genetic Algorithm (GA) and randomly initialize the population, completing the initialization configuration; Iteratively calculate the dry density increment, and check if the individual compaction degree and the number of iterations meet the requirements. If not, continue calculating the dry density increment. If the conditions are met, save the compaction degree sequence, then calculate the fitness values of the population individuals and sort them; Perform iterative optimization processes such as GA selection, crossover, and mutation, and output the final result. Further, the fitness value of the population individuals is calculated using the total compaction energy function, and the formula is as follows: n n fitness = Et = 2Aoi(W + MpAoi-n:3ff) i=l i=l Where fitness is the fitness function, n is the number of vibrations required for the entire compaction process, f is the dry density value of the (i-1 )th state, Aoi is the vibration parameter selected for the i-th state, Et is the vibration compaction energy of the i-th state, W is the static load, and Mp is the eccentricity. A parameter optimization system for high-speed railway filling vibration compaction is also provided, including a data input terminal, a background computing and processing terminal, and a data display terminal. The background computing and processing terminal is a computing and processing terminal that performs computing and processing using the parameter optimization method for high-speed railway filling vibration compaction according to any one of claims 1 -9. This invention discloses a parameter optimization method for high-speed railway filling vibration compaction based on the principle of minimum energy. By comprehensively using deep learning algorithms and target optimization algorithms to improve vibration compaction parameters in real-time during the compaction process, a hyperbolic model is first constructed through experiments and mathematical theory to calculate and construct the original data. A dry density increment prediction model is established using a BP neural network. Then, based on the principle of minimum energy, a genetic algorithm is used to construct a vibration parameter improvement method for the compaction process. The compaction parameters are adjusted in real-time based on the compaction state of the filling material, effectively improving the compaction quality and efficiency, enabling the filling material to reach the optimal compaction state better and faster. Finally, a parameter optimization system for high-speed railway filling vibration compaction based on the principle of minimum energy is developed. This invention solves the problems of traditional vibration compaction, such as the inability to adjust compaction vibration parameters in real-time, poor control of excitation energy, the tendency of vibration compaction equipment to experience "jumping vibration" phenomena, increased particle breakage, and uneven settlement diseases of the subgrade. The invention will be further described below in conjunction with the accompanying drawings and specific embodiments. Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention. Brief Description of the Drawings The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention and are not intended to be limiting. In the drawings: Fig. 1 is a flow chart of a parameter optimization method for high-speed railway filling vibration compaction. Fig. 2 is a flow chart of the construction of a dry density increment prediction model based on an improved BP neural network. Fig. 3 is the architecture of a BP neural network model based on an improved BP algorithm. Fig. 4 is a flow chart of the GA dynamic optimization model. Fig. 5 is a diagram of a 6-bit binary encoding of variables. Fig. 6 is the system interface of a parameter optimization system for high-speed railway filling vibration compaction based on the principle of minimum energy. Fig. 7 shows the loss function curves of the training set and validation set for two types of dry density increment models. Fig. 8 is a comparison diagram of the test results of the dry density increment prediction model before and after improvement. Fig. 9 is a diagram of the final vibration parameter optimization results. Fig. 10 is a diagram of the vibration optimization parameters and compaction curves for each time period. Detailed Implementation Method The present invention is described clearly and completely below with reference to the accompanying drawings. A person skilled in the art will be able to implement the invention based on this description. Before describing the invention in conjunction with the figures, it should be noted that: The technical solutions and features provided in various sections of this document, including the following description, may be combined without conflict, unless otherwise stated. Furthermore, the embodiments of the invention described below are generally only some examples, not all possible embodiments. Therefore, based on these embodiments, all other implementations obtained by a person skilled in the art without inventive effort fall within the scope of the present invention. Regarding terminology and units in this invention: The terms "comprising," "having," and their variations in the specification, claims, and related sections are intended to be inclusive and non-exclusive. The steps of a parameter optimization method for high-speed railway fill vibration compaction are shown in Figure 1: Step 1: Collect and construct raw data. Specifically, a hyperbolic model is used to fit the dry density during the compaction process, calculating and constructing the raw data; Step 2: Establish a dry density increment prediction model. Specifically, based on the obtained training data, a backpropagation (BP) neural network is used to establish a dry density increment prediction model; Step 3: Establish constraint conditions. Specifically, the degree of compaction is used to evaluate the compaction quality, establishing a compaction degree constraint; Step 4: Establish a genetic algorithm (GA)-based dynamic optimization model. Specifically, a GA-based algorithm solves the vibration parameter optimization process, establishing a dynamic optimization model; Step 5: Output the optimal solution. Specifically, the dynamic optimization results after parameter optimization are determined as the optimal vibration compaction solution. As shown in Figure 2, building the improved BP neural network-based dry density increment prediction model involves dataset construction, model training, and model testing. First, dataset construction includes these steps: raw data input, data preprocessing, and dataset partitioning into training, validation, and test sets. Next, model training involves: optimized BP neural network architecture construction and training resumption from breakpoints. If the validation set prediction error exceeds a predefined threshold, the process returns to architecture optimization. Otherwise, it proceeds to model testing. Finally, model testing involves: saving the model, calling the model, and testing the model. If the test error is within the predefined threshold, the final model is saved. Otherwise, the process returns to model training. Based on the training data, an improved BP neural network creates the dry density increment prediction model. The degree of compaction evaluates the compaction quality, and the compaction degree constraint ensures adequate fill compaction. A GA-based algorithm solves the vibration parameter optimization, and a GA-based dynamic optimization model, aiming to minimize vibration compaction energy, enables real-time parameter improvement and optimization. The dynamically optimized results after parameter optimization determine the optimal vibration compaction solution. In Step 1: Calculating and constructing the raw data involves three steps: 1) Determine the hyperbolic function relating dry density and vibration parameters during high-speed railway fill vibration compaction. 2) Experimentally determine the model parameters a and b. 3) Calculate the dry density dataset for different vibration frequencies (f) and amplitudes (A) combined with corresponding dry densities (p), using this as the raw data. In Step 1), determining the hyperbolic functional relationship between dry density and vibration parameters during high-speed railway fill vibration compaction involves three sub-steps: (D Establishing the basic hyperbolic function equation, as shown below: ax y* =---- 7 b + x where a and b are variable parameters. @ To eliminate the dimensional relationships between different variables, the dry density is normalized as follows: * y — ymin y =---- Vmax ymin where y* is the normalized value; y> ymin^ ymaxare the actual value, minimum value, and maximum value of the variable within the normalization range, respectively. ® Substituting the above equation into the equation in ® yields the fitting model for dry density and the number of vibrations: a(pmax - Po)n , ——b—+p« where pd is the current dry density; pmax is the dry density value at a stable compaction state under the current working condition; p0 is the initial dry density value; n is the number of vibrations; and a and b are the model parameters. In Step 2), the model parameters a and b are determined experimentally. The model parameters and fitting correlation coefficients under various working conditions are shown in Table 1 below: Table 1 Model Parameters and Correlation Coefficients Amplitude 0.15mm 0.3mm 0.6mm Vibration Frequency Fitting parameters R2 Fitting parameters R2 Fitting parameters R2 a b a b a b 10Hz 1.070 255.989 0.994 1.025 118.380 0.991 1.080 115.140 0.996 15HZ 0.978 99.018 0.991 1.011 87.520 0.999 1.146 290.864 0.990 20HZ 1.0052 15.912 0.999 0.956 46.721 0.990 1.062 95.367 0.992 25HZ 0.984 66.632 0.991 0.991 79.329 0.996 0.974 82.473 0.988 30HZ 0.996 69.242 0.991 0.987 80.219 0.997 0.969 126.715 0.985 As shown in Figure 2, Step (2), constructing the dry density increment prediction model, is mainly divided into three sub-steps: Constructing the high-speed railway fill vibration compaction parameter dataset. First, preprocess the data by normalizing it to the range [0, 1] using the following formula. Then, randomly divide the dataset into training, validation, and test sets with a data volume ratio of 70%, 15%, and 15%, respectively. X -^771177(0X15 = 0) xstd =-----------2----------- xmin(axis=<S) xmax(axis=0) X scaled = *std * + Where x is the data to be normalized, xmin^axis=0^ is the row vector composed of the minimum values in each column, xmax(axis=Q) is the row vector composed of the maximum values in each column, max is the maximum value of the interval to be mapped to, the default is 1, min is the minimum value of the interval to be mapped to, the default is 0, xstd is the standardization result, and xscaled is the normalization result. (D’ Building and optimizing the BP neural network model architecture. Input the training and validation sets to train the model. Determine whether the validation set error is less than the set error. If it is greater, modify the model architecture and retrain. If it is less, proceed to the next step. As shown in Figure 3, Step involves optimizing the BP neural network model architecture, which can be divided into input, hidden, and output layers. The input layer has 3 neurons, simultaneously inputting the current amplitude A_Oi, frequency fj, and the previous dry density p_di. The output layer has 1 neuron, outputting the current dry density increment Ap_di. The dry density increment prediction model establishes a nonlinear functional relationship between the model input and output, expressed as: Where f represents the mapping relationship, t40Z and ft represent the amplitude and frequency at the current time, respectively, pdi represents the dry density at the previous time, and APd. represents the increment of dry density at the current time. In Step ®’ , the improved BP neural network algorithm converges faster and has higher prediction accuracy than the traditional BP algorithm. By introducing the AdamOptimizer learning rate optimizer based on the traditional BP neural network, the problems of the traditional BP algorithm, such as easily falling into local optima, sample dependence, and non-adjustable learning rate, can be improved. The learning rate is adjusted reasonably according to the change in the loss function, accelerating model convergence and improving prediction accuracy. In Step , the AdamOptimizer is an optimized version of conventional optimization algorithms (e.g., SGD, AdaGrad). It adds first-order and second-order momentum based on the SGD algorithm and corrects bias based on the AdaGrad algorithm, featuring fast convergence and high stability. It dynamically adjusts the learning rate of each parameter using the first and second moment estimations of the gradient. The main improvement formula of the AdamOptimizer is as follows: Where lr0 is the initial learning rate, lrt is the learning rate at time t, is the exponential decay rate of the first-order moment estimation, and is the exponential decay rate of the second-order moment estimation. Use the divided test set to verify the prediction ability of the trained optimized BP neural network model. Determine whether the test set error is less than the set error using MSE and MAE (as shown below). If it is greater, return to Step (D’ and retrain the model. If it is less, save the final dry density increment prediction model. N 2 1 mse = n t=l MAE N where N represents the number of predicted samples, and yt and yt represent the actual and predicted values, respectively. In Step (3), the compaction quality is evaluated using the degree of compaction. According to the "High-Speed Railway Subgrade Specification," the degree of compaction of the bottom layer of the subgrade must reach at least 95%. Therefore, the individual degree of compaction should satisfy the constraint condition of being greater than 0.95 to ensure the compaction quality of the fill. The formula is as follows: P^-n >g pd — Where pdn is the dry density value after n vibrations, and pdmax is the maximum dry density. As shown in Figure 4, Step (4), obtaining the GA-based dynamic optimization model, involves the following sub-steps: Set the initial parameters of the GA, such as population size (PS), maximum number of generations (MAXGEN), selection probability (SP), crossover probability (CP), and mutation probability (Mu). Randomly initialize the population using binary encoding. Each chromosome in the population includes two genes (vibration frequency and amplitude). As shown in Figure 5, each variable occupies 6 bits of binary encoding; Iteratively calculate the dry density increment and determine whether the individual degree of compaction is greater than 0.95 or whether the number of iterations has reached the number of chromosomes. If not, continue calculating the dry density increment. If the condition is met, save the compaction degree sequence and calculate the fitness value of each individual in the population based on the total compaction energy function (as shown below), and sort them according to the principle of minimum energy; n n fitness = £ Ei = £ 2X^0 i=l i=l + MpAoin3fi2) Where fitness is the fitness function, n is the number of vibrations required for the entire compaction process, f is the dry density value of the (i-1)th state, Aoi is the vibration parameter selected for the i-th state, Et is the vibration compaction energy of the i-th state, W is the static load, and Mp is the eccentricity. Perform iterative optimization processes such as selection, crossover, and mutation in the genetic algorithm, and output the final optimized vibration parameters. The implementation example and process of the complete method according to the present invention are as follows: Calculate and construct the original data: Use the hyperbolic model to fit the dry density during the compaction process to obtain the original data, which consists of 7500 sets. Part of the data is shown in Table 2. Table 2 Frequency Amplitude Current Density, 10 0.15 1.8 10 0.15 1.818348498 10 0.15 1.828649231 10 0.15 1.835244597 10 0.15 1.83982915 10 0.15 1.843200842 10 0.15 1.845784746 10 0.15 1.847828079 10 0.15 1.849484413 10 0.15 1.850854181 10 0.15 1.852005832 10 0.15 1.852987622 Establish a dry density increment prediction model: First, calculate the dry density increment based on the current dry density, and normalize the entire data. Part of the processed data is shown in Table 3. Table 3 Frequency Amplitude Current Dry Density Dry Density Increment 0.100000 0.100000 0.810811 0.008265 0.100000 0.100000 0.819076 0.004640 0.100000 0.100000 0.823716 0.002971 0.100000 0.100000 0.826687 0.002065 0.100000 0.100000 0.828752 0.001519 0.100000 0.100000 0.830271 0.001164 0.100000 0.100000 0.831435 0.000920 0.100000 0.100000 0.832355 0.000746 0.100000 0.100000 0.833101 0.000617 0.100000 0.100000 0.833718 0.000519 0.100000 0.100000 0.834237 0.000442 0.100000 0.100000 0.834679 0.000381 0.100000 0.100000 0.835061 0.000332 0.100000 0.100000 0.835393 0.000292 0.100000 0.100000 0.835685 0.000259 0.100000 0.100000 0.835944 0.000231 0.100000 0.100000 0.836176 0.000207 0.100000 0.100000 0.836383 0.000187 The dataset was randomly divided into training, validation, and test sets with a 70%, 15%, and 15% split, respectively. Using Python 3.8.1, the Keras deep learning library and other third-party libraries were imported. An improved BP neural network model for dry density increment prediction was constructed with 3 input neurons, 100 hidden neurons, 1 output neuron, an initial learning rate of 0.001, and 100 training epochs. The model was trained using the training and validation sets and compared with a traditional BP neural network. Figure 7 shows the training loss (loss) and validation loss (valjoss) curves during training. The traditional BP network exhibited slow and oscillating loss reduction, with the loss curve not converging even after 100 epochs. The improved BP network, however, showed rapid loss reduction, with both loss and valjoss converging after only 5 epochs without oscillations. This improvement is attributed to the introduction of momentum, which accelerates convergence. Throughout the training process, valjoss remained lower than loss, and both converged to near zero, indicating good training performance and generalization ability on the validation set. Therefore, the improved BP network offers adjustable parameters and faster convergence, addressing the slow convergence and susceptibility to local optima of traditional BP networks. The trained improved BP model was then saved and tested using 1125 unseen test set samples. MSE and MAE were used to evaluate the prediction results and compare them with the traditional BP model. Figure 8 shows the predicted versus actual values. The traditional BP model predicted most data points reasonably well but struggled with some prominent dry density increment peaks. The improved BP model, on the other hand, accurately predicted the fluctuating trends of dry density increment. The improved model achieved an MSE of 4.5*10 6 and MAE of 1.4*10 3, while the traditional model had an MSE of 6.35*10 4 and MAE of 5.1*10 3, demonstrating the superior prediction accuracy of the improved BP network. Thus, the improved BP model not only performs well on the training and validation sets but also exhibits high prediction accuracy and strong generalization ability on the test set, making it well-suited for dry density increment prediction. Finally, the improved BP-based dry density increment prediction model was saved for later use in the dynamic optimization process. A compaction degree constraint of greater than 0.95 was established. A GA-based dynamic optimization model was then developed. The GA algorithm was implemented in Python 3.8.1. Experimentation and analysis of GA parameters led to the selection of the following optimal parameters: population size (PS) = 150, maximum generations (MAXGEN) = 200, selection probability (SP) = 0.9, crossover probability (CP) = 0.6, and mutation probability (Mu) = 0.05. Binary encoding was used to create 500 chromosomes, each containing two genes (vibration frequency and nominal amplitude), with each variable represented by a 6-bit binary string. The dry density increment prediction model was called to iteratively calculate the dry density increment. The process continued until the compaction degree exceeded 0.95 or the number of iterations reached the number of chromosomes. Upon meeting either condition, the compaction degree sequence was saved, and the fitness value of each individual in the population was calculated based on the total compaction energy function, ranked according to the principle of minimum energy. Selection, crossover, and mutation operations were then iteratively performed until the maximum number of generations was reached. The final optimized vibration parameters were outputted, as shown in Figure 9. Achieving the target compaction degree required 9 complete vibrations, equivalent to 180 cumulative vibration compactions. Figure 9(a) shows that the optimized vibration frequency gradually increases from an initial 10.15 Hz to 27.58 Hz during the vibration compaction process, which aligns with the existing two-degree-of-freedom vibration compaction model's relationship between optimal vibration frequency and soil-compactor interaction. Figure 9(b) illustrates the adjusted nominal amplitude scheme, showing that the optimized amplitudes are lower than the pre-improvement scheme and initially increase from 0.15 mm before stabilizing at 0.34 mm after 120 continuous vibrations. Figure 9(c) compares the energy output before and after improvement, revealing a reduction in energy output throughout the compaction process, with a total reduction of 127.58 J, representing 25.61% of the pre-improvement energy output. Therefore, the proposed GA-based dynamic optimization method selects appropriate vibration parameters based on the current soil compaction state and effectively reduces energy consumption during vibration while ensuring a 95% compaction degree, improving compaction efficiency and reducing wear on the equipment. The dynamically optimized parameters after optimization, as shown in Table 4, were determined as the optimal vibration compaction scheme. Table 4 Optimized Vibration Parameters Number of Vibrations, n 1 2 3 4 5 6 7 8 9 fopi / Hz 10.1 5 14.8 8 25.0 2 25.2 6 25.3 5 26.2 9 26.7 0 27.5 4 27.5 8 AO / mm 0.15 0.15 0.15 0.24 0.15 0.29 0.34 0.34 0.34 Correspondingly, after inputting parameters into the vibration compaction parameter optimization system and running the system, the optimized vibration parameters for each time period and the compaction curve can be outputted on the interface, as shown in Figure 10. Experimental results of high-speed railway (HSR) fill vibration compaction parameter optimization demonstrate that this optimization method effectively reduces energy output by 127.58 J, representing 25.61% of the pre-improvement energy output. This means the method effectively reduces energy consumption during the vibration process while ensuring a 95% compaction degree, thereby improving compaction efficiency and reducing wear on the equipment. This method can be applied to intelligent vibration compaction of HSR fills and serve the optimization of vibration compaction parameters. This invention discloses a method for optimizing HSR fill vibration compaction parameters based on the principle of minimum energy. By comprehensively utilizing deep learning and optimization algorithms to adjust vibration compaction parameters in real-time during the compaction process, the method first constructs a hyperbolic model based on experiments and mathematical theory to calculate and build the original data. A dry density increment prediction model is established using a BP neural network. Then, based on the principle of minimum energy, a genetic algorithm is employed to construct a method for improving vibration parameters during the compaction process. This method adjusts compaction parameters in real-time based on the fill's compaction state, effectively improving compaction quality and efficiency, enabling the fill to reach its optimal compacted state faster and more effectively. Finally, a system for optimizing HSR fill vibration compaction parameters based on the principle of minimum energy is developed. This invention addresses the problems of traditional vibration compaction, such as the inability to adjust vibration parameters in real-time, poor control of excitation energy, the tendency of vibration compaction equipment to "jump," increased particle breakage, and uneven settlement of subgrade. The foregoing describes the relevant content of this invention. A person of ordinary skill in the art will be able to implement this invention based on these descriptions. Based on the above content of this invention, all other embodiments obtained by a person of ordinary skill in the art without creative effort shall fall within the scope of protection of this invention.

Claims

1. A parameter optimization method for high-speed railway filling vibration compaction, characterized by comprising the following steps:(1) Using a hyperbolic model to fit the dry density during the compaction process, calculating and constructing the original data;(2) Based on the acquired training data, using a BP neural network to establish a dry density increment prediction model;(3) Adopting the compaction degree index to evaluate the compaction quality and establishing compaction degree constraints;(4) Solving the vibration parameter optimization process based on the GA algorithm and establishing a GA-based dynamic optimization model;(5) Determining the dynamic optimization result after parameter optimization processing as the optimal scheme for vibration compaction;The establishment steps of the dry density increment prediction model in step (2) include: a. Constructing a high-speed railway filling vibration compaction parameter dataset, including the steps: original data input, data preprocessing, and dividing the dataset into training, validation, and test sets;b. Building a BP neural network model architecture, inputting the training and validation sets to perform model training, and determining whether the validation set error is less than the set error; If it is greater, modify the model architecture and retrain; If it is less, proceed to the next step;c. Using the test set to verify the prediction ability of the BP neural network model, and determining whether the test set error is less than the set error; If it is greater, return to step b to retrain the model; If it is less, save the final dry density increment prediction model;In step (3), the formula for the compaction degree constraint is as follows:Where pd is the dry density value after n vibrations, and pd is the maximum dry TL T1XCLXdensity;The GA-based dynamic optimization model in step (4) is obtained by the following steps:A. Setting the initial parameters of the GA algorithm and randomly initializing the population,completing the initialization configuration;B. Calling the dry density increment prediction model, iteratively calculating the dry density increment, and determining whether the individual compaction degree meets the compaction degree constraint or whether the number of iterations reaches the number of chromosomes; If not, continue calculating the dry density increment; If the condition is met, save the compaction degree sequence and then calculate the fitness value of the population individuals and sort them;C. Performing iterative optimization processes such as genetic algorithm selection, crossover, and mutation, and outputting the final result.

2. The parameter optimization method for high-speed railway filling vibrationcompaction according to claim 1, characterized in that the formula of the hyperbolic modelin step (1) is:a^Pmax - Po)n n + b+ PoWhere pd is the current state density; pmax is the dry density value at the stable compaction state under the current working condition; p0 is the initial dry density value; n is the number of vibrations; and a and b are model parameters.

3. The parameter optimization method for high-speed railway filling vibration compaction according to claim 1, characterized in that the dry density increment prediction model constructs a nonlinear functional relationship between the model input and output, and the function is expressed as follows:&Pdi = f(Anfr-PdJWhere f represents the mapping relationship, 40j and ft represent the amplitude and frequency at the current time, respectively, pdi represents the dry density at the previous time, and &Pd. represents the increment of dry density at the current time.

4. The parameter optimization method for high-speed railway filling vibration compaction according to claim 1, characterized in that the data preprocessing adopts the data normalization method, and the formula is as follows:X ^min(axis=O) xstd — ~ ~~A7nin(axis=0) xmax(axis=O) ^scaled = xstd * (max - min) + minWhere x is the data to be normalized, xmin^axis=0-) is the row vector composed of the minimum values in each column, xmax(axis=0) is the row vector composed of the maximum values in each column, max is the maximum value of the interval to be mapped to, the default is 1, min is the minimum value of the interval to be mapped to, the default is 0, xstd is the standardization result, and xscaled is the normalization result.

5. The parameter optimization method for high-speed railway filling vibration compaction according to claim 1, characterized in that the BP neural network is an optimized BP neural network, and the optimized BP neural network is obtained by optimizing the traditional BP neural network by introducing the AdamOptimizer learning rate optimizer;The AdamOptimizer improvement algorithm is a conventional optimization algorithm that dynamically adjusts the learning rate of each parameter using the first-order moment estimation and second-order moment estimation of the gradient; The improvement formula of the AdamOptimizer improvement algorithm is:7„ _ 7 "t — 1 nt *Where lrQ is the initial learning rate, lrt is the learning rate at time t, is the exponential decay rate of the first-order moment estimation, and is the exponential decay rate of the second-order moment estimation.

6. The parameter optimization method for high-speed railway filling vibrationcompaction according to claim 1, characterized in that the fitness value of the populationindividuals is calculated using the total compaction energy function, and the formula is asfollows:n nfitness = Et = 2A„Oi=l i=lWhere fitness is the fitness function, n is the number of vibrations required for the entire compaction process, f is the dry density value of the (i-1)th state, Aoi is the vibrationparameter selected for the i-th state, Et is the vibration compaction energy of the i-th state, W is the static load, and Mp is the eccentricity.

7. A parameter optimization system for high-speed railway filling vibration compaction, characterized by including a data input end, a background computing processing end, and a data display end, wherein the background computing processing end is a computing processing end that performs computing processing using the parameter optimization method for high-speed railway filling vibration compaction according to any one of claims 1-6.

Images