Parameter optimization method and system for vibration compaction of high-speed rail filler
Patent Information
- Application Number
- GB2025001208
- Authority / Receiving Office
- GB · GB
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2022-06-30
- Filing Date
- 2023-06-09
- Publication Date
- 2025-03-12
AI Technical Summary
During the vibratory compaction process of traditional high-speed railway subgrade, vibration parameters cannot be adjusted in real time, resulting in low and uneven compaction efficiency, and the excitation energy is difficult to control, which can easily lead to equipment 'jumping' and particle breakage, resulting in uneven settlement of the subgrade.
The hyperbolic model is used to fit the dry density, and a BP neural network dry density increment prediction model is established. Vibration parameters are optimized based on genetic algorithms, and compaction parameters are adjusted in real time to achieve the principle of minimum energy and improve compaction quality and efficiency.
By adjusting vibration parameters in real time, the compaction state and compaction efficiency of the filler are improved, equipment wear and particle breakage are reduced, and the problem of uneven settlement of the roadbed is avoided.
Abstract
Description
A parameter optimization method and system for vibration compaction of high-speed iron fillers Technical Field
[0001] The present invention relates to the technical field of civil engineering, and in particular to a parameter optimization method and system for vibration compaction of high-speed iron fillers. Background Art
[0002] Subgrade performance is closely related to filler composition and vibratory compaction process control. To ensure high-speed rail subgrade safety and improve train comfort, strict standards are set for subgrade filler gradation, maximum particle size, and shape. Traditional high-speed rail subgrade vibration compaction quality testing typically uses test pit sampling, which ignores the discrete nature of the filler, leading to problems such as low compaction efficiency and uneven compaction.
[0003] The compaction state of filler materials changes continuously during the compaction process. Continuously adjusting vibration parameters based on the filler compaction state, rather than using fixed parameters, can improve compaction efficiency. To achieve this goal, intelligent compaction (IC), comprising continuous compaction control (CCC) and intelligent compaction (IC), has become a recent research hotspot. Current research on vibration parameters (amplitude, frequency, and time) focuses on dynamic response under varying vibration parameters. These findings primarily explore the impact of a single vibration parameter on compaction performance, but lack research on improving vibration parameters during the compaction process. This approach falls short of meeting the requirements for real-time adjustment of vibration parameters in intelligent compaction. With the advent of stepless amplitude and frequency modulation (AMFM) rollers, continuous frequency and amplitude modulation are now possible during compaction. However, adjusting vibration parameters to maximize the next compaction degree is a partial improvement. Real-time adjustment of vibration parameters during compaction is crucial to ensure optimal filler density is achieved more efficiently and effectively. Furthermore, excessive excitation energy can easily lead to "jumping" in vibratory compaction equipment, reducing its service life. At the same time, it will also cause increased particle breakage and form uneven roadbed settlement problems.
[0004] Summary of the Invention
[0005] The main purpose of the present invention is to provide a parameter optimization method and system for vibration compaction of high-speed railway subgrade fillers, so as to solve the problems in the conventional vibration compaction process in the prior art that the compaction vibration parameters cannot be adjusted in real time, the excitation energy cannot be well controlled, and the vibration compaction equipment is prone to "jump vibration", resulting in increased particle breakage and uneven subgrade settlement.
[0006] The present invention provides a parameter optimization method for high-speed iron filler vibration compaction, comprising the following steps:
[0007] (1) Using a hyperbola model to fit the dry density of the compaction process, calculate and construct the original data;
[0008] (2) Based on the acquired training data, a dry density increment prediction model was established using a BP neural network;
[0009] (3) Use compaction index to evaluate compaction quality and establish compaction constraint conditions;
[0010] (4) Solve the vibration parameter optimization process based on the GA algorithm and establish a GA-based dynamic optimization model;
[0011] (5) The dynamic optimization result after parameter optimization is determined as the best solution for vibration compaction.
[0012] Furthermore, the formula of the hyperbolic model in step (1) is:
[0013] Where, ρ d is the current state density; ρ max is the dry density value of the compacted stable state under the current working condition; ρ0 is the dry density value of the initial state; n is the number of vibrations; a and b are model parameters.
[0014] Furthermore, the step of establishing the dry density increment prediction model in step (2) includes:
[0015] a. Construct a high-speed rail filler vibration compaction parameter dataset, preprocess the data, and divide it into training set, validation set, and test set;
[0016] b. Build the BP neural network model architecture, input the training set and validation set to perform model training, and determine whether the validation set error is less than the set error. If it is, modify the model architecture and retrain. If it is less, proceed to the next step;
[0017] c. Use the test set to verify the prediction ability of the BP neural network model and determine whether the test set error is less than the set error. If it is, return to step b to retrain the model. If it is less than, save the final dry density increment prediction model.
[0018] Furthermore, the dry density increment prediction model constructs a nonlinear functional relationship between the model input and output, and the function is expressed as follows:
[0019] Where, f is the mapping relationship, A 0i 、f i are the amplitude and frequency at the current moment, p di is the dry density at the previous moment, is the dry density increment at the current moment.
[0020] Furthermore, the data is preprocessed using a data normalization method, the formula is as follows: xscaled =x std *(max-min)+min
[0021] In the formula, x is the data to be normalized, xmin (axis = 0) is the row vector composed of the minimum value in each column, xmax (axis = 0) is the row vector composed of the maximum value in each column, max is the maximum value of the interval to be mapped, the default is 1, min is the minimum value of the interval to be mapped, the default is 0, x std is the standardized result, x scaled is the normalized result.
[0022] Furthermore, 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 a learning rate improver AdamOptimizer optimization algorithm;
[0023] The AdamOptimizer improved algorithm is an improved algorithm for the conventional optimization algorithm that uses the first-order moment estimation and second-order moment estimation of the gradient to dynamically adjust the learning rate of each parameter. The improved formula of the AdamOptimizer improved algorithm is:
[0024] Where, l r0 is the initial learning rate, lr t is the learning rate at time t, is the exponential decay rate of the first-order moment estimate, is the exponential decay rate of the second-order moment estimate.
[0025] Furthermore, in step (3), the formula for the compaction constraint condition is as follows:
[0026] Where, is the dry density value after vibration n times, is the maximum dry density.
[0027] Furthermore, the GA-based dynamic optimization model in step (4) is obtained by the following steps:
[0028] Set the initial parameters of the GA algorithm and randomly initialize the population to complete the initial configuration;
[0029] Iteratively calculate the dry density increment to determine whether the individual compaction and the number of iterations meet the requirements. If not, continue to calculate the dry density increment. If the conditions are met, save the compaction sequence, and then calculate the fitness value of the population individuals and sort them.
[0030] Execute iterative optimization processes such as genetic algorithm selection, crossover, mutation, etc., and output the final results.
[0031] Furthermore, the fitness value of each individual in the population is calculated using the compacted total energy function, which is as follows:
[0032] Where fitness is the fitness function, n is the number of vibrations required for the entire compaction process, fi is the dry density value of the i-1 state, A0i is the vibration parameter selected for the i-th state, E i is the vibration compaction energy of the i-th state, W is the static load, M p is the eccentricity.
[0033] A parameter optimization system for vibration compaction of high-speed iron fillers is also provided, including a data input end, a background operation processing end and a data display end. The background operation processing end is an operation processing end that performs operation processing using a parameter optimization method for vibration compaction of high-speed iron fillers as described in any one of claims 1 to 9.
[0034] The present invention discloses a method for optimizing the vibration compaction parameters of high-speed iron fillers based on the principle of minimum energy. By integrating a deep learning algorithm and a target optimization algorithm, the vibration compaction parameters are improved in real time during the compaction process. First, a hyperbola model is constructed through experiments and mathematical theories to calculate and construct the original data. A dry density increment prediction model is established using a BP neural network. Then, a genetic algorithm is used to construct a method for improving the vibration parameters of the compaction process based on the principle of minimum energy. The compaction parameters are adjusted in real time based on the compaction state of the filler, effectively improving the compaction quality and compaction efficiency, so that the filler can reach the optimal dense state better and faster. Finally, a high-speed iron filler vibration compaction parameter optimization system based on the principle of minimum energy is developed. The method of the present invention solves the problems that the compaction vibration parameters cannot be adjusted in real time during the traditional vibration compaction process, the excitation energy cannot be well controlled, the vibration compaction equipment is prone to "jump vibration" phenomenon, resulting in increased particle breakage and the formation of uneven roadbed settlement diseases.
[0035] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. Additional aspects and advantages of the present invention will be partially given in the following description, partially become apparent from the following description, or be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] The drawings that constitute part of this invention are intended to assist in understanding the invention. The contents provided in the drawings and their related descriptions in the present invention may be used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:
[0037] Figure 1 is a flow chart of a parameter optimization method for vibration compaction of high-speed iron fillers.
[0038] Figure 2 is a flow chart for building a dry density increment prediction model based on the improved BP neural network.
[0039] Figure 3 shows the BP neural network model architecture based on the improved BP algorithm.
[0040] Figure 4 is a flow chart of the GA dynamic optimization model.
[0041] Figure 5 shows the 6-bit binary encoding diagram of the variable.
[0042] Figure 6 is the system interface of the high-speed rail filler vibration compaction parameter optimization system based on the energy minimization principle.
[0043] Figure 7 shows the loss function curves of the training set and validation set for the two types of dry density increment models.
[0044] Figure 8 is a comparison of the test results of the dry density increment prediction model before and after improvement.
[0045] Figure 9 shows the final vibration parameter optimization results.
[0046] Figure 10 shows the vibration optimization parameters and compaction curves at different time periods. DETAILED DESCRIPTION
[0047] 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 present invention based on these descriptions. Before describing the present invention with reference to the accompanying drawings, it should be noted that:
[0048] The technical solutions and technical features provided in each part of the present invention, including the following description, may be combined with each other unless there is any conflict.
[0049] In addition, the embodiments of the present invention described below are generally only part of the embodiments of the present invention, rather than all of the embodiments. Therefore, based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making any creative efforts should fall within the scope of protection of the present invention.
[0050] Regarding the terms and units in the present invention: The terms "include", "have" and any variations thereof in the description and claims of the present invention and the related parts are intended to cover non-exclusive inclusions.
[0051] The steps of a parameter optimization method for vibration compaction of high-speed iron filler are shown in Figure 1:
[0052] Step (1): collecting and constructing raw data, specifically using a hyperbola model to fit the dry density of the compaction process, and calculating and constructing raw data;
[0053] Step (2) establishing a dry density increment prediction model, specifically, using a BP neural network to establish a dry density increment prediction model based on the acquired training data;
[0054] Step (3) establishing constraint conditions, specifically taking a compaction index to evaluate compaction quality and establishing compaction constraint conditions;
[0055] Step (4) establishing a dynamic optimization model based on GA, specifically solving the vibration parameter optimization process based on the GA algorithm and establishing a dynamic optimization model based on GA;
[0056] Step (5) outputs the best solution, specifically determining the dynamic optimization result after parameter optimization as the best solution for vibration compaction.
[0057] As shown in Figure 2, the construction of the dry density increment prediction model based on the improved BP neural network includes, in sequence, data set construction, model training, and model testing. First, the data set is constructed, which includes, in sequence, the following steps: raw data input, data preprocessing, and data set division, where the data set is divided into a test set, a validation set, and a training set; then, model training is performed, which includes, in sequence, the steps of optimizing the BP neural network architecture, continuing training at breakpoints, and then determining whether the validation set prediction error is less than the set error value. If not, returning to optimizing the BP neural network architecture construction; if so, entering model testing; finally, model testing is performed, which includes, in sequence, the steps of saving the model, calling the model, and model testing. It is determined whether the test error is less than the set error value. If so, the final model is saved; if not, the model is returned to model training.
[0058] Based on the acquired training data, the improved BP neural network is used to establish a dry density increment prediction model.
[0059] Use compaction index to evaluate compaction quality and establish compaction constraint conditions to ensure the compaction quality of fillers;
[0060] The vibration parameter optimization process is solved based on the GA algorithm. A dynamic optimization model based on GA is established based on the principle of minimizing the energy of the vibration compaction process to achieve real-time improvement and optimization of the vibration compaction parameters.
[0061] The dynamic optimization results after parameter optimization are determined as the best solution for vibration compaction.
[0062] In the step (1), the calculation and construction of the original data are mainly divided into three steps: 1) determining the hyperbolic function relationship between the dry density and the vibration parameters during the vibration compaction process of the high-speed iron filler; 2) determining the model parameters a and b through experiments; 3) calculating and obtaining the dry density data set under different vibration frequencies f and amplitudes A and the corresponding dry densities p, and using them as the original data.
[0063] In the step 1), determining the hyperbolic function relationship between the dry density and the vibration parameters during the vibration compaction of the high-speed iron filler is divided into three steps:
[0064] ① Set the basic equation of the hyperbolic function as follows:
[0065] In the formula, a and b are variable parameters.
[0066] ② In order to eliminate the dimensional relationship between different variables, the dry density is normalized. The processing process is as follows:
[0067] Where y * are the standardized values; y, y min 、y max are the actual value of the variable, the maximum value, and the minimum value of the standardized interval, respectively.
[0068] ③Substitute the above formula into the formula in ① to obtain the fitting model of dry density and vibration number, as shown below:
[0069] Where, ρ d is the current state density; ρ max is the dry density value of the compacted stable state under the current working condition; ρ0 is the dry density value of the initial state; n is the number of vibrations; a and b are model parameters.
[0070] In the step 2), the model parameters a and b are determined by experiment, and the model parameters and fitting correlation coefficients under various working conditions are obtained as shown in Table 1 below:
[0071] Table 1 Model parameters and correlation coefficients
[0072] As shown in FIG2 , in step (2), the construction of the dry density increment prediction model is mainly divided into three steps:
[0073] ①'Construct a high-speed rail filler vibration compaction parameter dataset. First, preprocess the data and normalize it to [0, 1]. The specific formula is as follows; and randomly divide the dataset into training set, validation set, and test set according to the data volume ratio of 70%, 15%, and 15%. x scale d=x std *(max-min)+min
[0074] In the formula, x is the data to be normalized, xmin (axis = 0) is the row vector composed of the minimum value in each column, xmax (axis = 0) is the row vector composed of the maximum value in each column, max is the maximum value of the interval to be mapped, the default is 1, min is the minimum value of the interval to be mapped, the default is 0, x std is the standardized result, x scaled is the normalized result.
[0075] ②'Build an optimized BP neural network model architecture, input the training set and validation set to perform model training, and 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
[0076] As shown in FIG3 , in the step ②′, the optimized BP neural network model architecture can be divided into an input layer, a hidden layer, and an output layer. The number of neurons in the input layer is 3, and the amplitude A at the current moment is input. 0i , frequency f i and the dry density p at the previous moment di , the number of neurons in the output layer is 1, which outputs the dry density increment at the current moment Through the dry density increment prediction model, a nonlinear functional relationship between the model input and output can be constructed. The function is expressed as follows:
[0077] Where, f is the mapping relationship, A 0i 、f i are the amplitude and frequency at the current moment, p di is the dry density at the previous moment, is the dry density increment at the current moment
[0078] In the step ②': the improved BP neural network algorithm has the characteristics of fast convergence speed and high prediction accuracy compared with the traditional BP algorithm. The learning rate improver AdamOptimizer optimization algorithm is introduced on the basis of the traditional BP neural network, which can improve the problems of the traditional BP algorithm that easily fall into local optimality, sample dependence and unadjustable learning rate. The learning rate is reasonably adjusted according to the change of the loss function, thereby accelerating the model convergence speed and improving the prediction accuracy.
[0079] In step ②', the AdamOptimizer algorithm is an optimized version of a conventional optimization algorithm (e.g., SGD, AdaGrad). It adds first-order momentum and second-order momentum to the SGD algorithm, and adds a correction bias to the AdaGrad algorithm. It has the characteristics of fast convergence and high stability. It also uses the first-order moment estimation and second-order moment estimation of the gradient to dynamically adjust the learning rate of each parameter. The main improved formula of the AdamOptimizer algorithm is as follows:
[0080] Where, l r0 is the initial learning rate, lr t is the learning rate at time t, is the exponential decay rate of the first-order moment estimate, is the exponential decay rate of the second-order moment estimate.
[0081] ③' Use the divided test set to verify the prediction ability of the trained optimized BP neural network model. Use MSE and MAE (as shown in the following formula) to determine whether the test set error is less than the set error. If it is greater, return to step 2) to retrain the model. If it is less, save the final dry density increment prediction model.
[0082] Where N represents the number of prediction samples, y t and represent the actual value and the predicted value respectively.
[0083] In the step (3), the compaction index is used to evaluate the compaction quality. According to the requirement of the "High-speed Railway Roadbed Specifications" that the compaction of the subgrade layer must reach more than 95%, the individual compaction should meet the constraint condition of greater than 0.95 to ensure the compaction quality of the filler. The formula is as follows:
[0084] Where, is the dry density value after vibration n times, is the maximum dry density.
[0085] As shown in FIG4 , in step (4), the GA-based dynamic optimization model is obtained by the following steps:
[0086] The initial parameters of the GA algorithm, such as population size (PS), maximum genetic generations (MAXGEN), selection probability SP, crossover probability CP, and mutation probability Mu, are set, and the population is randomly initialized using binary coding. Each chromosome in the population includes two genes (vibration frequency and amplitude), as shown in Figure 5. Each variable occupies 6 bits of binary coding.
[0087] Iteratively calculate the dry density increment to determine whether the individual compaction degree is greater than 0.95 or whether the number of iterations reaches the number of chromosomes. If not, continue to calculate the dry density increment. If the conditions are met, save the compaction degree sequence, and then calculate the fitness value of the population individuals based on the total compaction energy function (as shown in the following formula), and sort them according to the minimum energy principle;
[0088] Where fitness is the fitness function, n is the number of vibrations required for the entire compaction process, fi is the dry density value of the i-1 state, A0i is the vibration parameter selected for the i-th state, E i is the vibration compaction energy of the i-th state, W is the static load, M p is the eccentricity.
[0089] Execute iterative optimization processes such as genetic algorithm selection, crossover, and mutation, and output the final vibration parameter optimization results;
[0090] The embodiment and implementation process of the complete method according to the present invention are as follows:
[0091] Calculate and construct original data:
[0092] The dry density of the compaction process was fitted using a hyperbola model to obtain the original data, which amounted to 7500 groups. Some of the data are shown in Table 2 below.
[0093] Table 2
[0094] Establish a dry density increment prediction model:
[0095] First, the dry density increment is calculated based on the current dry density, and the data is normalized as a whole. Part of the processed data is shown in Table 3 below.
[0096] Table 3
[0097] Next, the dataset was randomly divided into three parts: training, validation, and test sets, with data volumes accounting for 70%, 15%, and 15%, respectively. Using Python 3.8.1, the Keras deep learning library and other third-party libraries were imported. The number of input layer neurons was set to 3, the number of hidden layer neurons to 100, the number of output layer neurons to 1, the initial learning rate to 0.001, and the number of training iterations to 100. An improved BP neural network dry density increment prediction model architecture was constructed. The training and validation sets were then imported for model training. The model was then compared with a traditional BP neural network model. The changes in the training set loss (abbreviated as loss) and validation set loss (abbreviated as val_loss) during training are shown in Figure 7. It is clear that the traditional BP neural network model's loss and val_loss decrease slowly and oscillate continuously during training, with the loss curve still not converging at the end of 100 epochs. In contrast, the improved BP neural network model's loss and val_loss decrease very rapidly during training. After the 5th round of iteration, loss and val_loss began to converge, and no oscillation occurred. This is because the improved BP neural network introduced the momentum method, which achieved the purpose of accelerated convergence. At the same time, during the entire iterative operation process, val_loss was less than loss. At the end of the iteration, loss and val_loss had converged, and both values were close to 0, indicating that the improved BP neural network model had a good training effect and had good generalization ability on the validation set. In summary, the improved BP neural network model proposed in the present invention has the characteristics of adjustability and fast convergence, which solves the problem that the traditional BP neural network model converges slowly and is prone to falling into local optimal points.
[0098] Finally, the improved BP neural network dry density increment prediction model trained above was saved, and the 1125 test sets that did not participate in the training were input into the dry density increment prediction model for model testing. The prediction results were evaluated using MSE and MAE, and compared with the traditional BP neural network model. The comparison between the prediction results and the actual values is shown in Figure 8. It is easy to see that the traditional BP neural network model can predict most of the data well, but the prediction effect is not good for some prominent points of dry density increment; while the improved BP neural network model has a strong prediction ability and can well predict the fluctuation trend of dry density increment. The MSE and MAE values calculated by the improved BP neural network model are 4.5×10 -6 and 1.4×10 -3 , while the MSE and MAE values calculated by the traditional BP neural network model are 6.35×10 -4 and 5.1×10 -3, indicating that the improved BP neural network has higher prediction accuracy than the traditional BP neural network. In summary, the improved BP neural network model proposed in this paper not only performs well on the training set and test set, but also has high prediction accuracy and strong generalization ability on the test set, making it well suited for the prediction of dry density increment.
[0099] Finally, the final dry density increment prediction model based on the improved BP algorithm is saved and applied in the subsequent dynamic optimization process.
[0100] Establish a compaction constraint above 0.95.
[0101] Establish a dynamic optimization model based on GA.
[0102] First, we implemented the GA algorithm in Python 3.8.1. We then analyzed the genetic algorithm parameters experimentally to determine their optimal parameters. We set the optimal population size (PS) to 150, the maximum number of generations (MAXGEN) to 200, the selection probability (SP) to 0.9, the crossover probability (CP) to 0.6, and the mutation probability (Mu) to 0.05. Using binary encoding, we constructed 500 chromosomes, each containing two genes (vibration frequency and nominal amplitude), with each variable occupying a 6-bit binary code.
[0103] Secondly, the dry density increment prediction model is called to iteratively calculate the dry density increment to determine whether the individual compaction degree is greater than 0.95 or whether the number of iterations reaches the number of chromosomes. If not, the dry density increment calculation is continued. If the conditions are met, the compaction degree sequence is saved, and the fitness value of the population individual is calculated based on the total compaction energy function, and they are sorted according to the minimum energy principle.
[0104] Finally, an iterative optimization process consisting of selection, crossover, and mutation was performed. After reaching the maximum number of iterations, the final vibration parameter optimization results were output, as shown in Figure 9. Nine complete vibrations were required to achieve the specified compaction degree, i.e., a cumulative total of 180 vibration compactions. As shown in Figure 9(a), the optimized vibration frequency gradually increased from the initial 10.15 Hz to 27.58 Hz during the vibration compaction process. This optimization result is consistent with the existing two-degree-of-freedom vibration compaction model for the optimal vibration frequency relationship between the compaction device and the soil. Figure 9(b) shows the improved nominal amplitude adjustment scheme. It can be seen that the optimized vibration amplitudes are all smaller than the amplitude scheme before the improvement. As the vibration compaction process increases, the amplitude increases from 0.15 mm. After 120 consecutive vibrations, the amplitude change gradually stabilizes, ultimately reaching 0.34 mm. Figure 9(c) shows the comparative relationship between the energy output before and after the improvement. It can be seen that the energy output after the improvement is less than the energy output before the improvement throughout the compaction process, and ultimately effectively reduces the energy by 127.58 J, accounting for 25.61% of the energy output before the improvement. Therefore, the dynamic optimization method based on the GA algorithm proposed in the present invention can select appropriate vibration parameters according to the current soil density state, and effectively reduce the energy in the vibration process while ensuring a compaction degree of 95%, which can effectively improve the compaction efficiency and reduce the wear on the instrument.
[0105] The dynamic optimization result after parameter optimization is determined as the best solution for vibration compaction, as shown in Table 4 below.
[0106] Table 4 Vibration parameter optimization scheme
[0107] Accordingly, the parameters are input into the vibration compaction parameter optimization system. After the system is run, the vibration optimization parameters and compaction curves for each time period can be output on the interface, as shown in Figure 10.
[0108] The test results based on the optimization of vibration compaction parameters of high-speed iron fillers show that under the action of this optimization method, the energy output is effectively reduced by 127.58J, accounting for 25.61% of the energy output before improvement. That is, this method effectively reduces the energy in the vibration process while ensuring a compaction degree of 95%, which can effectively improve the compaction efficiency and reduce the wear on the instrument. It can be applied to the intelligent vibration compaction of high-speed iron fillers and serve the optimization application of vibration compaction parameters.
[0109] The present invention discloses a method for optimizing the vibration compaction parameters of high-speed iron fillers based on the principle of minimum energy. By integrating a deep learning algorithm and a target optimization algorithm, the vibration compaction parameters are improved in real time during the compaction process. First, a hyperbola model is constructed through experiments and mathematical theories to calculate and construct the original data. A dry density increment prediction model is established using a BP neural network. Then, a genetic algorithm is used to construct a method for improving the vibration parameters of the compaction process based on the principle of minimum energy. The compaction parameters are adjusted in real time based on the compaction state of the filler, effectively improving the compaction quality and compaction efficiency, so that the filler can reach the optimal dense state better and faster. Finally, a high-speed iron filler vibration compaction parameter optimization system based on the principle of minimum energy is developed. The method of the present invention solves the problems that the compaction vibration parameters cannot be adjusted in real time during the traditional vibration compaction process, the excitation energy cannot be well controlled, the vibration compaction equipment is prone to "jump vibration" phenomenon, resulting in increased particle breakage and the formation of uneven roadbed settlement diseases.
[0110] The above describes the relevant contents of the present invention. Based on this description, a person skilled in the art will be able to implement the present invention. Based on the above contents of the present invention, all other embodiments obtained by a person skilled in the art without making any creative efforts should fall within the scope of protection of the present invention.
Claims
1. A method for optimizing parameters of vibration compaction of high-speed rail fillers, characterized in that: It includes the following steps: (1) Fit the dry density during the compaction process using the hyperbolic model, and calculate and construct the original data; (2) Based on the obtained training data, establish a dry density increment prediction model using the BP neural network; (3) Evaluate the compaction quality using the compaction degree index, and establish the compaction degree constraint conditions; (4) Solve the vibration parameter optimization process based on the GA algorithm, and establish a GA-based dynamic optimization model; (5) Determine the dynamic optimization result after parameter optimization as the best vibration compaction plan; The steps for establishing the dry density increment prediction model in step (2) include: a. Construct a vibration compaction parameter dataset for high-speed rail fillers, including steps: inputting the original data, preprocessing the data, and dividing it into a training set, a validation set, and a test set; b. Build the BP neural network model architecture, input the training set and the validation set to perform model training, and 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; c. Use the test set to verify the prediction ability of the BP neural network model, and determine 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 the said step (3), the formula for the compaction degree constraint condition is as follows: In the formula, is the dry density value after vibrating n times, is the maximum value of the dry density; The GA-based dynamic optimization model in step (4) is obtained through the following steps: A. Set the initial parameters of the GA algorithm and randomly initialize the population to complete the initialization configuration; B. Call the dry density increment prediction model, iteratively calculate the dry density increment, and determine whether the individual compaction degree meets the compaction degree constraint conditions or whether the number of iterations reaches the number of chromosomes. If it does not meet, continue to calculate the dry density increment. If the conditions are met, save the compaction degree sequence, and then calculate and sort the fitness values of the population individuals; C. Execute the iterative optimization processes of genetic algorithm selection, crossover, mutation, etc., and output the final result.
2. The parameter optimization method for vibration compaction of high-speed rail fillers according to claim 1, characterized in that: The formula of the hyperbola model in the step (1) is as follows: where ρ d is the current state density; ρ max is the dry density value of the compaction stable state under the current working condition; ρ0 is the initial state is the dynamic dry density value; n is the number of vibrations; a and b are model parameters.
3. The parameter optimization method for vibration compaction of high-speed rail fillers according to claim 1, characterized in that: The dry density increment prediction model constructs a non-linear functional relationship between the model input and output, and the function is expressed as follows: where f is the mapping relationship, A 0i , f i are the amplitude and frequency at the current moment respectively, and p di is the dry density at the previous moment, is the dry density increment at the current moment.
4. A method for optimizing the parameters of vibration compaction of high-speed rail fillers according to claim 1, characterized in that: The data preprocessing adopts the data normalization method, and its formula is as follows: x scaled = x std *(max - min)+min 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 = 0) is the row vector composed of the maximum values in each column, max is the maximum value of the interval to which it is to be mapped, the default is 1, min is the minimum value of the interval to which it is to be mapped, the default is 0, and x std is the standardized result, and x scaled is the normalization result.
5. The parameter optimization method for vibration compaction of high-speed rail fillers 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 through the introduction of the learning rate improver AdamOptimizer optimization algorithm; The improved AdamOptimizer algorithm is an improved algorithm that uses the first-order moment estimation and second-order moment estimation of the gradient to dynamically adjust the learning rate of each parameter. The improved formula of the AdamOptimizer improved algorithm is as follows: where l r0 is the initial learning rate, lr t is the learning rate at time t, is the exponential decay rate of the first moment estimate, is the exponential decay rate of the second-order moment estimation.
6. The parameter optimization method for vibration compaction of high-speed railway fillers according to claim 1, characterized in that: The fitness value of each individual in the population is calculated using the compaction total energy function, and its formula is as follows: where fitness is the fitness function, n is the number of vibrations required for the entire compaction process, fi is the dry density value at the (i - 1)th state, A0i is the vibration parameter selected for the ith state, E i is the vibration compaction energy at the ith state, W is the static load, M p is the eccentricity.
7. A parameter optimization system for vibration compaction of high-speed rail fillers, characterized in that, It includes a data input end, a background operation processing end, and a data display end. The background operation processing end is an operation processing end that performs operation processing using a method for optimizing the parameters of vibration compaction of high-speed rail fillers according to any one of claims 1-6.