Multi-factor roadbed settlement prediction method based on deep learning model ECA-GRU

Abstract

The invention relates to the technical field of roadbed prediction, and discloses a multi-factor roadbed settlement prediction method based on a deep learning model ECA-GRU. According to the multi-factor roadbed settlement prediction method based on the deep learning model ECA-GRU, weight parameters and offset parameters in the model can be effectively optimized by applying an Adam optimizer, so that the purpose of minimizing loss is achieved, meanwhile, an effective model can be obtained by training the model, the ability of the model is evaluated by utilizing the model, and the prediction accuracy of roadbed settlement is improved. Compared with other models, the method can overcome the defects that a theoretical model adopts various assumptions, the universality is poor, and advanced model parameters are difficult to calibrate.

Description

Technical Field

[0001] The present application relates to the technical field of roadbed prediction, and in particular to a multi-factor roadbed settlement prediction method based on a deep learning model ECA-GRU. Background Art

[0002] The roadbed is the main body of the high-speed railway track structure and the train. In order to achieve high safety of train operation and high smoothness of the high-speed railway track structure, the post-construction settlement of the ballastless track high-speed railway roadbed in general sections is not more than 20mm, and the post-construction settlement of the ballasted track high-speed railway roadbed in general sections is not more than 50mm. This is a very big challenge for geotechnical engineering. Therefore, before the construction of the high-speed railway track project, a strict post-construction settlement assessment of the roadbed must be carried out.

[0003] At present, the post-construction settlement assessment of high-speed railway subgrade is to install monitoring equipment in the subgrade, collect settlement data through long-term observation, draw the "fill height-time-settlement" curve, and perform regression fitting on the post-construction settlement data based on methods such as hyperbola method, exponential curve method, Asaoka method or gray theory, and then predict the long-term post-construction settlement of the subgrade. These methods play an important role in the post-construction settlement assessment of high-speed railway subgrade, but are greatly affected by long-term fluctuations. In actual projects, the geotechnical engineering environment of the subgrade is complex and changeable, and the settlement is affected by factors such as temperature, groundwater and load changes. The subgrade in the loess area is mostly in an unsaturated state, and it is easy to have excessive post-construction settlement under dry-wet cycles and freeze-thaw cycles. However, the traditional single-factor regression fitting settlement prediction models such as hyperbola method, exponential curve method, Asaoka method have obvious time dependence and cannot consider the influence of factors such as temperature and precipitation. Although the gray theory method can make up for the shortcomings of poor information and uncertainty, it has volatility.

[0004] Practice has proved that the single factor regression fitting settlement prediction model has a high accuracy in predicting the post-construction settlement of soft soil roadbed, while the prediction results of post-construction settlement of loess roadbed still require long-term observation and evaluation. For the millimeter-level roadbed settlement control of high-speed railways, the roadbed post-construction settlement prediction method still needs to be improved, and it is very necessary to improve its prediction accuracy. Although deep learning models have been used for roadbed settlement prediction, multi-factor deep learning settlement prediction methods and systems that consider environmental and geotechnical properties are still blank. The influence of environmental and geotechnical properties at different time series on roadbeds is different, and the existing deep learning settlement prediction methods have not yet realized the calculation of the influence of environmental and geotechnical properties on settlement. Summary of the invention

[0005] This application proposes a multi-factor roadbed settlement prediction method based on the deep learning model ECA-GRU, which has the advantages of overcoming the defects of theoretical models adopting multiple assumptions, poor universality and difficulty in calibrating advanced model parameters, and is used to solve the problems raised by the background technology.

[0006] To achieve the above objectives, the present application adopts the following technical solution: a multi-factor roadbed settlement prediction method based on a deep learning model ECA-GRU, comprising the following steps:

[0007] S1. Establish the original data set and preprocess the data, that is, standardize the data. The calculation method formula for standardization is as follows:

[0008]

[0009] Among them, x represents the original data, x * represents the standardized data, x min and x max Respectively represent the minimum and maximum values ​​of the feature;

[0010] S2. Establish a deep learning model through GRU;

[0011] S3, carrying the attention module through the ECAnet attention mechanism;

[0012] S4, model training, the training steps include:

[0013] a. Parameter initialization: Parameters such as weights and learning rates need to be set before they can be correctly applied to the model;

[0014] b. Establish the loss function and set the stopping condition: The loss function is an important learning criterion that can be used to measure the performance of a network and is often related to the optimization amount. The mean square error loss function is used to calculate the mean square error between them. The calculation formula is as follows:

[0015]

[0016] Where N is the length of the input sequence, is the predicted value of the model, y i is the true value;

[0017] c. By applying the Adam optimizer, the weight parameters and bias parameters in the model can be effectively optimized;

[0018] d. By carefully adjusting the parameters and functions, an effective network training process can be achieved. That is, according to the pre-set number of training times, the parameters of the model are continuously adjusted until a certain threshold is reached. In this way, even if the expected number of training times is not reached, the model can stop running when the threshold is met. After 3 epochs, if the loss function has not been significantly improved, the deep learning model will be terminated.

[0019] Furthermore, the GRU in S2 processes the data information, and the basic architecture includes:

[0020] Update gate formula: z t =g(W (x) x t +U (x) h t-1 )

[0021] Reset gate formula: r t =g(W (N) x t +U (r) h t-1 )

[0022] Current memory content: h' i =tanh(Wx i +r i ) i-1 )

[0023] Current time final memory: h t =z t ⊙h t-1 +(1-z t )⊙h.

[0024] Furthermore, the implementation steps of the ECAnet attention mechanism in S3 are as follows:

[0025] (1) Global average pooling: Perform a global average pooling operation on the input feature map, compressing it from the [h, w, c] dimension to a [1, 1, c] vector;

[0026] (2) Adaptive kernel size: An adaptive convolution kernel size is calculated based on the dimension of the feature map to capture key information between channels;

[0027] (3) Apply 1*1 convolution: Use the calculated convolution kernel size to process the vector obtained in the previous step in a 1x1 convolution layer to generate a weight value for each channel;

[0028] (4) Generate weighted feature map: Multiply these weights with the original input feature map channel by channel to generate the final weighted feature map.

[0029] The multi-factor roadbed settlement prediction method based on the deep learning model ECA-GRU provided in this application can effectively optimize the weight parameters and bias parameters in the model by applying the Adam optimizer, so as to achieve the purpose of minimizing the loss. At the same time, by training the model, an effective model can be obtained, and these models can be used to evaluate their capabilities for comparison with other models. This method can overcome the defects of theoretical models adopting multiple assumptions, poor universality and difficulty in calibrating advanced model parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings, which constitute a part of the specification, illustrate embodiments disclosed in the present application and, together with the description, serve to explain the principles disclosed in the present application.

[0031] The present disclosure may be more clearly understood from the following detailed description with reference to the accompanying drawings, in which:

[0032] Figure 1 It is a schematic diagram of the GRU of the present invention;

[0033] Figure 2 It is a schematic diagram of ECAnet of the present invention. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0035] See also Figure 1-2 ,The multi-factor roadbed settlement prediction method based on the deep learning model ECA-GRU includes the following steps:

[0036] S1. Establish the original data set and preprocess the data, that is, standardize the data. The calculation method formula for standardization is as follows:

[0037]

[0038] Among them, x represents the original data, x * represents the standardized data, x min and x max Respectively represent the minimum and maximum values ​​of the feature;

[0039] S2. Establish a deep learning model through GRU;

[0040] S3, carrying the attention module through the ECAnet attention mechanism;

[0041] S4, model training, the training steps include:

[0042] a. Parameter initialization: Parameters such as weights and learning rates need to be set before they can be correctly applied to the model;

[0043] b. Establish the loss function and set the stopping condition: The loss function is an important learning criterion that can be used to measure the performance of a network and is often related to the optimization amount. The mean square error loss function is used to calculate the mean square error between them. The calculation formula is as follows:

[0044]

[0045] Where N is the length of the input sequence, is the predicted value of the model, y i is the true value;

[0046] c. By applying the Adam optimizer, the weight parameters and bias parameters in the model can be effectively optimized;

[0047] d. By carefully adjusting the parameters and functions, an effective network training process can be achieved. That is, according to the pre-set number of training times, the parameters of the model are continuously adjusted until a certain threshold is reached. In this way, even if the expected number of training times is not reached, the model can stop running when the threshold is met. After 3 epochs, if the loss function has not been significantly improved, the deep learning model will be terminated.

[0048] Among them, the GRU in S2 processes the data information, and the basic architecture is as follows:

[0049] Update gate formula: z t =g(W (x) x t +U (x) h t-1 )

[0050] Reset gate formula: r t =g(W (N) x t +U (r) h t-1 )

[0051] Current memory content: h' i =tanh(Wx i +r i ) i-1 )

[0052] Current time final memory: h t =z t ⊙h t-1+(1-z t )⊙h.

[0053] The implementation steps of the ECAnet attention mechanism in S3 are as follows:

[0054] (1) Global average pooling: Perform a global average pooling operation on the input feature map, compressing it from the [h, w, c] dimension to a [1, 1, c] vector;

[0055] (2) Adaptive kernel size: An adaptive convolution kernel size is calculated based on the dimension of the feature map to capture key information between channels;

[0056] (3) Apply 1*1 convolution: Use the calculated convolution kernel size to process the vector obtained in the previous step in a 1x1 convolution layer to generate a weight value for each channel;

[0057] (4) Generate weighted feature map: Multiply these weights with the original input feature map channel by channel to generate the final weighted feature map.

[0058] LSTM can capture long-term dependencies and is very suitable for processing time series problems. However, its complex internal structure leads to a long training time. The GRU model has a similar structure to LSTM, but only contains update gates and reset gates, which reduces the number of parameters and converges faster. The function of the update gate is to determine how much historical information from the previous moment is retained to the current state. The reset gate affects the combination of the current state and the previous state information. GRU maintains the memory of neurons while increasing the training speed. The update gate controls how much previous information the model passes to subsequent processing, while the reset gate determines how much previous information is discarded. In this process, new information and old information are complementary. If more new information is retained, correspondingly, less original information will be considered, and vice versa.

Claims

1. A multi-factor roadbed settlement prediction method based on the deep learning model ECA-GRU is characterized by: The method comprises the following steps: S1. Establish the original data set and preprocess the data, that is, standardize the data. The calculation method formula for standardization is as follows: Among them, x represents the original data, x * represents the standardized data, x min and x max Respectively represent the minimum and maximum values ​​of the feature; S2. Establish a deep learning model through GRU; S3, carrying the attention module through the ECAnet attention mechanism; S4, model training, the training steps include: a. Parameter initialization: Parameters such as weights and learning rates need to be set before they can be correctly applied to the model; b. Establish the loss function and set the stopping condition: The loss function is an important learning criterion that can be used to measure the performance of a network and is often related to the optimization amount. The mean square error loss function is used to calculate the mean square error between them. The calculation formula is as follows: Where N is the length of the input sequence, is the predicted value of the model, y i is the true value; c. By applying the Adam optimizer, the weight parameters and bias parameters in the model can be effectively optimized; d. By carefully adjusting the parameters and functions, an effective network training process can be achieved. That is, according to the pre-set number of training times, the parameters of the model are continuously adjusted until a certain threshold is reached. In this way, even if the expected number of training times is not reached, the model can stop running when the threshold is met. After 3 epochs, if the loss function has not been significantly improved, the deep learning model will be terminated.

2. The multi-factor roadbed settlement prediction method based on the deep learning model ECA-GRU according to claim 1 is characterized in that: The GRU in S2 processes the data information, and the basic architecture is as follows: Update gate formula: z t =g(W (x) x t +U (x) h r-1 ) Reset gate formula: r t =g(W (N) x t +U (r) h t-1 ) Current memory content: h' i =tanh(Wx i +r i ) i-1 ) Current time final memory: h t =z t ⊙h t-1 +(1-z t )⊙h.

3. The multi-factor roadbed settlement prediction method based on the deep learning model ECA-GRU according to claim 1 is characterized in that: The implementation steps of the ECAnet attention mechanism in S3 are as follows: (1) Global average pooling: Perform a global average pooling operation on the input feature map, compressing it from the [h, w, c] dimension to a [1, 1, c] vector; (2) Adaptive kernel size: An adaptive convolution kernel size is calculated based on the dimension of the feature map to capture key information between channels; (3) Apply 1*1 convolution: Use the calculated convolution kernel size to process the vector obtained in the previous step in a 1x1 convolution layer to generate a weight value for each channel; (4) Generate weighted feature map: Multiply these weights with the original input feature map channel by channel to generate the final weighted feature map.

Images