Method for performing tunnel deformation data filling on basis of improved gain

Abstract

The present invention relates to the field of tunnel deformation monitoring data filling. Disclosed are a method for performing tunnel deformation data filling on the basis of an improved GAIN. The method comprises: an auto-encoder performing an initial imputation method Autoencoders-LSGAIN on an LSGAIN, wherein Autoencoders is composed of two parts, that is, an encoder and a decoder, the encoder is a function which converts original data into a code, and the encoder comprises a full-connection neural network and a convolutional neural network; and the decoder reversely restores the code into an input, so as to generate an output most similar to the original data. The tunnel deformation data filling method performed on the basis of an improved GAIN that is used in the present invention improves the convergence speed of a GAIN method, and shortens a calculation time required for filling, such that a filling process can be completed within a relatively short time, and thus the overall efficiency of data processing can be improved, and subsequent analysis and modeling work can be developed more quickly, and the method has good performance in tunnel deformation monitoring data filling.

Description

A tunnel deformation data filling method based on improved GAIN Technical Field

[0001] The present invention relates to the field of tunnel deformation data filling, and in particular to a tunnel deformation data filling method based on improved GAIN. Background Art

[0002] Methods for handling missing data generally include deletion and completion. Deletion involves deleting samples or variables with missing data. While this method maintains formal integrity and authenticity, it loses a significant amount of useful information, adversely affecting data analysis. Completion, based on traditional statistical theories such as single-value interpolation using unconditional mean, median, mode, and zero, fills in the missing data with estimated values ​​to maintain data consistency. Yong Huang et al. proposed using Bayesian compressed sensing to recover health monitoring data. Yongchao Yang and Satish Nagarajaiah proposed using sparse recovery based on minimization theory and low-rank matrix completion based on nuclear norm minimization for randomly missing structural vibration response data. Choi et al. used relational analysis to estimate missing strain values ​​using measured strain data during construction. These methods are prone to sample distortion and data distortion, reducing data robustness. Furthermore, interpolated data can easily be confused with identical values ​​in the original data, disrupting the distribution of the original data.

[0003] With the development of the field of machine learning, methods such as MF (Miss Forest), K-Nearest Neighbor (KNN), and Denoising Auto-encoder (DAE) have been proposed. These methods make the interpolated values ​​closer to the true values, but these methods also have some disadvantages. The DAE method requires complete data of the same type during training; KNN and MF are not easy to obtain highly correlated information for small sample data, and are easily limited by hardware resources when processing large samples.

[0004] Generative Adversarial Imputation Nets (GAIN) effectively solve the problems of the above methods. Its generative approach does not rely on complete data samples and has no requirements on the data relationships in the samples involved in training.

[0005] Although the GAIN method can simultaneously fill missing values ​​for multiple features, it also has some shortcomings. The GAIN method defaults to using a value of 0 as the initial value for missing data. However, this value of 0 can cause the model to misunderstand or bias the data when processing features with very small values. This can affect the authenticity of tunnel deformation monitoring data when the GAIN method is used to fill in tunnel deformation monitoring data with a value no greater than 1. At the same time, the GAIN method requires training both the generator and the discriminator, which makes the training of the method unstable. Moreover, when filling in data, the method may cause the generated dataset results to be biased towards a single mode in the real data distribution, ignoring other modes in the distribution, resulting in mode collapse in the generated samples. This leads to a decrease in the quality of the filled data when the GAIN method is used to fill in tunnel deformation monitoring data, which in turn affects the performance and generalization ability of the final model and is not conducive to subsequent data mining work.

[0006] Therefore, it is necessary to provide a tunnel deformation data filling method based on improved GAIN to solve the above problems.

[0007] Summary of the Invention

[0008] The purpose of the present invention is to provide a tunnel deformation data filling method based on improved GAIN, which uses an autoencoder to perform initial interpolation on the GAIN method to solve the problem of the GAIN method using 0 value as the initial interpolation. When the missing rate of tunnel deformation monitoring data is not higher than 50%, the tunnel deformation monitoring data filling method has good performance.

[0009] To achieve the above objectives, the present invention provides a tunnel deformation data filling method based on improved GAIN, and a method for performing initial interpolation of LSGAIN by autoencoders, namely Autoencoders-LSGAIN. Autoencoders consists of two parts: an encoder and a decoder. The encoder is a function that converts raw data into a code, and the encoder includes a fully connected neural network and a convolutional neural network. The decoder reversely restores the code back to the input to generate an output that is most similar to the original data.

[0010] Preferably, the training process of Autoencoders includes the following steps:

[0011] S1: Randomly initialize weights;

[0012] S2: Calculate the reconstruction error gradient through the back propagation method;

[0013] S3: Use optimization methods to iteratively minimize the error.

[0014] Preferably, the process of Autoencoders is as follows: the encoding of the original data x from the input layer to the hidden layer is shown in formula (1), and the decoding process from the hidden layer to the output layer is shown in formula (2); h = g1(A) (1)

[0015] Preferably, the Autoencoders method performs initial interpolation on GAIN, comprising the following steps:

[0016] S1A: The missing tunnel deformation monitoring data are first filled using Autoencoders;

[0017] S1B: Using Autoencoders to fill in missing values ​​in tunnel deformation monitoring data;

[0018] S1C: Fill the data filled by Autoencoders into GAIN.

[0019] Preferably, the cross entropy loss function in GAIN is replaced by the least squares loss function, and LSGAIN based on the least squares method is proposed; the loss function of GAIN is as follows,

[0020] The loss function of LSGAIN is shown in the following formula;

[0021] Among them, M is a mask vector, which represents a random variable with a value of {0,1}, H is a random variable that depends on the hint matrix, G is a generator used to generate data, and D is used to generate the probability of whether the data is true or false. Indicates expected value.

[0022] Therefore, the present invention adopts the above-mentioned tunnel deformation data filling method based on improved GAIN, which has the following beneficial effects:

[0023] (1) Compared with traditional methods, the present invention has certain advantages in filling missing tunnel deformation monitoring data, and it is found that the tunnel deformation monitoring data filled by this method will not affect the reliability and authenticity of the tunnel deformation monitoring data.

[0024] (2) The present invention reduces the computational time required for filling, so that the filling process can be completed in a shorter time, which can improve the overall efficiency of data processing and enable subsequent analysis and modeling work to be carried out more quickly.

[0025] (3) When the data missing rate is not higher than 50%, the interpolated tunnel deformation monitoring data of the present invention basically meets the changing trend of the tunnel deformation monitoring data. When the missing rate is higher than 50%, although the size of the interpolated tunnel deformation monitoring data is somewhat distorted relative to the original data, the displacement change missing under different working conditions has a certain similarity with the change of normal data.

[0026] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] FIG1 is a framework diagram of a tunnel deformation data filling method based on improved GAIN according to the present invention;

[0028] Fig. 2 is a framework diagram of the Autoencoders method of the present invention;

[0029] FIG3 is a framework diagram of the AE-GAIN method of the present invention;

[0030] FIG4 is a framework diagram of the LSGAIN method of the present invention. DETAILED DESCRIPTION

[0031] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0032] Unless otherwise defined, technical or scientific terms used in the present invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0033] The words “include” or “comprising” and similar words used in the present invention mean that the elements before the word include the elements listed after the word, and do not exclude the possibility of also including other elements. The orientation or position relationship indicated by the terms “inside”, “outside”, “upper”, “lower”, etc. is based on the orientation or position relationship shown in the accompanying drawings. It is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation of the present invention. When the absolute position of the described object changes, the relative position relationship may also change accordingly. In the present invention, unless otherwise clearly stipulated and limited, the terms such as “attachment” should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral whole; it can be directly connected or indirectly connected through an intermediate medium, and it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0034] As shown in Figure 1, the present invention provides a tunnel deformation data filling method based on improved GAIN, and a method for performing initial interpolation of LSGAIN by an autoencoder, Autoencoders-LSGAIN. Autoencoders consists of two parts: an encoder and a decoder. The encoder is a function that converts raw data into a code. The encoder includes a fully connected neural network and a convolutional neural network. The decoder reversely restores the code back to the input to generate an output that is most similar to the original data.

[0035] The training process of Autoencoders includes the following steps:

[0036] S1: Randomly initialize weights;

[0037] S2: Calculate the reconstruction error gradient through the back propagation method;

[0038] S3: Use optimization methods to iteratively minimize the error.

[0039] The process of Autoencoders is shown in Figure 2. The encoding of the original data x from the input layer to the hidden layer is shown in Equation (1), and the decoding process from the hidden layer to the output layer is shown in Equation (2); h = g1(A) (1)

[0040] As shown in Figure 3, the Autoencoders method performs initial interpolation of GAIN, including the following steps:

[0041] S1A: The missing tunnel deformation monitoring data are first filled using Autoencoders;

[0042] S1B: Using Autoencoders to fill in missing values ​​in tunnel deformation monitoring data;

[0043] S1C: Fill the data filled by Autoencoders into GAIN.

[0044] By replacing the cross-entropy loss function in GAIN with a least-squares loss function, we propose LSGAIN, a least-squares-based algorithm. This solves the vanishing gradient and mode collapse issues associated with the cross-entropy loss function in the original GAIN method. Furthermore, it reveals more information related to the generator's objective function, making the interpolation method more accurate. Figure 4 shows the framework of LSGAIN.

[0045] The least squares loss function is a convex function with good mathematical properties. By differentiating the loss function, the optimal analytical solution for the parameters can be obtained, which directly calculates the optimal parameter estimates, simplifying the model solution process. The least squares loss function is an optimization criterion that fits data by minimizing the difference between the observed values ​​and the model's predicted values. Under certain assumptions, the least squares solution is unbiased and efficient, and it can obtain the minimum variance estimate of the parameters, providing the best fit to the dataset. A global optimal solution is obtained: For the least squares loss function, there is a unique global optimal solution. When using the least squares method for parameter estimation, it ensures a globally optimal model fit, reducing the influence of local optimal solutions on the results. The least squares loss function is not only used for data fitting but also for statistical inference. Statistical analyses such as hypothesis testing and confidence interval estimation of model parameters can be used to assess the significance and reliability of the model.

[0046] The loss function of GAIN is as follows,

[0047] The loss function of LSGAIN is shown in the following formula;

[0048] Among them, M is a mask vector, which represents a random variable with a value of {0,1}, H is a random variable that depends on the hint matrix, G is a generator used to generate data, and D is used to generate the probability of whether the data is true or false. Indicates expected value.

[0049] Example

[0050] The process of Autoencoders is shown in Figure 2.

[0051] An autoencoder is an unsupervised learning method that automatically learns a feature representation from data and compresses the input data into a low-dimensional encoded form. Autoencoders consist of two parts: an encoder and a decoder. The encoder is a function that converts raw data into an encoded form. This is achieved through various methods, including fully connected neural networks and convolutional neural networks. The encoder aims to reduce the dimensionality of the data and extract the key features of the data.

[0052] The decoder's task is to translate the encoded data back into the input, generating an output that is most similar to the original data. Decoders are implemented using various methods, such as backpropagation neural networks. The decoder's goal is to restore the input data as accurately as possible using the information obtained from the encoder, while fully utilizing the information in the original data.

[0053] Therefore, the present invention adopts the above-mentioned tunnel deformation data filling method based on the improved GAIN. When the data missing rate is no more than 50%, the interpolated tunnel deformation monitoring data basically meets the changing trend of the tunnel deformation monitoring data. When the missing rate is higher than 50%, although the size of the interpolated tunnel deformation monitoring data is somewhat distorted relative to the original data, the displacement changes under different working conditions are somewhat similar to the changes in normal data. In summary, when the tunnel deformation monitoring data missing rate is no more than 50%, this paper has a good performance in filling tunnel deformation monitoring data.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A tunnel deformation data filling method based on improved GAIN, characterized by: Autoencoders-LSGAIN is a method for initial interpolation of LSGAIN by autoencoders. Autoencoders consists of two parts: encoder and decoder. The encoder is a function that converts raw data into code. The encoder includes a fully connected neural network and a convolutional neural network. The decoder reverses the encoding back to the input, producing an output that is most similar to the original data.

2. The tunnel deformation data filling method based on improved GAIN according to claim 1 is characterized by: The training process of Autoencoders includes the following steps: S1: Randomly initialize weights; S2: Calculate the reconstruction error gradient by back propagation method; S3: Use optimization methods to iteratively minimize the error.

3. The tunnel deformation data filling method based on improved GAIN according to claim 2 is characterized by: The process of Autoencoders is as follows: the encoding of the original data x from the input layer to the hidden layer is shown in equation (1), and the decoding process from the hidden layer to the output layer is shown in equation (2); h = g1(A) (1) 4. The tunnel deformation data filling method based on improved GAIN according to claim 3 is characterized by: The Autoencoders method performs initial interpolation on GAIN, including the following steps: S1A: The missing tunnel deformation monitoring data are first filled using Autoencoders; S1B: Using Autoencoders to fill in missing values ​​in tunnel deformation monitoring data; S1C: Fill the data filled by Autoencoders into GAIN.

5. The tunnel deformation data filling method based on improved GAIN according to claim 1, characterized in that: The cross entropy loss function in GAIN is replaced by the least squares loss function, and LSGAIN based on the least squares method is proposed; the loss function of GAIN is as follows: The loss function of LSGAIN is shown in the following formula; Among them, M is a mask vector, which represents a random variable with a value of {0,1}, H is a random variable that depends on the hint matrix, G is a generator used to generate data, and D is used to generate the probability of whether the data is true or false. Indicates expected value.

Images