Method and system for recovering lost data of multi-channel vibration signals in rail transit

Abstract

The present invention provides a method and system for recovering lost data of multi-channel vibration signals of rail transit, belonging to the technical field of rail transit signal data processing, including taking into account the characteristic trend of data changing over time, applying self-compiled mixed data to expand long and short-term neural networks, building a sequence-to-sequence data recovery model, and performing regression analysis based on multi-channel vibration continuous monitoring data collected on-site, using adjacent channel data of the lost channel to obtain the relationship between the lost channel data and the adjacent channel data, establishing a time series data prediction model under the influence of multiple channels, and repairing the lost channel data. On the basis of the traditional single-channel time series data prediction neural network, the present invention fully considers the multi-channel time series relationship, and on this basis uses mixed data to train the recovery model, and proposes a mixed data expansion long and short-term neural network recovery model. Compared with traditional data recovery methods, it has obvious improvements in efficiency and accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of rail transit signal data processing, and in particular to a method and system for recovering lost data of rail transit multi-channel vibration signals. Background Art

[0002] With the continuous development of urban rail transit construction, the lines in many cities have developed rapidly. With the increasing demand for transportation facility construction, subway evaluation, on-site testing, experimental verification and simulation simulation all require the collection of a large amount of measured data. In the process of data collection, due to the long-term vibration of the data acquisition equipment, the sensor is damaged, causing the collected data to be abnormal and lost, resulting in a large amount of data waste. This phenomenon is particularly obvious in the scenario where multiple sensors are used at one time to collect multi-channel data. When one of the sensors in a group of sensors is abnormal, the single-channel data collected is damaged, causing the entire group of multi-channel data to be discarded, which greatly reduces the amount of data available per unit time.

[0003] In order to solve the problem of data loss, a data recovery model is usually constructed based on historical data. At present, the research methods of data recovery can be roughly divided into three categories: the first category is the traditional mathematical model, the second is the intelligent recovery method, and the third is the combination model, which refers to the combination of two or more data processing methods. Although the traditional method can achieve accurate recovery of a single missing data set, the important information in the long time series data is often ignored in addition to the long estimation time, which will lead to data loss and distortion, and have a greater impact on subsequent data analysis. Intelligent recovery methods, especially the use of neural networks, can identify the characteristics, trends and development laws of variable changes from time series, thereby achieving effective prediction and recovery of missing data. Intelligent recovery methods mainly use machine learning algorithms to identify and recover abnormal and missing data, and are widely used in data recovery. At present, neural networks are mainly used for the prediction of single-channel time series data based on historical data. However, there is a lack of research on methods for recovering data from one channel using multi-channel data when collecting multi-channel data, especially when there is a certain data correlation between channels. Summary of the invention

[0004] The object of the present invention is to provide a method and system for recovering lost data of multi-channel vibration signals of rail transit, so as to solve at least one technical problem existing in the above-mentioned background technology.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In a first aspect, the present invention provides a method for recovering lost data of a multi-channel vibration signal of rail transit, comprising:

[0007] Input the channel data to be recovered into the trained data recovery model, specify the channel with complete data as the prediction sequence, and specify the channel that needs data recovery as the response sequence;

[0008] The input channel data to be restored is processed using a data recovery model, and the adjacent channels of any channel of missing data of interest are used to restore the data of the missing channel; wherein, the data recovery model training includes: preprocessing the vibration data collected on site and dividing it into a training set and a test set; dividing the data in the training set into two groups, one of which is subjected to Fourier transform; establishing a sequence-to-sequence RNN encoder-decoder hybrid data expansion long-short-time neural network recovery model, and in each group of multi-channel data, defining the channels as response sequences and feature sequences respectively; allocating the multi-channel data in the training set into two groups of sequences; creating a sequence input layer, a long-short-time neural network layer, a fully connected layer, a discard layer and a regression output layer according to the total amount of data collected and the number of channel inputs; using a user-defined solver, specifying the number of iterations, the number of samples for each training and the learning rate value; using the training set to train the model, and using the validation set data to verify the accuracy of the model, to obtain the data recovery model.

[0009] Optionally, the vibration data collected on-site is preprocessed and divided into a training set and a test set, including: designating the data group collected by each individual sensor as a channel data, and the data collected by multiple sensors at the same time as a group of multi-channel data; dividing each group of collected multi-channel data into two collections, selecting multi-channel data with complete data for training the model, and selecting multi-channel data with missing data in a certain channel for model recovery; according to the amount of data, allocating part of the complete multi-channel data as the training set for training the model, and the remaining data is used as the verification set for testing the model.

[0010] Optionally, define the channels with lost data that are expected to be recovered as the response sequence, and the remaining channels with complete data as the feature sequence.

[0011] Optionally, the value of the input layer is equal to the number of complete data channels, the value of the fully connected layer is equal to the number of missing data channels, and the value of the long short-term neural network layer is equal to 10 times the number of complete data channels and is not less than 50.

[0012] Optionally, the number of iterations entered is equal to 70, the sample size for each training is one percent of the sample size collected for a single channel and is not greater than 200, and the learning rate value is 0.03.

[0013] Optionally, use the data in training set one to train the model, perform inverse Fourier transform on the training results and mix them with the data in training set two; use the mixed data to train the model again, and finally use the validation set data to validate the twice-trained model.

[0014] In a second aspect, the present invention provides a system for recovering lost data of multi-channel vibration signals of rail transit, comprising:

[0015] The input module is used to input the channel data that needs to be restored into the trained data recovery model, specify the channel with complete data as the prediction sequence, and specify the channel that needs data recovery as the response sequence;

[0016] A processing module is used to use a data recovery model to process the input channel data that needs to be restored, and use the adjacent channels of any interested missing data channel to restore the data of the missing channel; wherein the data recovery model training includes: preprocessing the vibration data collected on site and dividing it into a training set and a test set; dividing the data in the training set into two groups, one of which is subjected to Fourier transform; establishing a sequence-to-sequence RNN encoder-decoder hybrid data expansion long-short-time neural network recovery model, and in each group of multi-channel data, defining the channels as response sequences and feature sequences respectively; allocating the multi-channel data in the training set into two groups of sequences; creating a sequence input layer, a long-short-time neural network layer, a fully connected layer, a discard layer and a regression output layer according to the total amount of data collected and the number of channel inputs; using a user-defined solver, specifying the number of iterations, the number of samples for each training and the learning rate value; using the training set to train the model, and using the validation set data to verify the model accuracy, to obtain the data recovery model.

[0017] In a third aspect, the present invention provides a non-transitory computer-readable storage medium, which is used to store computer instructions. When the computer instructions are executed by a processor, the method for recovering lost data of multi-channel vibration signals of rail transit as described in the first aspect is implemented.

[0018] In a fourth aspect, the present invention provides a computer device comprising a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the method for recovering lost data of multi-channel vibration signals of rail transit as described in the first aspect.

[0019] In a fifth aspect, the present invention provides an electronic device comprising: a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes instructions for implementing the method for recovering lost data of multi-channel vibration signals of rail transit as described in the first aspect.

[0020] Beneficial effects of the present invention: The data recovery model trained with mixed data has an accuracy of 95% for recovering multi-channel time series data, which can better reflect the internal relationship between adjacent channels, thereby making the lost data recovery model have higher accuracy.

[0021] Additional advantages of the present invention will be more clearly given in the following description or learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 The present invention is a flowchart of a method for recovering multi-channel measured lost data according to an embodiment of the present invention.

[0024] Figure 2 This is a diagram showing the arrangement of vibration acceleration sensor measurement points in a subway tunnel according to an embodiment of the present invention.

[0025] Figure 3 This is a comparison diagram of the pre-trained data and the predicted data of channel 8 after Fourier transformation of the training set 1 according to an embodiment of the present invention.

[0026] Figure 4 This is a comparison chart between the predicted data and the measured data of channel 8 after the inverse Fourier transform described in an embodiment of the present invention.

[0027] Figure 5 This is a comparison chart of the predicted data and measured data of channel 8 after secondary training using mixed data according to an embodiment of the present invention.

[0028] Figure 6 This is a comparison chart of the predicted data and measured data of channel 8 of the validation model using the validation set as described in an embodiment of the present invention.

[0029] Figure 7 This is a comparison chart of the predicted data and the measured data of the validation set channel 8 after Fourier transformation according to an embodiment of the present invention. DETAILED DESCRIPTION

[0030] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below by the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0031] It should be understood by those skilled in the art that unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which this invention belongs.

[0032] It should also be understood that terms, such as those defined in commonly used dictionaries, should be understood to have a meaning consistent with that in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless as defined herein.

[0033] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements and / or groups thereof.

[0034] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0035] To facilitate understanding of the present invention, the present invention is further explained below with reference to specific embodiments in conjunction with the accompanying drawings, and the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0036] Those skilled in the art should understand that the drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily necessary for implementing the present invention.

[0037] The present invention aims to solve the problem that in actual testing, when multiple sensors are used at a time to collect multi-channel data, when a single sensor fails or is damaged, single channel data in a set of multi-channel data may be abnormal or lost. A method for recovering lost data of multi-channel vibration signals measured in rail transit is found to overcome the above problem.

[0038] Example 1

[0039] In this embodiment 1, a system for recovering lost data of multi-channel vibration signals of rail transit is first provided, including: an input module, used to input channel data to be recovered into a trained data recovery model, designate channels with complete data as prediction sequences, and designate channels for data recovery as response sequences; a processing module, used to use the data recovery model to process the input channel data to be recovered, and use adjacent channels of any channel of interest with missing data to recover the data of the missing channel; wherein the data recovery model training includes: preprocessing the vibration data collected on site, dividing it into a training set and a test set; dividing the data in the training set into The invention discloses a method for recovering a data set of a RNN multi-channel data set, wherein the data set is divided into two groups, one of which is subjected to Fourier transform; a sequence-to-sequence RNN encoder-decoder hybrid data augmentation long-short-time neural network recovery model is established, and in each group of multi-channel data, the channels are defined as response sequences and feature sequences respectively; the multi-channel data in the training set are distributed into two groups of sequences; a sequence input layer, a long-short-time neural network layer, a fully connected layer, a discard layer and a regression output layer are created according to the total amount of data collection and the number of channel inputs; a user-defined solver is used to specify the number of iterations, the amount of samples for each training and the learning rate value; the model is trained using the training set, and the accuracy of the model is verified using the validation set data, so as to obtain the data recovery model.

[0040] In this embodiment 1, the above-mentioned system is used to implement a method for recovering lost data of multi-channel vibration signals of rail transit, including: inputting the channel data to be recovered into a trained data recovery model, designating the channels with complete data as prediction sequences, and designating the channels for which data recovery is required as response sequences; using the data recovery model to process the input channel data to be recovered, and using the adjacent channels of any channel of interest with missing data to recover the data of the missing channel; wherein the data recovery model training includes: preprocessing the vibration data collected on-site, dividing it into a training set and a test set; dividing the data in the training set into two groups, A group of them is subjected to Fourier transform; a sequence-to-sequence RNN encoder-decoder hybrid data expansion long-short-time neural network recovery model is established, and in each group of multi-channel data, the channels are defined as response sequences and feature sequences respectively; the multi-channel data in the training set are distributed to two groups of sequences; a sequence input layer, a long-short-time neural network layer, a fully connected layer, a discard layer and a regression output layer are created according to the total amount of data collection and the number of channel inputs; a user-defined solver is used to specify the number of iterations, the number of samples for each training and the learning rate value; the model is trained using the training set, and the accuracy of the model is verified using the validation set data to obtain the data recovery model.

[0041] The vibration data collected on-site are preprocessed and divided into a training set and a test set, including: designating the data group collected by each individual sensor as a channel data, and the data collected by multiple sensors at the same time as a group of multi-channel data; dividing each group of multi-channel data collected into two collections, selecting multi-channel data with complete data for training the model, and selecting multi-channel data with missing data in a certain channel for model recovery; according to the amount of data, allocating part of the complete multi-channel data as the training set for training the model, and the remaining data is used as the verification set for testing the model.

[0042] The channels with lost data and expected to be recovered are defined as response sequences, and the remaining channels with complete data are defined as feature sequences.

[0043] The value of the input layer is equal to the number of complete data channels, the value of the fully connected layer is equal to the number of lost data channels, and the value of the long and short time neural network layer is equal to 10 times the number of complete data channels and is not less than 50.

[0044] The number of input iterations is equal to 70, the sample size for each training is one percent of the sample size collected for a single channel and is no greater than 200, and the learning rate value is 0.03.

[0045] Use the data in training set one to train the model, perform inverse Fourier transform on the training results and mix them with the data in training set two; use the mixed data to train the model again, and finally use the validation set data to verify the twice-trained model.

[0046] Example 2

[0047] In this embodiment 2, in order to address the problem that when multiple sensors are used at a time to collect multi-channel data in actual testing, a single sensor fails or is damaged, resulting in abnormal or lost data of a single channel in a set of multi-channel data, a method for recovering lost data of multi-channel vibration signals measured in rail transit is provided to overcome the above problem.

[0048] The method for recovering lost data of multi-channel vibration signals measured in rail transit described in this embodiment 2 can be used to recover lost data of the target channel using the measured data of adjacent channels. The idea is clear and the process is clear, which reflects the feasibility of the recovery method.

[0049] The invention discloses a method for recovering lost data of multi-channel time series measured in urban rail transit, comprising: preprocessing data collected on site, establishing a data recovery model, training and debugging a neural network, and using the model to recover lost channel data.

[0050] S1. Preprocessing of vibration data collected on site, this step includes:

[0051] S11. Designate the data group collected by each individual sensor as a channel data, and the data collected by multiple sensors at the same time as a group of multi-channel data.

[0052] S12. Divide each group of collected multi-channel data into two collections, select the multi-channel data with complete data for training the model, and select the multi-channel data with missing data in a certain channel for use in model recovery.

[0053] S13. According to the amount of data, part of the complete multi-channel data is allocated as a training set for training the model, and the remaining data is used as a validation set for testing the model.

[0054] S14. Divide the data in the training set into two groups, and perform Fourier transform on one of them.

[0055] Wherein, in step S13, 80% of the complete multi-channel data is allocated as a training set for training the model, and 20% of the data is used as a validation set for testing the model.

[0056] In step S14, 30% of the training set data is allocated for Fourier transform, which is called training set 1, and 70% of the training set data is not processed, which is called training set 2.

[0057] S2. Establishing a data recovery model, this step includes:

[0058] S21. In each set of multi-channel data, define the channels as response sequences and feature sequences respectively. Allocate the multi-channel data in the training set into two sets of sequences.

[0059] S22. Create a sequence input layer, a long-short time neural network layer, a fully connected layer, a discard layer, and a regression output layer based on the total amount of data collected and the number of channel inputs.

[0060] S23. Use a user-defined solver, specifying the number of iterations, the size of each training sample, and the learning rate value.

[0061] S24. Use the training data to train the model and use the validation set data to verify the model accuracy.

[0062] In step S21, the channels with lost data and expected to be restored are defined as response sequences, and the remaining channels with complete data are defined as feature sequences.

[0063] In step S22, the value of the input layer is equal to the number of complete data channels, the value of the LSTM fully connected layer is equal to the number of lost data channels, and the long short-term neural network layer is equal to 10 times the number of complete data channels and is not less than 50.

[0064] The number of iterations input in step S23 is equal to 70, the sample size for each training is one percent of the sample size collected by a single channel and is not greater than 200, and the value of the learning rate is 0.03.

[0065] In step S24, the model is first trained using the data in training set 1, and the training results are mixed with the data in training set 2 after inverse Fourier transformation. The model is then trained again using the mixed data, and finally the model trained again is verified using the verification set data.

[0066] S3. Use the model to restore the lost channel data. This step includes:

[0067] S31. Input the channel data that needs to be restored into the trained model, designate the channel with complete data as the prediction sequence, and designate the channel that needs data restoration as the response sequence.

[0068] S32. Use the model to restore the lost channel data.

[0069] Example 3

[0070] like Figure 1 As shown, in this embodiment 3, a method for recovering lost data of multi-channel time series measured in urban rail transit is provided, which sequentially includes preprocessing of on-site collected data, establishing a data recovery model, training and debugging of a neural network, and using the model to recover lost channel data.

[0071] S1. Preprocessing of vibration data collected on site, this step includes:

[0072] S11. Designate the data group collected by each individual sensor as a channel data, and the data collected by multiple sensors at the same time as a group of multi-channel data.

[0073] S12. Divide each group of collected multi-channel data into two collections, select the multi-channel data with complete data for training the model, and select the multi-channel data with missing data in a certain channel for use in model recovery.

[0074] S13. According to the amount of data, part of the complete multi-channel data is allocated as a training set for training the model, and the remaining data is used as a validation set for testing the model.

[0075] S14. Divide the data in the training set into two groups, and perform Fourier transform on one of them.

[0076] S2. Establishing a data recovery model, this step includes:

[0077] S21. In each set of multi-channel data, define the channels as response sequences and feature sequences respectively. Allocate the multi-channel data in the training set into two sets of sequences.

[0078] S22. Create a sequence input layer, a long-short time neural network layer, a fully connected layer, a discard layer, and a regression output layer based on the total amount of data collected and the number of channel inputs.

[0079] S23. Use a user-defined solver, specifying the number of iterations, the size of each training sample, and the learning rate value.

[0080] S24. Use the training data to train the model and use the validation set data to verify the model accuracy.

[0081] S3. Use the model to restore the lost channel data. This step includes:

[0082] S31. Input the channel data that needs to be restored into the trained model, designate the channel with complete data as the prediction sequence, and designate the channel that needs data restoration as the response sequence.

[0083] S32. Use the model to restore the lost channel data.

[0084] 1. Preprocessing of field data

[0085] like Figure 2 As shown, the multi-channel measured data is collected using the INV3062V network distributed data collector and the LC0161BG piezoelectric acceleration sensor. The tunnel cross-section measurement points are arranged on the side away from the platform, including 8 measurement points of the rail, the roadbed, and the tunnel wall. In order to ensure the effectiveness of the network model, 100 groups of 8-channel data and 12,800 sampling points of each channel are selected as the model training data. A group of 8-channel data, in which 8 channels of data are lost, is designated as the model recovery data.

[0086] All data were normalized by subtracting the mean and dividing by the standard deviation to reduce the features. After processing, all values ​​had a mean of 0 and a unit square. 80 sets of data were designated as training sets and 20 sets of data were designated as validation sets.

[0087] 2. Establish a data recovery model

[0088] The subway vibration signal prediction model was established using MATLAB, and channels 1-7 were defined as feature sequences and channel 8 as response sequence.

[0089] Create a long short-term neural network, set the network to contain an LSTM layer with 80 hidden units, a fully connected layer of size 1, a dropout layer with a dropout probability of 0.3, and a regression output layer.

[0090] Specify the training options. Use the solver 'adam', set the number of epochs to 70, the number of samples per training to 128, and the learning rate to 0.03. To prevent gradient explosion, set the gradient threshold to 1.

[0091] Select 24 sets of training data and designate them as training set 1, and the remaining 56 sets of data are designated as training set 2. Perform Fourier transform on training set 1, input the transformed data into the defined data recovery neural network, perform model training, and obtain the training results as follows Figure 3 shown.

[0092] like Figure 4 As shown, the training results are inverse Fourier transformed and the transformed data are remixed with the data in the second training set.

[0093] The mixed data is input into the data recovery neural network for secondary training, and the training results are as follows: Figure 5 shown.

[0094] like Figure 6 As shown, the trained data recovery model is verified using the validation set. The vibration signal predicted by the adjacent channel time series loss data recovery model proposed in the present invention is basically consistent with the real vibration signal curve. The data of the predicted sequence channel 8 in the validation set is compared with the data of the measured channel 8, and the root mean square error is 0.3457.

[0095] like Figure 7 As shown, the vibration acceleration signal is processed by one-third octave band, so as to obtain a more intuitive comparison chart of the lost data and the measured vibration acceleration amplitude of channel 8 after recovery. The lost data recovery value and the true value have a high degree of consistency at all frequencies, and the error values ​​at all frequencies are less than 10%, and the overall accuracy rate is as high as 95%.

[0096] 3. Use the model to restore lost channel data

[0097] Input a set of channel data that needs to be restored into the trained model, specify channels 1-7 with complete data as the prediction sequence, specify the channels that need data restoration as the response sequence, and use the trained model to restore the lost channel data.

[0098] In summary, the method for recovering lost data of multi-channel vibration signals measured in rail transit described in Example 3. When collecting multi-channel data, adjacent channel data is used to recover lost time series data. The present invention uses mixed data in the time domain and frequency domain to perform secondary training on the data recovery model. Its advantage is that the data recovery model trained with mixed data has an accuracy of 95% for recovering multi-channel time series data, which can better reflect the internal relationship between adjacent channels, thereby making the lost data recovery model have higher accuracy. This data recovery method is feasible in actual production and is widely used in rail transit data collection and processing, so it has high theoretical value and commercial promotion prospects.

[0099] Example 4

[0100] This embodiment 4 provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method for recovering lost data of a multi-channel vibration signal of rail transit as described above is implemented. The method includes:

[0101] Input the channel data to be recovered into the trained data recovery model, specify the channel with complete data as the prediction sequence, and specify the channel that needs data recovery as the response sequence;

[0102] The input channel data to be restored is processed using a data recovery model, and the adjacent channels of any channel of missing data of interest are used to restore the data of the missing channel; wherein, the data recovery model training includes: preprocessing the vibration data collected on site and dividing it into a training set and a test set; dividing the data in the training set into two groups, one of which is subjected to Fourier transform; establishing a sequence-to-sequence RNN encoder-decoder hybrid data expansion long-short-time neural network recovery model, and in each group of multi-channel data, defining the channels as response sequences and feature sequences respectively; allocating the multi-channel data in the training set into two groups of sequences; creating a sequence input layer, a long-short-time neural network layer, a fully connected layer, a discard layer and a regression output layer according to the total amount of data collected and the number of channel inputs; using a user-defined solver, specifying the number of iterations, the number of samples for each training and the learning rate value; using the training set to train the model, and using the validation set data to verify the accuracy of the model, to obtain the data recovery model.

[0103] Example 5

[0104] This embodiment 5 provides a computer device, including a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, the processor calls the program instructions to execute the above-mentioned method for recovering lost data of multi-channel vibration signals of rail transit, the method comprising:

[0105] Input the channel data to be recovered into the trained data recovery model, specify the channel with complete data as the prediction sequence, and specify the channel that needs data recovery as the response sequence;

[0106] The input channel data to be restored is processed using a data recovery model, and the adjacent channels of any channel of missing data of interest are used to restore the data of the missing channel; wherein, the data recovery model training includes: preprocessing the vibration data collected on site and dividing it into a training set and a test set; dividing the data in the training set into two groups, one of which is subjected to Fourier transform; establishing a sequence-to-sequence RNN encoder-decoder hybrid data expansion long-short-time neural network recovery model, and in each group of multi-channel data, defining the channels as response sequences and feature sequences respectively; allocating the multi-channel data in the training set into two groups of sequences; creating a sequence input layer, a long-short-time neural network layer, a fully connected layer, a discard layer and a regression output layer according to the total amount of data collected and the number of channel inputs; using a user-defined solver, specifying the number of iterations, the number of samples for each training and the learning rate value; using the training set to train the model, and using the validation set data to verify the accuracy of the model, to obtain the data recovery model.

[0107] Example 6

[0108] This embodiment 6 provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes instructions for implementing the method for recovering lost data of a multi-channel vibration signal of rail transit as described above, the method comprising:

[0109] Input the channel data to be recovered into the trained data recovery model, specify the channel with complete data as the prediction sequence, and specify the channel that needs data recovery as the response sequence;

[0110] The input channel data to be restored is processed using a data recovery model, and the adjacent channels of any channel of missing data of interest are used to restore the data of the missing channel; wherein, the data recovery model training includes: preprocessing the vibration data collected on site and dividing it into a training set and a test set; dividing the data in the training set into two groups, one of which is subjected to Fourier transform; establishing a sequence-to-sequence RNN encoder-decoder hybrid data expansion long-short-time neural network recovery model, and in each group of multi-channel data, defining the channels as response sequences and feature sequences respectively; allocating the multi-channel data in the training set into two groups of sequences; creating a sequence input layer, a long-short-time neural network layer, a fully connected layer, a discard layer and a regression output layer according to the total amount of data collected and the number of channel inputs; using a user-defined solver, specifying the number of iterations, the number of samples for each training and the learning rate value; using the training set to train the model, and using the validation set data to verify the accuracy of the model, to obtain the data recovery model.

[0111] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0112] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0113] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0114] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0115] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative labor on the basis of the technical solution disclosed in the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for recovering lost data of multi-channel vibration signals of rail transit, characterized in that: include: Input the channel data to be recovered into the trained data recovery model, specify the channel with complete data as the prediction sequence, and specify the channel that needs data recovery as the response sequence; The input channel data to be restored is processed using a data recovery model, and the adjacent channels of any interested missing data channel are used to restore the missing data channel data; wherein, the data recovery model training includes: preprocessing the vibration data collected on site, dividing it into a training set and a test set; dividing the data in the training set into two groups, one of which is subjected to Fourier transform; establishing a sequence-to-sequence RNN encoder-decoder hybrid data expansion long-short-time neural network recovery model, and in each group of multi-channel data, defining the channels as response sequences and feature sequences respectively; allocating the multi-channel data in the training set into two groups of sequences; creating a sequence input layer, a long-short-time neural network layer, a fully connected layer, a discard layer, and a regression output layer according to the total amount of data collected and the number of channel inputs; using a user-defined solver, specifying the number of iterations, the number of samples for each training, and the learning rate value; using the training set to train the model, and using the validation set data to verify the accuracy of the model, to obtain the data recovery model; The vibration data collected on-site are preprocessed and divided into a training set and a test set, including: designating the data group collected by each individual sensor as a channel data, and the data collected by multiple sensors at the same time as a group of multi-channel data; dividing each group of multi-channel data collected into two collections, selecting multi-channel data with complete data for training the model, and selecting multi-channel data with missing data in a certain channel for model recovery; according to the amount of data, allocating part of the complete multi-channel data as the training set for training the model, and the remaining data is used as the verification set for testing the model.

2. The method for recovering lost data of multi-channel vibration signals of rail transit according to claim 1, characterized in that: The channels with lost data and expected to be recovered are defined as response sequences, and the remaining channels with complete data are defined as feature sequences.

3. The method for recovering lost data of multi-channel vibration signals of rail transit according to claim 1, characterized in that: The value of the input layer is equal to the number of complete data channels, the value of the fully connected layer is equal to the number of lost data channels, and the value of the long and short time neural network layer is equal to 10 times the number of complete data channels and is not less than 50.

4. The method for recovering lost data of multi-channel vibration signals of rail transit according to claim 1, characterized in that: The number of input iterations is equal to 70, the sample size for each training is one percent of the sample size collected for a single channel and is no greater than 200, and the learning rate value is 0.

03.

5. The method for recovering lost data of multi-channel vibration signals of rail transit according to claim 1, characterized in that: Use the data in training set one to train the model, perform inverse Fourier transform on the training results and mix them with the data in training set two; use the mixed data to train the model again, and finally use the validation set data to verify the twice-trained model.

6. A system for recovering lost data of multi-channel vibration signals of rail transit, characterized in that: include: The input module is used to input the channel data that needs to be restored into the trained data recovery model, specify the channel with complete data as the prediction sequence, and specify the channel that needs data recovery as the response sequence; A processing module is used to use a data recovery model to process the input channel data that needs to be restored, and use the adjacent channels of any interested missing data channel to restore the missing data channel data; wherein, the data recovery model training includes: preprocessing the vibration data collected on site, dividing it into a training set and a test set; dividing the data in the training set into two groups, one of which is subjected to Fourier transform; establishing a sequence-to-sequence RNN encoder-decoder hybrid data expansion long-short-time neural network recovery model, and in each group of multi-channel data, defining the channel as a response sequence and a feature sequence respectively; allocating the multi-channel data in the training set into two groups of sequences; creating a sequence input layer, a long-short-time neural network layer, a fully connected layer, a discard layer, and a regression output layer according to the total amount of data collected and the number of channel inputs; using a user-defined solver, specifying the number of iterations, the number of samples for each training, and the learning rate value; using the training set to train the model, and using the validation set data to verify the model accuracy, to obtain the data recovery model; The vibration data collected on-site are preprocessed and divided into a training set and a test set, including: designating the data group collected by each individual sensor as a channel data, and the data collected by multiple sensors at the same time as a group of multi-channel data; dividing each group of multi-channel data collected into two collections, selecting multi-channel data with complete data for training the model, and selecting multi-channel data with missing data in a certain channel for model recovery; according to the amount of data, allocating part of the complete multi-channel data as the training set for training the model, and the remaining data is used as the verification set for testing the model.

7. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method for recovering lost data of multi-channel vibration signals of rail transit as described in any one of claims 1 to 5 is implemented.

8. A computer device, characterized in that: It includes a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the method for recovering lost data of multi-channel vibration signals of rail transit as described in any one of claims 1-5.

9. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes instructions for implementing the method for recovering lost data of multi-channel vibration signals of rail transit as described in any one of claims 1-5.

Images