Train wheel slip detection method, device, equipment and medium

Abstract

The invention relates to a train wheel slip detection method, device and equipment and a medium, the method is realized based on deep learning, and the method comprises the following steps: S1, collecting a vehicle-mounted driving log, extracting slip data, and sorting the slip-related data into a data set; s2, constructing a train wheel slip detection model based on the combination of the long and short term memory network and an auto-encoder; s3, making a data set, performing normalization processing on the data, and dividing the data into a training set, a verification set and a test set; s4, using the training set to train the constructed model until the loss value calculated by the loss function converges, and obtaining a trained model; and S5, reconstructing to-be-detected data by using the trained model to obtain output data, and comparing the output data with the original data to judge a slip section. Compared with the prior art, the method has the advantages of effectively saving the cost and the like.

Description

Technical Field

[0001] The present invention relates to a rail transit signal system, and in particular to a train wheel slip detection method, device, equipment and medium based on deep learning. Background Art

[0002] The safe operation of trains depends on accurate speed measurement and positioning, which is especially important in the automatic train protection system, as it directly affects driving decisions. The results of the axle speed sensors currently widely used are easily affected by wheel slippage, resulting in inaccurate speed measurement and positioning deviation. If the error is large, it may cause serious accidents. Therefore, accurately judging the wheel slippage state is extremely critical to improving positioning accuracy and ensuring driving safety.

[0003] After searching, Chinese patent publication number CN 107525505A discloses a method for detecting idling and sliding of locomotive wheels. Through inertial navigation and speed sensor data, the speed difference, acceleration difference and sliding rate are calculated to determine that the running state of the locomotive wheels is the sliding state, so as to detect whether the locomotive wheels are idling or sliding.

[0004] Chinese patent publication number CN108216168A discloses a method for detecting train slippage or idling, which obtains speed data through axle sensors and Doppler radars, calculates the speed difference ratio, and determines slippage or idling based on the difference ratio and train status, thereby improving the accuracy of detection and speed compensation.

[0005] Chinese patent publication number CN102991489 A discloses a train speed and distance measurement system for detecting and compensating for idling and slipping, including an acceleration sensor, a speed sensor and multiple determination modules, which can accurately detect the wheel status and compensate for the speed.

[0006] Chinese patent publication number CN105242060 A discloses a real-time monitoring method for train wheel idling and sliding. By installing a camera and a plane mirror, the wheel and train speeds are measured, the wheel running status is determined, and the monitoring efficiency is improved.

[0007] In summary, the existing skid identification technology relies on sensor data comparison and threshold judgment, and its core lies in the appropriate threshold setting; if the threshold is too high, slight skid may be ignored and the reaction to significant skid may be slow, affecting safety; if the threshold is too low, it may easily lead to false alarms and damage operating efficiency; a unified threshold may not be suitable for all lines and may lead to misjudgment; in addition, the introduction of multiple sensors increases costs and requires additional consideration of the failure of these new sensors.

[0008] Therefore, how to improve the accuracy of wheel slip detection without significantly increasing costs has become a technical problem that needs to be solved urgently. Summary of the invention

[0009] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a train wheel slip detection method, device, equipment and medium based on deep learning.

[0010] The purpose of the present invention can be achieved by the following technical solutions:

[0011] According to a first aspect of the present invention, a train wheel slip detection method is provided, which is implemented based on deep learning and comprises the following steps:

[0012] Step S1, collecting the vehicle driving log and extracting the data of the skidding, and arranging the data related to the skidding into a data set;

[0013] Step S2, constructing a train wheel slip detection model based on a combination of a long short-term memory network and an autoencoder;

[0014] Step S3, creating a data set, normalizing the data, and dividing the data into a training set, a validation set, and a test set;

[0015] Step S4, using the training set to train the model constructed in step S2 until the loss value calculated by the loss function converges, thereby obtaining a trained model;

[0016] Step S5, using the model trained in step S4 to reconstruct the data to be detected, obtain output data, and compare the output data with the original data to determine the slip section.

[0017] As a preferred technical solution, step S1 specifically includes the following steps:

[0018] Step S1-1, extracting data related to wheel slip measured by sensors from the onboard driving log of an existing train;

[0019] Step S1-2, cleaning the data to remove obviously abnormal data;

[0020] Step S1-3, use the sliding window method to sample the cleaned data, and save the normal data and the slip data separately.

[0021] As a preferred technical solution, the data measured by the sensor includes velocity, displacement and acceleration data.

[0022] As a preferred technical solution, the sliding window method specifically includes: sampling the used data according to the set window size and time step.

[0023] As a preferred technical solution, the autoencoder of the model in step S2 includes an encoder and a decoder both of which are built by long short-term memory networks, wherein the encoder is composed of two layers of long short-term memory networks, and the decoder is composed of two layers of long short-term memory networks and one fully connected layer.

[0024] As a preferred technical solution, the long short-term memory network of the model in step S2 is used to extract hidden features of the time series, and the autoencoder is used to identify the difference between abnormal data and normal data.

[0025] As a preferred technical solution, step S3 specifically includes the following steps:

[0026] Step S3-1, normalize all data involved in training and testing;

[0027] In step S3-2, 70% of the normal data are used as a training set, 15% are used as a validation set, and the remaining 15% and all the slip data form a test set.

[0028] As a preferred technical solution, the normalization process in step S3-1 is specifically as follows:

[0029]

[0030] Among them, x is the data to be processed, x min is the minimum value among all data, x max is the maximum value of all data, and x' is the data after normalization.

[0031] As a preferred technical solution, step S4 specifically includes the following steps:

[0032] Step S4-1, reducing the dimension of the training set data through the encoder so that the model can learn and capture the characteristics of normal data;

[0033] Step S4-2, reconstructing the data after the encoder dimensionality reduction through the decoder and restoring it to the original dimension;

[0034] Step S4-3, calculating the reconstruction error using the loss function;

[0035] Step S4-4, after each training round, the trained model is reconstructed and predicted using the validation set, and the loss value of the validation set is calculated using the loss function;

[0036] Step S4-5, using gradient descent, reduces the loss values ​​of the training set and the validation set in multiple training rounds until the loss values ​​converge.

[0037] As a preferred technical solution, the encoder body propagation process in step S4-1 includes:

[0038] Step S4-1-1, the first LSTM network layer receives input of n_features dimension and outputs hidden state of hidden_dim dimension;

[0039] Step S4-1-2, the second long short-term memory network layer receives the output of the first layer and generates an embedding of the embedding_dim dimension.

[0040] As a preferred technical solution, the specific propagation process of the decoder in step S4-2 includes:

[0041] Step S4-2-1, the first LSTM network layer receives the input of the embedding_dim dimension transmitted by the encoder and outputs the hidden state of the same dimension;

[0042] Step S4-2-2, the second long short-term memory network layer receives the output of the first layer and generates a hidden state of the hidden_dim dimension;

[0043] Step S4-2-3, the next layer is a fully connected layer that maps the hidden state to n_features dimensions.

[0044] As a preferred technical solution, the loss function in step S4-3 is a mean square error loss function, which is used to calculate the average value of the sum of squares of the differences between the predicted value and the true value. The specific formula is as follows:

[0045]

[0046] Among them, y i is the value after normalization of the training data, The data y is used for the model i The reconstructed value after reconstruction.

[0047] As a preferred technical solution, step S5 specifically includes the following steps:

[0048] Step S5-1, input the data to be detected into the trained model to obtain reconstructed data;

[0049] Step S5-2, after the normal data is reconstructed by the model, the reconstruction error between the normal data and the original data is less than the set threshold, but when the slipped data is input into the model for reconstruction, the difference between the reconstructed data and the original data is greater than the set threshold;

[0050] Step S5-3, comparing and analyzing the reconstructed data with the original data, so as to identify the section where the slip occurs.

[0051] According to a second aspect of the present invention, there is provided a device for the train wheel slip detection method, comprising a data extraction module, a data storage module, a model storage module and a slip detection module;

[0052] The data extraction module is connected to the vehicle driving log storage system, and is used to extract the data related to the skidding in the driving log, and transmit the extracted data to the data storage module;

[0053] The data storage module is used to store the data to be detected and the data that has been detected and the detection results;

[0054] The model storage module stores a trained train wheel slip detection model based on a deep learning network;

[0055] The slip detection module stores relevant judgment conditions for slip detection, calls the model stored in the model storage module, compares the data reconstructed by the model with the original data, detects the section where slip occurs, and transmits the processed data and results to the data storage module.

[0056] According to a third aspect of the present invention, there is provided an electronic device, comprising a memory and a processor, wherein a computer program is stored in the memory, and the method described above is implemented when the processor executes the program.

[0057] According to a fourth aspect of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, wherein the program implements the method described when executed by a processor.

[0058] Compared with the prior art, the present invention has the following advantages:

[0059] 1) The present invention relies only on wheel speed sensors and improves the detection accuracy of slippage through deep learning techniques, thereby improving the operating efficiency of the train while ensuring the safe operation of the train;

[0060] 2) The present invention combines the real data collected by sensors in the train controller in actual projects, such as vehicle speed, displacement, acceleration, etc., and uses deep learning technology to integrate and analyze these data to improve the accuracy of skidding detection;

[0061] 3) The present invention uses unsupervised deep learning technology for skidding detection, and only normal data is needed to train the model; this method is suitable for the actual situation where the proportion of skidding data in all data is extremely low, and solves the problem of traditional supervised learning methods relying on abnormal data;

[0062] 4) The present invention optimizes the accuracy of the train positioning system by only using an algorithm without adding hardware equipment, which effectively saves costs; compared with existing methods, this reduces dependence on other sensors. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 Flow chart of a train skidding detection method based on deep learning in an embodiment of the present invention;

[0064] Figure 2 A schematic diagram of a train wheel slip detection model constructed by the present invention;

[0065] Figure 3 A schematic diagram of a loss curve after logarithmization when training a train wheel slip detection model based on a deep learning network according to the present invention;

[0066] Figure 4 The figure is a schematic diagram of the recognition effect of the present invention on a partial slipping section, and the shaded part in the figure is the detected slipping section. (a) is a schematic diagram of the detection effect of the model for a situation where the train wheel slips severely and then slips again after several consecutive slips, (b) is a schematic diagram of the detection effect of the model for a situation where the train wheel slips continuously for a long time, (c) is a schematic diagram of the detection effect of the model for a single and severe slip of the train wheel, and (d) is a schematic diagram of the detection effect of the model for a situation where the train wheel slips continuously and intensively;

[0067] Figure 5 This is a structural diagram of a train skid detection device based on deep learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0069] Figure 1 Flow chart of a train skidding detection method based on deep learning in an embodiment of the present invention, the specific process is as follows;

[0070] Step S1, select driving log data from actual projects, extract data related to wheel slip transmitted by sensors, including speed, displacement and acceleration, clean the data, remove obvious abnormal data such as transponder missing, etc. Then use the sliding window method to sample the used data according to the set window size and time step, and save the normal data and slip data separately for further analysis.

[0071] Step S2, building a train wheel slip detection model based on a combination of a long short-term memory network and an autoencoder;

[0072] Step S3, make a data set, normalize all the data involved in training and testing, use 70% of the normal data as the training set, 15% as the validation set, and the remaining 15% and all the slip data as the test set. Divide the data into training set, validation set and test set, and neither the validation set nor the test set participates in training;

[0073] Step S4, use the training set to train the model until the loss value calculated by the loss function converges to obtain a trained model. Step S2 The autoencoder is composed of an encoder and a decoder, and the training set is composed of normal data. The data used for training is first reduced in dimension by the encoder so that the model can learn and capture the characteristics of normal data, and then reconstructed by the decoder to restore it to the original dimension. The reconstruction error is then calculated using the loss function, and the loss function used in the method is the mean square error. During the training process, after each training round, the model is reconstructed and predicted for the validation set, and the loss value of the validation set is calculated using the loss function. The validation set does not participate in the training, which can effectively prevent overfitting. Using gradient descent, the loss values ​​of the training set and the validation set are reduced in multiple training rounds until the loss value converges.

[0074] Step S5, input the data to be detected into the trained model, the model reconstructs the input data to obtain output data, and compares and analyzes the output data with the original data to detect the slip section.

[0075] Figure 2 This is a structural diagram of a train wheel slip detection model based on a deep learning network constructed in the present invention, wherein the train wheel slip detection model is established by combining a long short-term memory network LSTM and an autoencoder.

[0076] The autoencoder is the core of the model. The autoencoder consists of an encoder and a decoder. The encoder maps the input data to a low-dimensional hidden representation space, which can capture the key features of the input data. The decoder attempts to restore the reduced-dimensional data to the original input data. The goal of the decoder is to reconstruct the input data so that the reconstructed data is as close to the original input as possible. The difference between the reconstructed data and the original data is determined by the loss function. The logic of using this model for anomaly detection is that compared to normal data, abnormal data is more difficult to be represented by a low-dimensional feature space. Therefore, when encountering abnormal data, the reconstruction error of the autoencoder will become very large, so the reconstruction error can be used to determine whether there is anomaly in the data.

[0077] The encoder and decoder of the autoencoder in the model are respectively composed of two layers of LSTM networks. Among them, the LSTM network is a special recursive neural network. The LSTM network is composed of a series of LSTM units. Each LSTM unit contains three gates: input gate, forget gate, output gate and a memory unit. These structures enable LSTM to remember and forget information. The LSTM network controls the flow of information through these gates, thereby effectively solving the problem of long-term dependence.

[0078] Figure 3 The loss curve after logarithmization is shown when the present invention is training a train wheel slip detection model based on a deep learning network. Since the data is normalized, it is not intuitive to directly use its loss value to draw the loss curve, so it is logarithmized to amplify the trend.

[0079] from Figure 3 It can be seen that the loss value drops rapidly in the initial stage, then converges, and remains at a low level after convergence. This shows that the model gradually learns and adapts to the data characteristics during the training process, the reconstruction error is significantly reduced, and finally reaches a stable low error level.

[0080] Figure 4 The present invention shows the recognition effect of some slipping sections. The data to be detected is reconstructed using the trained model. The difference between the normal data and the original data after reconstruction is small. However, when the slipping data is input into the model for reconstruction, there will be significant differences between the reconstructed data and the original data. The reconstructed data and the original data are compared and analyzed to identify the slipping section. Figure 4 The shaded part in the middle is the identified slipping section. It can be seen from the figure that this method has a good detection effect on obvious slipping, slipping sections with a long duration, and continuous small slipping.

[0081] The above is an introduction to a method embodiment. The following is a further explanation of the solution of the present invention through an apparatus embodiment.

[0082] Figure 5The structure diagram of the train skidding detection device based on deep learning in the embodiment of the present invention includes four modules: data extraction module 1, data storage module 2, model storage module 3, and skidding detection module 4. The data extraction module 1 has an interface connected to the vehicle-mounted driving log storage system, which is used to extract the data related to skidding in the driving log and transmit the extracted data to the data storage module 2; the data storage module 2 is used to store the data to be detected and the data that has been detected and the detection results; the model storage module 3 stores the trained train wheel skidding detection model 4 based on the deep learning network; the skidding detection module 4 stores the relevant judgment conditions for skidding detection, calls the model stored in the model storage module, compares the data reconstructed by the model with the original data, detects the section where skidding occurs, and transmits the processed data and results to the data storage module 2.

[0083] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0084] The embodiment of the present invention also provides an electronic device including a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or computer program instructions loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0085] Multiple components in the device are connected to the I / O interface, including: input units, such as keyboards, mice, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as disks, optical disks, etc.; and communication units, such as network cards, modems, wireless communication transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunication networks.

[0086] The processing unit performs the various methods and processes described above, such as methods S1 to S5. For example, in some embodiments, methods S1 to S5 may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via a ROM and / or a communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of methods S1 to S5 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute methods S1 to S5 in any other appropriate manner (e.g., by means of firmware).

[0087] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0088] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0089] In the context of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0090] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A train wheel slip detection method, It is characterized in that The method is implemented based on deep learning and comprises the following steps: Step S1, collecting the vehicle driving log and extracting the data of the skidding, and arranging the data related to the skidding into a data set; Step S2, constructing a train wheel slip detection model based on a combination of a long short-term memory network and an autoencoder; Step S3, creating a data set, normalizing the data, and dividing the data into a training set, a validation set, and a test set; Step S4, using the training set to train the model constructed in step S2 until the loss value calculated by the loss function converges, thereby obtaining a trained model; Step S5, using the model trained in step S4 to reconstruct the data to be detected, obtain output data, and compare the output data with the original data to determine the slip section.

2. A train wheel slip detection method according to claim 1, It is characterized in that The step S1 specifically includes the following steps: Step S1-1, extracting data related to wheel slip measured by sensors from the onboard driving log of an existing train; Step S1-2, cleaning the data to remove obviously abnormal data; Step S1-3, use the sliding window method to sample the cleaned data, and save the normal data and the slip data separately.

3. A train wheel slip detection method according to claim 2, It is characterized in that The data measured by the sensor include velocity, displacement and acceleration data.

4. A train wheel slip detection method according to claim 2, It is characterized in that The sliding window method is specifically as follows: sampling the used data according to the set window size and time step.

5. A train wheel slip detection method according to claim 1, It is characterized in that The autoencoder of the model in step S2 includes an encoder and a decoder both of which are built by long short-term memory networks, wherein the encoder is composed of two layers of long short-term memory networks, and the decoder is composed of two layers of long short-term memory networks and one fully connected layer.

6. A train wheel slip detection method according to claim 1, It is characterized in that The long short-term memory network of the model in step S2 is used to extract hidden features of the time series, and the autoencoder is used to identify the difference between abnormal data and normal data.

7. A train wheel slip detection method according to claim 1, It is characterized in that The step S3 specifically comprises the following steps: Step S3-1, normalize all data involved in training and testing; In step S3-2, 70% of the normal data are used as a training set, 15% are used as a validation set, and the remaining 15% and all the slip data form a test set.

8. A train wheel slip detection method according to claim 7, It is characterized in that The normalization process in step S3-1 is specifically as follows: Among them, x is the data to be processed, x min is the minimum value among all data, x max is the maximum value of all data, and x' is the data after normalization.

9. A train wheel slip detection method according to claim 5, It is characterized in that The step S4 specifically comprises the following steps: Step S4-1, reducing the dimension of the training set data through the encoder so that the model can learn and capture the characteristics of normal data; Step S4-2, reconstructing the data after the encoder dimensionality reduction through the decoder and restoring it to the original dimension; Step S4-3, calculating the reconstruction error using the loss function; Step S4-4, after each training round, the trained model is reconstructed and predicted using the validation set, and the loss value of the validation set is calculated using the loss function; Step S4-5, using gradient descent, reduces the loss values ​​of the training set and the validation set in multiple training rounds until the loss values ​​converge.

10. A train wheel slip detection method according to claim 9, It is characterized in that The encoder body propagation process in step S4-1 includes: Step S4-1-1, the first LSTM network layer receives input of n_features dimension and outputs hidden state of hidden_dim dimension; Step S4-1-2, the second long short-term memory network layer receives the output of the first layer and generates an embedding of the embedding_dim dimension.

11. A train wheel slip detection method according to claim 9, It is characterized in that The specific propagation process of the decoder in step S4-2 includes: Step S4-2-1, the first LSTM network layer receives the input of the embedding_dim dimension transmitted by the encoder and outputs the hidden state of the same dimension; Step S4-2-2, the second long short-term memory network layer receives the output of the first layer and generates a hidden state of the hidden_dim dimension; Step S4-2-3, the next layer is a fully connected layer that maps the hidden state to n_features dimensions.

12. A train wheel slip detection method according to claim 9, It is characterized in that The loss function in step S4-3 is a mean square error loss function, which is used to calculate the average value of the sum of squares of the differences between the predicted value and the true value. The specific formula is as follows: Among them, y i is the value after normalization of the training data, The data y is used for the model i The reconstructed value after reconstruction.

13. A train wheel slip detection method according to claim 1, It is characterized in that The step S5 specifically comprises the following steps: Step S5-1, input the data to be detected into the trained model to obtain reconstructed data; Step S5-2, after the normal data is reconstructed by the model, the reconstruction error between the normal data and the original data is less than the set threshold, but when the slipped data is input into the model for reconstruction, the difference between the reconstructed data and the original data is greater than the set threshold; Step S5-3, comparing and analyzing the reconstructed data with the original data, so as to identify the section where the slip occurs.

14. A device for the train wheel slip detection method according to claim 1, It is characterized in that It includes a data extraction module, a data storage module, a model storage module and a slip detection module; The data extraction module is connected to the vehicle driving log storage system, and is used to extract the data related to the skidding in the driving log, and transmit the extracted data to the data storage module; The data storage module is used to store the data to be detected and the data that has been detected and the detection results; The model storage module stores a trained train wheel slip detection model based on a deep learning network; The slip detection module stores relevant judgment conditions for slip detection, calls the model stored in the model storage module, compares the data reconstructed by the model with the original data, detects the section where slip occurs, and transmits the processed data and results to the data storage module.

15. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory. It is characterized in that When the processor executes the program, the method according to any one of claims 1 to 13 is implemented.

16. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the program is executed by a processor, the method according to any one of claims 1 to 13 is implemented.

Images