A transient disturbance and fault identification method, system, device and storage medium for vehicle network electrical system
Through discrete Hilbert transformation and CNN technology, electrical transient disturbances in the vehicle network system and train power supply system failures are identified, which solves the problems that cannot be identified in the existing technology and ensures the safe operation of the vehicle network electrical system.
Patent Information
- Application Number
- CN202410673284.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-28
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-05-28
AI Technical Summary
The prior art cannot simultaneously identify electrical transient disturbances in the vehicle network system and failures of multiple types of electrical equipment in the train power supply system, resulting in threats to the safe operation of the train and making it difficult to eliminate driving risks.
Using a method based on discrete Hilbert transform and CNN, a data set, preprocessing module, feature extraction network and feature fusion network are constructed to extract and fuse feature data to identify transient disturbances and faults of the electric system of the vehicle network.
It has realized the cause analysis of the abnormal transient working conditions of the electric system in the electric network, accurately identified multiple types of transient disturbances and faults, evaluated the operating status of the electric equipment in the train, checked equipment safety hazards, and ensured the smooth and safe operation of the electric system in the electric network.
Smart Images

Figure CN118535953B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of transient disturbance and fault identification of a vehicle network electrical system, and in particular to a method, system, device and storage medium for transient disturbance and fault identification of a vehicle network electrical system. Background Art
[0002] There is a close electrical coupling relationship between the train and the traction network, and the normal operation of the train network electrical system ensures the smooth operation of the train. Electric locomotives and EMU trains mostly use AC-DC or AC-DC-AC traction drive systems to convert the single-phase 25kV AC from the traction network into power for the normal operation of the train. The train draws current through the roof pantograph, and the single-phase AC passes through high-voltage transformers, main circuit breakers, lightning arresters and other high-voltage electrical equipment and then connects to the on-board transformer, and then provides power to the traction motor after the converter acts. When an electrical device fails, it may interfere with other equipment, thereby affecting the overall operation of the train and even posing a threat. On the other hand, when the train passes through the split phase area or under other special working conditions, transient disturbances such as excitation surge current may occur, which may affect trains running in the same power supply arm.
[0003] There is currently no method to simultaneously consider abnormal operating conditions caused by electrical transient disturbances in the train-grid system and failures of multiple types of electrical equipment in the train power supply system, which creates difficulties in ensuring the safe operation of trains and eliminating driving hazards. Summary of the invention
[0004] In order to solve the problems existing in the prior art, the present invention provides a method, system, device and storage medium for transient disturbance and fault identification of a vehicle network electrical system, which can simultaneously consider two major abnormal operating conditions: electrical transient disturbance of the vehicle network system and multiple types of electrical equipment failures in the train power supply system, so as to realize the cause analysis of abnormal transient conditions of the vehicle network electrical system, and solve the problems mentioned in the above background technology.
[0005] To achieve the above object, the present invention provides the following technical solution: a method for transient disturbance and fault identification of a vehicle network electrical system, comprising the following steps:
[0006] S1. Construct a data set of electrical transient disturbances of the train-grid system and equipment failure conditions of the train power supply system;
[0007] S2. Construct a preprocessing module based on discrete Hilbert transform, take all samples in the data set as input samples, perform discrete Hilbert transform on the input samples, amplify the typical features of disturbance or fault data, and enhance the differences between different types of data;
[0008] S3. Construct a CNN-based feature extraction network to transform the dimension and size of the input samples, and finally obtain four sets of feature data of transient disturbances and fault conditions of different dimensions and sizes;
[0009] S4. Construct a feature fusion network, and use four corresponding feature transformation modules to perform dimension transformation and size transformation on four sets of feature data of transient disturbances and fault conditions of different dimensions and sizes, and finally obtain four sets of feature data of the same dimension and size. The four sets of feature data are spliced according to the dimensions to obtain the fused feature data of the input sample;
[0010] S5. Based on the obtained fusion feature data, a CNN-based classification network is constructed, and the classification network is used to predict the type of input samples, thereby realizing the identification of electrical transient disturbances and faults in the vehicle network.
[0011] Preferably, in step S1, the following is specifically included: collecting the current signal on the high voltage 25kV side of the train, dividing the collected current signal, each data sample after division is a tensor of size 1×12800, and contains the current waveform data of the high voltage 25kV side of the train within 0.5s; arranging the data samples to ensure that the number of samples of different types is close, and then forming a data set;
[0012] The data set includes electrical transient disturbance data of the vehicle-grid system and equipment failure data of the train power supply system; among them, the types of electrical transient disturbances of the vehicle-grid system include excitation inrush current and response inrush current; the types of equipment failures of the train power supply system include bow-net offline, grounding on the high-voltage side of the train, short circuit between turns of the train main transformer and open circuit failure of the train rectifier IGBT.
[0013] Preferably, the preprocessing module in step S2 processes the one-dimensional time series X1∈R 1×12800 Perform discrete Hilbert transform, the transformed one-dimensional time series X2∈R 1×12800 In the transient disturbance or fault condition, the amplitude and phase of the current signal will change. Under normal conditions, the amplitude of the current signal remains approximately unchanged and the phase shifts by about 90°, thus highlighting the characteristics of the transient disturbance or fault.
[0014] Preferably, in step S3, the CNN-based feature extraction network includes a downsampling module and three feature extraction modules, namely, a first feature extraction module, a second feature extraction module, and a third feature extraction module, which gradually change the data dimension and data size of the input sample, and gradually extract data features of different dimensions and sizes, and finally obtain four groups of transient disturbance and fault condition feature data of different dimensions and sizes; wherein the reasoning process of the feature extraction module is expressed as F:x i →f i , where xi represents the i-th sample, f i Indicates the feature vector corresponding to the i-th sample;
[0015] The downsampling module is composed of two stacked maximum pooling layers; the three feature extraction modules have the same structure, and each feature extraction module is composed of a stack of a one-dimensional convolution layer, a batch normalization layer, a non-linear activation layer, and a maximum pooling layer; wherein the non-linear activation layer adopts the LeakyReLU activation function.
[0016] Preferably, in step S4, the four feature transformation modules are respectively a first feature transformation module, a second feature transformation module, a third feature transformation module, and a fourth feature transformation module; the feature data input by the four feature transformation modules come from the downsampling module, the first feature extraction module, the second feature extraction module, and the third feature extraction module respectively; the feature data output by the four feature transformation modules have the same dimension and size, and are spliced according to the dimensions to obtain fused feature data of the input samples; assuming that the feature data output by the four feature transformation modules have dimensions b and sizes 1×m, the dimension of the fused feature data is 4b and the size is 1×m.
[0017] Preferably, the first feature transformation module is composed of a stack of one-dimensional convolutional layers, batch normalization layers, nonlinear activation layers, and downsampling layers; the second feature transformation module is composed of a stack of one-dimensional convolutional layers, batch normalization layers, and nonlinear activation layers; the third feature transformation module and the fourth feature transformation module are composed of a stack of one-dimensional convolutional layers, batch normalization layers, nonlinear activation layers, upsampling layers, one-dimensional convolutional layers, batch normalization layers, and nonlinear activation layers. The nonlinear activation layer uses the LeakyReLU activation function; the downsampling layer uses the maximum pooling method; and the upsampling layer uses the linear interpolation method.
[0018] Preferably, in step S5, the CNN-based classification network includes a CNN-based feature extraction module, a Flatten layer, a fully connected layer and a softmax layer; the feature extraction module is a fourth feature extraction module, which is stacked by a one-dimensional convolution layer, a batch normalization layer, a non-linear activation layer, and a maximum pooling layer; wherein the non-linear activation layer adopts a LeakyReLU activation function;
[0019] After the fourth feature extraction module, the Flatten layer, the fully connected layer, and the softmax layer are stacked in sequence to form a classification network, thereby realizing the prediction of the type of the input sample. The prediction result using one-hot encoding is expressed as a1,a2,...,a i Only one value is 1, and the rest are 0. i 1 indicates that the sample belongs to this type of transient disturbance or fault, a iA value of 0 indicates that the sample does not belong to this type of transient disturbance or fault.
[0020] On the other hand, in order to achieve the above-mentioned purpose, the present invention also provides the following technical solution: a transient disturbance and fault identification system of a vehicle network electrical system, the system comprising:
[0021] Dataset construction module, constructing data sets of electrical transient disturbances of the train-grid system and equipment failure conditions of the train power supply system;
[0022] Preprocessing module: Construct a preprocessing module based on discrete Hilbert transform, take all samples in the data set as input samples, perform discrete Hilbert transform on the input samples, amplify the typical features of disturbance or fault data, and enhance the differences between different types of data;
[0023] Feature extraction network module, builds a CNN-based feature extraction network, performs dimension transformation and size transformation on the input samples, and finally obtains four sets of feature data of transient disturbances and fault conditions of different dimensions and sizes;
[0024] Feature fusion network module, constructing feature fusion network, using four corresponding feature transformation modules to perform dimension transformation and size transformation on four sets of feature data of transient disturbance and fault conditions of different dimensions and sizes, and finally obtaining four sets of feature data of the same dimension and size, and splicing the four sets of feature data by dimension to obtain the fused feature data of the input sample;
[0025] The classification network module builds a CNN-based classification network based on the obtained fusion feature data, and uses the classification network to predict the type of input samples, thereby realizing the identification of electrical transient disturbances and faults in the vehicle network.
[0026] On the other hand, to achieve the above-mentioned purpose, the present invention also provides the following technical solution: an electronic device, the electronic device comprising: a processor; and a memory for storing one or more programs;
[0027] When the one or more programs are executed by the processor, the processor executes the transient disturbance and fault identification method of the vehicle network electrical system.
[0028] On the other hand, to achieve the above-mentioned purpose, the present invention also provides the following technical solution: a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the transient disturbance and fault identification method of the vehicle network electrical system is implemented.
[0029] The beneficial effects of the present invention are:
[0030] 1) The transient disturbance and fault identification method of the vehicle network electrical system provided by the present invention samples and analyzes the current on the high-voltage 25kV side of the train, and can simultaneously consider two major abnormal operating conditions, namely, the electrical transient disturbance of the vehicle network system and the faults of various types of electrical equipment in the train power supply system, to achieve transient disturbance and fault identification of the vehicle network electrical system; the present invention can solve the problem that the current fault diagnosis method of the vehicle network electrical system cannot identify the electrical transient disturbance of the vehicle network system and the faults of various types of electrical equipment in the train power supply system, judge the faulty train electrical equipment and the fault type, provide a reference for evaluating the operating status of the train electrical equipment and troubleshooting the safety hazards of the equipment, and is conducive to the smooth and safe operation of the vehicle network electrical system.
[0031] 2) Transient disturbances and faults in the vehicle-grid electrical system are of various types and of short duration. Some fault types are somewhat hidden, and their waveform characteristics are similar to normal operating data. The transient disturbance and fault identification method of the vehicle-grid electrical system provided by the present invention adopts a unique network structure. The model can capture and accurately identify the subtle features of disturbances or faults, ensuring the recognition effect of multiple types of transient disturbances and faults. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 Schematic diagram of the topological structure of an electrified railway network system in an embodiment of the present invention;
[0033] Figure 2 This is a schematic diagram of the main circuit of a high-speed train in an embodiment of the present invention;
[0034] Figure 3 It is a schematic diagram of the principle structure of a transient disturbance and fault identification method of a vehicle network electrical system in an embodiment of the present invention;
[0035] Figure 4 It is a schematic flow chart of the steps of a method for identifying transient disturbance and fault of a vehicle network electrical system in an embodiment of the present invention;
[0036] Figure 5 Schematic diagram of the structure of a feature extraction module in an embodiment of the present invention;
[0037] Figure 6 Schematic diagram of the structure of a feature conversion module in an embodiment of the present invention;
[0038] Figure 7 It is a schematic diagram of a transient disturbance and fault identification system module of a vehicle network electrical system in an embodiment of the present invention;
[0039] Figure 8 This is a schematic diagram of the structure of an electronic device in an embodiment of the present invention;
[0040] In the figure, 110 - data set construction module; 120 - preprocessing module; 130 - feature extraction network module; 140 - feature fusion network module; 150 - classification network module; 210 - processor; 220 - storage. DETAILED DESCRIPTION
[0041] 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 only 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 are within the scope of protection of the present invention.
[0042] Example 1
[0043] A method for identifying transient disturbances and faults in a vehicle network electrical system comprises the following steps:
[0044] S1. Construct a data set of electrical transient disturbances of the train-grid system and equipment failure conditions of the train power supply system, and divide the data set into a training set and a test set;
[0045] In step S1, the current signal on the high-voltage 25kV side of the train is collected, and the collected current signal is divided. Each data sample after division is a tensor of size 1×12800, which contains the current waveform data of the high-voltage 25kV side of the train within 0.5s. The data samples are sorted to ensure that the number of samples of different types in the data set is relatively close, and then the samples in the data set are divided into a training set and a test set at a ratio of 4:1;
[0046] The data set includes the electrical transient disturbance data of the vehicle-grid system and the equipment failure data of the train power supply system. The types of electrical transient disturbances of the vehicle-grid system include magnetizing inrush current and response inrush current; the types of equipment failures of the train power supply system include bow-grid offline, grounding of the high-voltage side of the train, short circuit between turns of the train main transformer and open circuit failure of the train rectifier IGBT;
[0047] S2. Construct a preprocessing module based on discrete Hilbert transform to perform discrete Hilbert transform on input samples, amplify the typical features of disturbance or fault data, and enhance the differences between different types of data;
[0048] The preprocessing module is used to process the one-dimensional time series X1∈R 1×12800 Perform discrete Hilbert transform, the transformed one-dimensional time series X2∈R 1×12800 In the transient disturbance or fault condition, the amplitude and phase of the current signal will change greatly. Under normal conditions, the amplitude of the current signal remains unchanged and the phase shifts by about 90°, thus highlighting the characteristics of the transient disturbance or fault.
[0049] S3. Construct a CNN-based feature extraction network, which includes a downsampling module and three feature extraction modules, which respectively perform dimension transformation and size transformation on the input samples, and finally obtain four sets of feature data of transient disturbances and fault conditions of different dimensions and sizes;
[0050] The CNN-based feature extraction network includes a downsampling module and three feature extraction modules, namely the first feature extraction module, the second feature extraction module, and the third feature extraction module, which gradually change the data dimension and data size of the input sample, and gradually extract data features of different dimensions and sizes, and finally obtain four groups of transient disturbance and fault condition feature data of different dimensions and sizes. Among them, the reasoning process of the feature extraction module is expressed as F:x i →f i , where x i represents the i-th sample, f i Indicates the eigenvector corresponding to the i-th sample;
[0051] The downsampling module is composed of two stacked maximum pooling layers; the feature extraction modules have the same structure, each of which is composed of a stack of one-dimensional convolutional layer, batch normalization layer, nonlinear activation layer, and maximum pooling layer. The nonlinear activation layer uses the LeakyReLU activation function;
[0052] S4. Construct a feature fusion network, and use four corresponding feature transformation modules to perform dimension transformation and size transformation on four sets of feature data of transient disturbances and fault conditions of different dimensions and sizes, and finally obtain four sets of feature data of the same dimension and size. The four sets of feature data are spliced according to the dimensions to obtain the fused feature data of the input sample;
[0053] The four groups of transient disturbance and fault condition feature data of different dimensions and sizes obtained in step S3 are respectively transformed in dimension and size using corresponding feature transformation modules, and finally four groups of feature data of the same dimension and size are obtained, and the fused feature data of the input sample is obtained by splicing them according to the dimensions;
[0054] The four feature transformation modules are respectively a first feature transformation module, a second feature transformation module, a third feature transformation module, and a fourth feature transformation module, and the input feature data are respectively from the downsampling module, the first feature extraction module, the second feature extraction module, and the third feature extraction module. The feature data output by the four feature transformation modules have the same dimension and size;
[0055] The first feature transformation module is composed of a stack of one-dimensional convolutional layers, batch normalization layers, nonlinear activation layers, and downsampling layers; the second feature transformation module is composed of a stack of one-dimensional convolutional layers, batch normalization layers, and nonlinear activation layers; the third feature transformation module and the fourth feature transformation module are composed of a stack of one-dimensional convolutional layers, batch normalization layers, nonlinear activation layers, upsampling layers, one-dimensional convolutional layers, batch normalization layers, and nonlinear activation layers. The nonlinear activation layer uses the LeakyReLU activation function; the downsampling layer uses the maximum pooling method; and the upsampling layer uses the linear interpolation method.
[0056] The feature data output by the four feature transformation modules are concatenated according to the dimensions to obtain the fused feature data of the input sample. Assuming that the feature data output by the four feature transformation modules have dimensions b and sizes 1×m, the dimension of the fused feature data is 4b and the size is 1×m;
[0057] S5. Based on the obtained fusion feature data, a CNN-based classification network is constructed. A CNN-based feature extraction module is used for feature extraction, and a Flatten layer, a fully connected layer, and a softmax layer are added at the end to form a classification network, thereby realizing the identification of electrical transient disturbances and faults in the vehicle network.
[0058] The feature extraction module is the fourth feature extraction module, which is composed of a one-dimensional convolution layer, a batch normalization layer, a nonlinear activation layer, and a maximum pooling layer. The nonlinear activation layer uses the LeakyReLU activation function. After the fourth feature extraction module, the Flatten layer, the fully connected layer, and the softmax layer are stacked in sequence to form a classification network, thereby realizing the prediction of the type of the input sample. The prediction result using one-hot encoding is expressed as a1,a2,...,a i Only one value is 1, and the rest are 0. i 1 indicates that the sample belongs to this type of transient disturbance or fault, a i A value of 0 indicates that the sample does not belong to this type of transient disturbance or fault;
[0059] The steps S3, S4 and S5 together constitute a network model. In each round of the network training process, the three networks in S3, S4 and S5 complete a training process. The loss function in the training process adopts a cross entropy loss function.
[0060] Example 2
[0061] First, the applicable scenarios of the present invention are introduced. The present invention can be applied to the identification of electrical transient disturbances in the train-grid system and electrical equipment faults in the train power supply system.
[0062] The topology of the electrified railway network system is as follows: Figure 1As shown in the figure, the traction substation converts the three-phase electricity on the power system side into a single-phase 27.5kV to supply power to the overhead contact system. While the overhead contact provides energy to the train, it is also affected by the train. When the train passes through the phase separation section or the main circuit breaker is closed under no-load conditions in other working conditions, it may cause excitation surge current, resulting in obvious current distortion in the power supply arm, and may also cause other trains running in the same power supply arm to have a response surge current phenomenon. High-speed trains generally use AC-DC-AC traction drive systems. The schematic diagram of the train main circuit is shown in the figure below. Figure 2 As shown. There are a large number of electrical equipment inside the train, including lightning arresters, voltage transformers, current transformers, main circuit breakers, on-board transformers, rectifiers, etc. The normal operation of each device ensures the smooth operation of the train. When a certain device fails during operation, such as the bow network is offline, the high-voltage side is grounded, the transformer turns are short-circuited, and the rectifier IGBT is open-circuited, it will affect the normal operation of the train and cause current distortion on the 25kV side. If measures are not taken in time, other equipment may fail, causing the fault to expand, and even triggering overcurrent protection, causing the train to be unable to run.
[0063] There is currently no method to simultaneously consider abnormal operating conditions caused by electrical transient disturbances in the train-grid system and electrical equipment failures in the train power supply system. It is impossible to effectively identify transient disturbances and failures that occur during train operation, and it is difficult to analyze factors that interfere with the normal operation of trains.
[0064] In summary, the identification of transient disturbances and faults in the train-grid electrical system can identify the transient disturbances and fault conditions of the train-grid electrical system from the overall train level, provide a reference for further troubleshooting of train electrical equipment, and facilitate the safe and smooth operation of the train.
[0065] Based on this, the embodiment of the present invention provides a method for transient disturbance and fault identification of a vehicle network electrical system, the principle structure is as follows: Figure 3 As shown, the steps are as follows Figure 4 As shown, including:
[0066] S1. Construct a data set of electrical transient disturbances of the vehicle-grid system and fault conditions of electrical equipment in the train power supply system. Consider that transient disturbance types include magnetizing inrush current and response inrush current; electrical equipment fault types include pantograph-grid offline, high-voltage side grounding, on-board transformer turn-to-turn short circuit, and rectifier IGBT open circuit. In order to simplify the transient disturbance or fault identification method, the labels of the working conditions of different types of data are simplified. The specific labeling method is shown in Table 1.
[0067] Table 1 Simplified labels for transient disturbances or faults in the vehicle-grid electrical system
[0068]
[0069]
[0070] Finally, we get the dataset D all ={(x i ,y i )|x i ∈X all ,y i ∈Y all}, X all is the sample data, Y all is the type of transient disturbance or fault corresponding to the sample data.
[0071] Based on this, we build a training dataset D tr ={(x i ,y i )|x i ∈X tr ,y i ∈Y all} and test dataset D te ={(x i ,y i )|x i ∈X te ,y i ∈Y all}. Among them, X tr , X te The data for the training set and test set samples are:
[0072]
[0073] The sampling frequency of the current signal is 25600 Hz, and the data of each sample is a tensor of size 1×12800, which contains the waveform data of the current I on the high-voltage 25kV side of the train within 0.5s.
[0074] S2. Construct a preprocessing module based on discrete Hilbert transform to perform discrete Hilbert transform on the input samples, amplify the typical features of disturbance or fault data, and enhance the differences between different types of data. This preprocessing module performs discrete Hilbert transform on all samples in the original data set one by one: for any sample (x i ,y i ), keep the type y of the sample i unchanged, for the sample data x i Perform discrete Hilbert transform, sample data x i It can be expressed as: i [n], n∈1,2,…,12800, then the sample data x after discrete Hilbert transform i The calculation process of ' can be expressed as:
[0075]
[0076] After discrete Hilbert transform, the training set becomes D tr '={(x i ',y i )|x i '∈X tr ',y i ∈Y all}, the test set becomes D te '={(x i ',y i )|x i '∈X te ',y i ∈Y all}, where X tr ', X te ' are the training set and test set sample data after discrete Hilbert transform.
[0077] The preprocessing module is used to process the one-dimensional time series X1∈R 1×12800 Perform discrete Hilbert transform, the transformed one-dimensional time series X2∈R 1×12800 In the data, the amplitude of the current signal under normal conditions is approximately unchanged, and the phase is shifted by about 90°. However, the amplitude and phase of the current signal under transient disturbance or fault conditions will change significantly, which is obviously different from the current signal under normal conditions, thus highlighting the data characteristics of transient disturbances or faults and laying the foundation for the identification of transient disturbances or faults.
[0078] S3. Build a CNN-based feature extraction network, which includes a downsampling module and three feature extraction modules.
[0079] The downsampling module consists of two stacked maximum pooling layers, which changes the input sample size from 1×12800 to 1×800 to avoid the phenomenon of excessive model calculation caused by too complex network structure and too many parameters. The three feature extraction modules are the first feature extraction module, the second feature extraction module, and the third feature extraction module. The three feature extraction modules have the same structure, such as Figure 5 As shown in the figure, they are all stacked by one-dimensional convolution layer, batch normalization layer, nonlinear activation (LeakyReLU) layer, and maximum pooling layer, which gradually change the data dimension and data size of the input sample, and gradually extract data features of different dimensions and sizes, and finally obtain four sets of transient disturbance and fault condition feature data of different dimensions and sizes. Among them, the reasoning process of the feature extraction module is expressed as F:x i →f i , where x i represents the i-th sample, f i Indicates the eigenvector corresponding to the i-th sample;
[0080] S4. Construct a feature fusion network, input four sets of feature data of transient disturbances and fault conditions of different dimensions and sizes, use corresponding feature transformation modules to perform dimension transformation and size transformation respectively, and finally obtain four sets of feature data of the same dimension and size, and then splice them according to the dimensions to obtain the fused feature data of the input sample;
[0081] The four feature transformation modules used are respectively a first feature transformation module, a second feature transformation module, a third feature transformation module and a fourth feature transformation module, and the input feature data come from a downsampling module, a first feature extraction module, a second feature extraction module and a third feature extraction module respectively;
[0082] The structures of the four feature transformation modules are as follows: Figure 6 As shown. The first feature transformation module is composed of a stack of one-dimensional convolutional layers, batch normalization layers, non-linear activation (LeakyReLU) layers, and downsampling layers, and its function is to increase the dimension and reduce the size of the input feature data; the second feature transformation module is composed of a stack of one-dimensional convolutional layers, batch normalization layers, and non-linear activation (LeakyReLU) layers, and only reduces the dimension of the input feature data while keeping the size unchanged; the third feature transformation module and the fourth feature transformation module are composed of a stack of one-dimensional convolutional layers, batch normalization layers, non-linear activation (LeakyReLU) layers, upsampling layers, one-dimensional convolutional layers, batch normalization layers, and non-linear activation (LeakyReLU) layers, and their function is to reduce the dimension and increase the size of the input feature data. Among them, the downsampling layer adopts the maximum pooling method; the upsampling layer adopts the linear interpolation method;
[0083] The feature data output by the four feature transformation modules have the same dimension and size. After splicing according to the dimension, the fused feature data of the input sample is obtained. Assuming that the dimension of the feature data output by the four feature transformation modules is b and the size is 1×m, the dimension of the fused feature data is 4b and the size is 1×m;
[0084] S5. Based on the obtained fusion feature data, a CNN-based classification network is constructed. A CNN-based feature extraction module is used for feature extraction, and a Flatten layer, a fully connected layer, and a softmax layer are added at the end to form a classification network, thereby realizing the identification of electrical transient disturbances and faults in the vehicle network.
[0085] The feature extraction module is the fourth feature extraction module, and its structure is as follows: Figure 5 As shown in the figure, it is composed of a one-dimensional convolutional layer, a batch normalization layer, a nonlinear activation (LeakyReLU) layer, and a maximum pooling layer. After the fourth feature extraction module, the Flatten layer, the fully connected layer, and the softmax layer are stacked in sequence to form a classification network, thereby realizing the prediction of the type of the input sample. The prediction result using one-hot encoding is expressed as a1,a2,...,a i Only one value is 1, and the rest are 0. i 1 indicates that the sample belongs to this type of transient disturbance or fault, a i A value of 0 indicates that the sample does not belong to this type of transient disturbance or fault;
[0086] Steps S3, S4 and S5 together constitute the recognition network model. In each round of the network training process, the three networks in S3, S4 and S5 complete a training process. During the training process, the Adam algorithm is selected as the optimization strategy, and the loss function adopts the cross entropy loss function.
[0087] The present invention can solve the problem that the current fault diagnosis method of the vehicle network electrical system is unable to identify the electrical transient disturbance of the vehicle network system and various types of electrical equipment faults in the train power supply system, and judge the faulty train electrical equipment and the fault type, so as to provide a reference for evaluating the operating status of the train electrical equipment and troubleshooting the safety hazards of the equipment, which is beneficial to the smooth and safe operation of the vehicle network electrical system.
[0088] Based on the same inventive concept as the above method embodiment, the present application embodiment also provides a transient disturbance and fault identification system for a vehicle network electrical system, which can implement the functions provided by the above method embodiment, such as Figure 7 As shown, the system includes the following modules:
[0089] A data set construction module 110 constructs a data set of electrical transient disturbances of the train-grid system and equipment failure conditions of the train power supply system;
[0090] Preprocessing module 120, constructing a preprocessing module based on discrete Hilbert transform, taking all samples in the data set as input samples, performing discrete Hilbert transform on the input samples, amplifying typical features of disturbance or fault data, and enhancing the differences between different types of data;
[0091] The feature extraction network module 130 constructs a feature extraction network based on CNN, performs dimension transformation and size transformation on the input samples respectively, and finally obtains four sets of feature data of transient disturbances and fault conditions of different dimensions and sizes;
[0092] The feature fusion network module 140 constructs a feature fusion network, and uses four corresponding feature transformation modules to perform dimension transformation and size transformation on four sets of feature data of transient disturbances and fault conditions of different dimensions and sizes, and finally obtains four sets of feature data of the same dimension and size, and splices the four sets of feature data by dimension to obtain fused feature data of the input sample;
[0093] The classification network module 150 constructs a CNN-based classification network based on the obtained fusion feature data, and uses the classification network to predict the type of the input sample, thereby realizing the identification of electrical transient disturbances and faults in the vehicle network.
[0094] Based on the same inventive concept as the above method embodiment, the present application embodiment also provides an electronic device, such as Figure 8 As shown, the device includes: a processor 210; and a memory 220, for storing one or more programs;
[0095] When the one or more programs are executed by the processor 210, the processor executes the transient disturbance and fault identification method of the vehicle network electrical system.
[0096] The transient disturbance and fault identification method of the vehicle network electrical system comprises the following steps:
[0097] Construct a data set of electrical transient disturbances in the train-grid system and equipment failure conditions in the train power supply system;
[0098] Construct a preprocessing module based on discrete Hilbert transform, take all samples in the data set as input samples, perform discrete Hilbert transform on the input samples, amplify the typical features of disturbance or fault data, and enhance the differences between different types of data;
[0099] A CNN-based feature extraction network is constructed to transform the dimension and size of the input samples, and finally obtain four sets of feature data of transient disturbances and fault conditions of different dimensions and sizes;
[0100] Construct a feature fusion network, and use the corresponding four feature transformation modules to perform dimension transformation and size transformation on four sets of feature data of transient disturbances and fault conditions of different dimensions and sizes, and finally obtain four sets of feature data with the same dimensions and sizes. The four sets of feature data are spliced according to the dimensions to obtain the fused feature data of the input sample;
[0101] Based on the obtained fusion feature data, a CNN-based classification network is constructed, and the classification network is used to predict the type of input samples, thereby realizing the identification of electrical transient disturbances and faults in the vehicle network.
[0102] Based on the same inventive concept as the above method embodiment, the embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by the processor 210, the transient disturbance and fault identification method of the vehicle network electrical system is implemented.
[0103] The transient disturbance and fault identification method of the vehicle network electrical system comprises the following steps:
[0104] Construct a data set of electrical transient disturbances in the train-grid system and equipment failure conditions in the train power supply system;
[0105] Construct a preprocessing module based on discrete Hilbert transform, take all samples in the data set as input samples, perform discrete Hilbert transform on the input samples, amplify the typical features of disturbance or fault data, and enhance the differences between different types of data;
[0106] A CNN-based feature extraction network is constructed to transform the dimension and size of the input samples, and finally obtain four sets of feature data of transient disturbances and fault conditions of different dimensions and sizes;
[0107] Construct a feature fusion network, and use the corresponding four feature transformation modules to perform dimension transformation and size transformation on four sets of feature data of transient disturbances and fault conditions of different dimensions and sizes, and finally obtain four sets of feature data with the same dimensions and sizes. The four sets of feature data are spliced according to the dimensions to obtain the fused feature data of the input sample;
[0108] Based on the obtained fusion feature data, a CNN-based classification network is constructed, and the classification network is used to predict the type of input samples, thereby realizing the identification of electrical transient disturbances and faults in the vehicle network.
[0109] The present invention can capture and accurately identify the minute features of disturbances or faults, thereby ensuring the recognition effect of multiple types of transient disturbances and faults.
[0110] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for identifying transient disturbances and faults in a vehicle-grid electrical system, characterized in that: The steps include: S1. Construct a data set of electrical transient disturbances of the train-grid system and equipment failure conditions of the train power supply system; S2. Construct a preprocessing module based on discrete Hilbert transform, take all samples in the data set as input samples, perform discrete Hilbert transform on the input samples, amplify the typical features of disturbance or fault data, and enhance the differences between different types of data; S3. Construct a CNN-based feature extraction network to transform the dimension and size of the input samples, and finally obtain four sets of feature data of transient disturbances and fault conditions of different dimensions and sizes; S4. Construct a feature fusion network, and use four corresponding feature transformation modules to perform dimension transformation and size transformation on four sets of feature data of transient disturbances and fault conditions of different dimensions and sizes, and finally obtain four sets of feature data of the same dimension and size. The four sets of feature data are spliced according to the dimensions to obtain the fused feature data of the input sample; The four feature transformation modules are respectively a first feature transformation module, a second feature transformation module, a third feature transformation module, and a fourth feature transformation module; the feature data input by the four feature transformation modules come from the downsampling module, the first feature extraction module, the second feature extraction module, and the third feature extraction module respectively; the feature data output by the four feature transformation modules have the same dimension and size, and the fused feature data of the input sample is obtained by splicing according to the dimension; assuming that the dimension of the feature data output by the four feature transformation modules is b and the size is 1×m, then the dimension of the fused feature data is 4b and the size is 1×m; The first feature transformation module is composed of a stack of a one-dimensional convolution layer, a batch normalization layer, a nonlinear activation layer, and a downsampling layer; The second feature transformation module is composed of a stack of one-dimensional convolutional layers, batch normalization layers, and nonlinear activation layers; The third feature transformation module and the fourth feature transformation module are stacked by a one-dimensional convolution layer, a batch normalization layer, a nonlinear activation layer, an upsampling layer, a one-dimensional convolution layer, a batch normalization layer, and a nonlinear activation layer; wherein the nonlinear activation layer adopts a LeakyReLU activation function; and the downsampling layer adopts a maximum pooling method; The upsampling layer uses linear interpolation; S5. Based on the obtained fusion feature data, a CNN-based classification network is constructed, and the classification network is used to predict the type of input samples, thereby realizing the identification of electrical transient disturbances and faults in the vehicle network.
2. The method for transient disturbance and fault identification of a vehicle network electrical system according to claim 1, characterized in that: In step S1, the following is specifically included: collecting the current signal on the high-voltage 25kV side of the train, dividing the collected current signal, and each data sample after division is a tensor of size 1×12800, containing the current waveform data of the high-voltage 25kV side of the train within 0.5s; arranging the data samples to ensure that the number of samples of different types is close, and then forming a data set; The data set includes electrical transient disturbance data of the vehicle-grid system and equipment failure data of the train power supply system; among them, the types of electrical transient disturbances of the vehicle-grid system include excitation inrush current and response inrush current; the types of equipment failures of the train power supply system include bow-net offline, grounding on the high-voltage side of the train, short circuit between turns of the train main transformer and open circuit failure of the train rectifier IGBT.
3. The method for transient disturbance and fault identification of a vehicle network electrical system according to claim 1, characterized in that: The preprocessing module in step S2 processes the one-dimensional time series X1∈R 1×12800 Perform discrete Hilbert transform, the transformed one-dimensional time series X2∈R 1×12800 In the transient disturbance or fault condition, the amplitude and phase of the current signal will change. Under normal conditions, the amplitude of the current signal remains approximately unchanged and the phase shifts by about 90°, thus highlighting the characteristics of the transient disturbance or fault.
4. The method for transient disturbance and fault identification of a vehicle-grid electrical system according to claim 1, characterized in that: In step S3, the CNN-based feature extraction network includes a downsampling module and three feature extraction modules, namely, a first feature extraction module, a second feature extraction module, and a third feature extraction module, which gradually change the data dimension and data size of the input sample, and gradually extract data features of different dimensions and sizes, and finally obtain four groups of transient disturbance and fault condition feature data of different dimensions and sizes; wherein the reasoning process of the feature extraction module is expressed as F:x i →f i , where x i represents the i-th sample, f i Indicates the feature vector corresponding to the i-th sample; The downsampling module is composed of two stacked maximum pooling layers; the three feature extraction modules have the same structure, and each feature extraction module is composed of a stack of a one-dimensional convolution layer, a batch normalization layer, a non-linear activation layer, and a maximum pooling layer; wherein the non-linear activation layer adopts the LeakyReLU activation function.
5. The method for transient disturbance and fault identification of a vehicle network electrical system according to claim 1, characterized in that: In step S5, the CNN-based classification network includes a CNN-based feature extraction module, a Flatten layer, a fully connected layer and a softmax layer; the feature extraction module is the fourth feature extraction module, which is stacked by a one-dimensional convolution layer, a batch normalization layer, a non-linear activation layer, and a maximum pooling layer; wherein the non-linear activation layer adopts a LeakyReLU activation function; After the fourth feature extraction module, the Flatten layer, the fully connected layer, and the softmax layer are stacked in sequence to form a classification network, thereby realizing the prediction of the type of the input sample. The prediction result using one-hot encoding is expressed as a1,a2,...,a i Only one value is 1, and the rest are 0. i 1 indicates that the sample belongs to this type of transient disturbance or fault, a i A value of 0 indicates that the sample does not belong to this type of transient disturbance or fault.
6. A system according to the method for identifying transient disturbance and fault of a vehicle network electrical system according to any one of claims 1 to 5, characterized in that: The system comprises the following: A data set construction module (110) constructs a data set of electrical transient disturbances of the train network system and equipment failure conditions of the train power supply system; A preprocessing module (120) is constructed based on discrete Hilbert transform, taking all samples in the data set as input samples, performing discrete Hilbert transform on the input samples, amplifying typical features of disturbance or fault data, and enhancing the differences between different types of data; A feature extraction network module (130) is used to construct a CNN-based feature extraction network, and to perform dimension transformation and size transformation on the input samples, thereby finally obtaining four sets of feature data of transient disturbances and fault conditions of different dimensions and sizes; A feature fusion network module (140) constructs a feature fusion network, and uses four corresponding feature transformation modules to perform dimension transformation and size transformation on four groups of feature data of transient disturbances and fault conditions of different dimensions and sizes, and finally obtains four groups of feature data of the same dimension and size, and splices the four groups of feature data according to the dimensions to obtain fused feature data of the input sample; The classification network module (150) constructs a classification network based on CNN based on the obtained fusion feature data, and uses the classification network to predict the type of the input sample, thereby realizing the identification of electrical transient disturbances and faults of the vehicle network.
7. An electronic device, characterized in that: The electronic device comprises: a processor (210); and a memory (220) for storing one or more programs; When the one or more programs are executed by the processor (210), the processor executes the transient disturbance and fault identification method of the vehicle network electrical system as described in any one of claims 1 to 5.
8. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by the processor (210), the method for transient disturbance and fault identification of a vehicle network electrical system as described in any one of claims 1 to 5 is implemented.
