GSM-R service quality analysis processing method and device based on neural network

Abstract

The invention provides a GSM-R service quality analysis processing method and device based on a neural network, and relates to the technical field of communication technology analysis processing, and the method comprises the steps: obtaining GSM-R waveform data, inputting the GSM-R waveform data to a service quality classification recognition model of a GSM-R system, and carrying out the classification recognition of the service quality of the GSM-R system; analyzing to obtain a corresponding service quality problem of the GSM-R system under the condition of the input waveform data; the service quality classification and identification model adopts a ResNet architecture, and the training mode is as follows: marking the category of a service quality problem, and dividing a marked sample set; training the artificial intelligence model by using the training set, performing mapping learning from the GSM-R waveform data to the category of the service quality problem on the artificial intelligence model in combination with the category of the service quality problem, and performing verification and testing by using the verification set and the test set to obtain a service quality classification identification model; and the GSM-R waveform data and the service quality problem are displayed through the railway facility monitoring platform.

Description

Technical Field

[0001] The present invention relates to the field of communication technology analysis and processing technology, and in particular to a GSM-R service quality analysis and processing method and device based on neural network. Background Art

[0002] This section is intended to provide a background or context for embodiments of the present invention. No description herein is admitted to be prior art by virtue of its inclusion in this section.

[0003] As a digital wireless communication system designed specifically for railway communications, GSM-R not only has excellent security and efficiency, but also can support a variety of data communication services. It has effectively promoted the modernization of railway transportation production command means, met the comprehensive needs of train dispatching, control and railway informatization development, and provided solid communication guarantee for the digital transformation and intelligent upgrading of the railway industry. As of the end of December 2020, GSM-R covered tens of thousands of kilometers of lines. Through in-depth research and analysis of the service quality indicators of the GSM-R system, potential problems in the network can be identified, and then corresponding optimization measures can be taken. These optimization measures include but are not limited to adjusting network parameters, optimizing network layout, upgrading hardware equipment, and improving software algorithms. Through continuous practice and adjustment, the service quality of the network can be gradually improved to meet the growing business needs. The electromagnetic environment along the railway is complex, and GSM-R is seriously interfered by mobile operators (GSM). During the communication detection process, testers need to report four types of problems: dropped calls, switching failures, switching abnormalities, and poor voice quality according to the waveform diagram of the GSM-R service quality detection. Despite the continuous advancement of technology, traditional GSM-R service quality issues still rely on the rich professional knowledge and experience of testers to carefully identify and judge various problems with the naked eye, and to promptly discover and deal with complex situations that are difficult for automated tools to detect. In order to further improve the accuracy and efficiency of detection, we are actively exploring the possibility of combining advanced technology with manual detection to achieve better quality and intelligent detection results.

[0004] With the continuous development of deep learning technology, the railway industry has adopted a large number of deep learning methods in defect identification, which greatly optimizes the efficiency and accuracy of defect identification. For example, the self-attention mechanism neural network is used to identify the important components of the overhead contact power supply line of the railway, such as the string clamp bolt, to improve the recognition accuracy while reducing the amount of calculation and parameters. At the same time, the image is expanded to address the problem of insufficient defect samples in the data set. The improved component-level segmentation method of railway bridge point cloud based on RandLA-Net solves the key problem in the construction progress management of railway bridges, that is, how to more accurately extract component-level information from point cloud data. By adopting the self-attention mechanism, the network can better capture the long-range dependencies and semantic relationships between points. In order to innovate the lost property search method in the railway passenger transport scenario, based on the analysis of the needs and difficulties of lost property search in railway passenger transport, combined with the cutting-edge technical achievements of face recognition and deep learning, a lost property search framework based on image search is established, and the security inspection lost property search scheme and non-security inspection lost property search scheme for railway passenger transport scenarios are designed. The research results show that this method can further improve the intelligence level of railway passenger transport and optimize the efficiency of lost property search. Based on the data characteristics of the waveform report problem of GSM-R service quality detection, we draw on the deep learning theory that has achieved significant performance improvements in computer vision tasks to improve detection efficiency. With the continuous development of deep learning technology, the neural network structure has gradually developed from shallow to deep. However, traditional neural networks will experience performance degradation when the number of layers increases, which is mainly attributed to the gradient disappearance and gradient explosion problems. In order to solve these problems, the ResNet network was proposed in 2015. In addition to achieving significant performance in image classification tasks, ResNet also performs well in various tasks such as target detection, image segmentation, and natural language processing. This proves that ResNet has strong generalization ability and adaptability and can be applied to various deep learning tasks.

[0005] In summary, deep learning has been widely used in many fields. However, there is no relevant research on the problem of waveform reporting based on GSM-R service quality detection. There is an urgent need for a technical solution that can overcome the above defects, improve the GSM-R service quality detection method, and improve the detection effect. Summary of the invention

[0006] In order to solve the problems existing in the prior art, the present invention proposes a GSM-R service quality analysis and processing method and device based on neural network. Aiming at the classification of GSM-R service quality problems, the present invention proposes a GSM-R image recognition algorithm based on ResNet network; the GSM-R image recognition algorithm is constructed by using ResNet network, and the case library is enriched for various service quality problems, providing strong technical support for service quality analysis.

[0007] In a first aspect of an embodiment of the present invention, a GSM-R service quality analysis and processing method based on a neural network is proposed, comprising:

[0008] Get GSM-R waveform data;

[0009] The GSM-R waveform data is input into the service quality classification and recognition model of the GSM-R system, and the service quality problem corresponding to the GSM-R system under the input waveform data condition is obtained through analysis; wherein the service quality classification and recognition model adopts a ResNet architecture and is pre-trained in the following manner:

[0010] For the GSM-R waveform data sample set, the categories of service quality issues are marked, and the marked sample set is divided into a training set, a validation set, and a test set;

[0011] The artificial intelligence model is trained using the training set, and in combination with the category of the service quality problem, the artificial intelligence model is subjected to mapping learning from GSM-R waveform data to the category of the service quality problem, and the validation set and the test set are used for validation and testing to obtain a service quality classification and recognition model; wherein the service quality classification and recognition model takes the GSM-R waveform data as input and takes the category of the service quality problem as output;

[0012] The railway facility monitoring platform displays the acquired GSM-R waveform data and the service quality issues obtained through analysis.

[0013] In a second aspect of an embodiment of the present invention, a GSM-R service quality analysis and processing device based on a neural network is proposed, comprising:

[0014] A data acquisition module, used to acquire GSM-R waveform data;

[0015] The service quality analysis module is used to input the GSM-R waveform data into the service quality classification and recognition model of the GSM-R system, and obtain the service quality problem corresponding to the GSM-R system under the input waveform data condition through analysis; wherein, the service quality classification and recognition model adopts the ResNet architecture and is pre-trained in the following manner:

[0016] For the GSM-R waveform data sample set, the categories of service quality issues are marked, and the marked sample set is divided into a training set, a validation set, and a test set;

[0017] The artificial intelligence model is trained using the training set, and in combination with the category of the service quality problem, the artificial intelligence model is subjected to mapping learning from GSM-R waveform data to the category of the service quality problem, and the validation set and the test set are used for validation and testing to obtain a service quality classification and recognition model; wherein the service quality classification and recognition model takes the GSM-R waveform data as input and takes the category of the service quality problem as output;

[0018] The display module is used to display the GSM-R waveform data obtained through the railway facility monitoring platform and the service quality problems obtained through analysis.

[0019] In a third aspect of an embodiment of the present invention, a computer device is proposed, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements a neural network-based GSM-R service quality analysis and processing method when executing the computer program.

[0020] In a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is proposed, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, a GSM-R service quality analysis and processing method based on a neural network is implemented.

[0021] In a fifth aspect of an embodiment of the present invention, a computer program product is proposed. The computer program product includes a computer program. When the computer program is executed by a processor, a GSM-R service quality analysis and processing method based on a neural network is implemented.

[0022] The neural network-based GSM-R service quality analysis and processing method and device proposed in the present invention analyzes a large number of GSM-R service quality problem pictures, sorts out the abnormal patterns of GSM-R service quality problem pictures, determines the characteristic laws of its abnormalities, and constructs the optimal algorithm flow for this characteristic law. The relevant parameters and thresholds in the present invention are determined based on the results of a large number of experiments with the best effects. Compared with the prior art, it has at least the following technical effects: based on the data set generalization method of transfer learning, it extracts the characteristics of GSM-R service quality problem pictures and enriches the multi-source heterogeneous data samples of data analysts; through the GSM-R image recognition algorithm based on the ResNet network, it assists detection personnel to realize automatic GSM-R service quality problem identification and improves detection efficiency. By enriching the GSM-R service quality problem sample library, it can provide valuable experimental materials for data analysts and developers and provide strong technical support for the identification of multi-category communication problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1 It is a flowchart of a GSM-R service quality analysis and processing method based on a neural network according to an embodiment of the present invention.

[0025] Figure 2 It is a schematic diagram of reporting a dropped call problem according to a specific embodiment of the present invention.

[0026] Figure 3 It is a schematic diagram of not reporting a dropped call problem in a specific embodiment of the present invention.

[0027] Figure 4 It is a schematic diagram of handover failure reporting according to a specific embodiment of the present invention.

[0028] Figure 5 It is a schematic diagram of a switching failure not being reported according to a specific embodiment of the present invention.

[0029] Figure 6 It is a schematic diagram of switching abnormality reporting according to a specific embodiment of the present invention.

[0030] Figure 7 It is a schematic diagram of switching abnormality reporting according to another specific embodiment of the present invention.

[0031] Figure 8 It is a schematic diagram of reporting poor voice quality according to a specific embodiment of the present invention.

[0032] Fig. 9 It is a flowchart of model training according to a specific embodiment of the present invention.

[0033] Fig.10 It is a schematic diagram of the ResNet network structure of a specific embodiment of the present invention.

[0034] Fig.11 It is a schematic diagram of the ResNet18 network structure of a specific embodiment of the present invention.

[0035] Fig.12 It is a schematic diagram of the relationship of jump connections in a specific embodiment of the present invention.

[0036] Fig.13A It is a schematic diagram of the loss value of each iteration of a specific embodiment of the present invention.

[0037] Fig. 13B It is a schematic diagram of the exact value of each iteration of a specific embodiment of the present invention.

[0038] Fig.14 It is a schematic diagram of the processing flow of a disease case library according to a specific embodiment of the present invention.

[0039] Fig.15 It is a schematic diagram of a query interface of a communication professional disease database according to a specific embodiment of the present invention.

[0040] Fig.16 It is a schematic diagram of the interface for importing a communication professional disease library according to a specific embodiment of the present invention.

[0041] Fig.17 It is a schematic diagram of the architecture of a GSM-R service quality analysis and processing device based on a neural network according to an embodiment of the present invention.

[0042] Fig.18 It is a schematic diagram of the architecture of a GSM-R service quality analysis and processing device based on a neural network according to another embodiment of the present invention.

[0043] Fig.19 It is a schematic diagram of the structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.

[0045] Those skilled in the art will appreciate that the embodiments of the present invention may be implemented as a system, device, apparatus, method or computer program product. Therefore, the present disclosure may be specifically implemented in the following forms, namely: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0046] According to the implementation mode of the present invention, a method and device for analyzing and processing GSM-R service quality based on neural network is proposed. The present invention relates to the application of intelligent classification and decision-making in the field of railway detection and monitoring by railway communication technology professionals. As the digital wireless communication system of railways, GSM-R (railway dedicated mobile communication system) is interfered by the electromagnetic environment. The detection personnel report four types of problems, namely, dropped calls, switching failures, switching anomalies, and poor voice quality, based on the waveform diagram of GSM-R service quality detection. The GSM-R image recognition and intelligent classification and processing of service quality problems based on the ResNet network are of great significance to further improve the accuracy and efficiency of detection.

[0047] The principle and spirit of the present invention are explained in detail below with reference to several representative embodiments of the present invention.

[0048] Figure 1 FIG. 1 is a flow chart of a GSM-R service quality analysis and processing method based on a neural network according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0049] S101, obtaining GSM-R waveform data;

[0050] S102, inputting the GSM-R waveform data into a service quality classification and recognition model of the GSM-R system, and obtaining the service quality problem corresponding to the GSM-R system under the input waveform data condition through analysis; wherein the service quality classification and recognition model adopts a ResNet architecture and is pre-trained in the following manner:

[0051] For the GSM-R waveform data sample set, the categories of service quality issues are marked, and the marked sample set is divided into a training set, a validation set, and a test set;

[0052] The artificial intelligence model is trained using the training set, and in combination with the category of the service quality problem, the artificial intelligence model is subjected to mapping learning from GSM-R waveform data to the category of the service quality problem, and the validation set and the test set are used for validation and testing to obtain a service quality classification and recognition model; wherein the service quality classification and recognition model takes the GSM-R waveform data as input and takes the category of the service quality problem as output;

[0053] S103, displaying the acquired GSM-R waveform data and the service quality issues obtained through analysis through the railway facility monitoring platform.

[0054] The neural network-based GSM-R service quality analysis and processing method and device proposed in the present invention analyzes a large number of GSM-R service quality problem pictures, sorts out the abnormal patterns of GSM-R service quality problem pictures, determines the characteristic laws of its abnormalities, and constructs the optimal algorithm flow for this characteristic law. The relevant parameters and thresholds in the present invention are determined based on the results of a large number of experiments with the best effects. Compared with the prior art, it has at least the following technical effects: based on the data set generalization method of transfer learning, it extracts the characteristics of GSM-R service quality problem pictures and enriches the multi-source heterogeneous data samples of data analysts; through the GSM-R image recognition algorithm based on the ResNet network, it assists detection personnel to realize automatic GSM-R service quality problem identification and improves detection efficiency. By enriching the GSM-R service quality problem sample library, it can provide valuable experimental materials for data analysts and developers and provide strong technical support for the identification of multi-category communication problems.

[0055] In order to explain the above GSM-R service quality analysis and processing method based on neural network more clearly, each step is described in detail below.

[0056] In one embodiment, the categories of the service quality issues include at least: dropped calls, handover failures, handover anomalies, and poor voice quality.

[0057] Specifically, for the service quality problem of dropped calls, when the test interface prompts that the connection is lost and there is continuous voice quality reaching the set level, the corresponding position is determined to be dropped; when the test interface prompts that the connection is lost and there is no voice quality problem, the line history data is played back. If the same phenomenon occurs at the same position, the corresponding position is determined to be dropped; if the same phenomenon does not occur, it is determined that the test system is abnormal;

[0058] For the service quality problem of switching failure, when the test interface prompts that the switching has failed, it is determined that the corresponding position has failed to switch;

[0059] For the service quality problem of abnormal handover, when no handover occurs between two base stations or multiple handovers occur, it is judged as abnormal handover;

[0060] For the service quality problem of poor voice quality, when the voice quality on the test interface is within the set level range and no call drop occurs, it is judged that the voice quality is poor.

[0061] Different service quality issues are described below in conjunction with a specific embodiment.

[0062] 1. Dropped calls:

[0063] When the test interface prompts that the connection is lost and there is a continuous voice quality level of 7, it is determined that the call is dropped at this location. Figure 2 FIG. 1 is a schematic diagram of reporting a call drop problem according to a specific embodiment of the present invention. Device 1001 corresponds to Lushan, device 1012 corresponds to JJK-LS01, and device 1002 corresponds to JJN-JJX01.

[0064] When the test interface prompts that the connection is lost, but there is no poor voice quality, in order to avoid false alarms, it is necessary to replay the historical data of the line last month. If the same phenomenon occurs at the same location last month, it is determined that the call is dropped at this location; if there is no abnormality last month, it is determined that the test system has a sudden abnormality, and the problem will not be reported for the time being and continued observation. Figure 3 FIG. 1 is a schematic diagram of a specific embodiment of the present invention in which a call drop problem is not reported. Device 1012 corresponds to ZX-BJ03.

[0065] 2. Switching failure:

[0066] When the test interface prompts that the switch fails, it is determined that the position switch fails, such as Figure 4 As shown, device 1004 corresponds to Luo source 02, device 1002 corresponds to Luo source 01, and device 1007 corresponds to Luo source.

[0067] When the test module works for a long time, occasional exceptions may occur, which are manifested as the test interface prompting a switch failure and the "switch command" and "switch failure" signals at the same timestamp. In this case, do not report the problem for the time being and continue to observe. Figure 5 shown.

[0068] 3. Switching abnormality:

[0069] There should be one handover between two base stations. If there is no handover or multiple handovers between the two base stations, it is considered as a handover anomaly (co-line or hub area is considered separately). For this type of problem, the following two aspects need to be considered when reporting the problem:

[0070] If the ping-pong switching is caused by continuous cross-coverage of field strength, for C3 lines, if the overall interference rate index is qualified, it will not be reported, but the person in charge of maintenance of the railway bureau still needs to be informed; for C2 lines and conventional lines, if it does not affect the voice quality, it will not be reported, but the person in charge of maintenance of the railway bureau still needs to be informed.

[0071] If the current service cell is forced to switch from high level to low level, it is considered to be caused by equipment and interference, and needs to be reported, such as Figure 6 , 7 As shown. Figure 6 In the , device 1001 corresponds to DXD-YZ04, device 1003 corresponds to DXD-YZ03, and device 1009 corresponds to DXB-YZ02. Figure 7 In the figure, device 1008 corresponds to TJN-CZX24, device 1004 corresponds to TJN-CZX23, device 1000 corresponds to TJN-CZX22, and device 1013 corresponds to TJN-CZX21.

[0072] 4. Poor voice quality:

[0073] When the test interface has a voice quality of 6-7, but no call drops occur, it is judged as poor voice quality. This type of problem has different treatment methods for different lines. For C3 lines, if the overall interference rate index is qualified, no report is required, but the person in charge of the railway bureau maintenance still needs to be informed; for C2 lines and ordinary speed lines, if the voice quality is 5-7 for more than 500m, a report is required, such as Figure 8 As shown, device 1003 corresponds to HeFeiXiA, device 1010 corresponds to HFN-CAJ02-BU, and device 1004 corresponds to HFN-CAJ01.

[0074] In one embodiment, for S101 , GSM-R waveform data is acquired.

[0075] Furthermore, the GSM-R waveform data is subjected to image preprocessing; wherein, the GSM-R waveform data is subjected to random image cropping, and the size of the cropped image is adjusted to a set pixel specification; during the training process, the image is horizontally flipped with a set probability; the image is normalized, and the pixel values ​​of the image are converted to a distribution with a mean of 0 and a standard deviation of 1.

[0076] In one embodiment, for S102, the GSM-R waveform data is input into a service quality classification and recognition model of the GSM-R system, and the service quality problem corresponding to the GSM-R system under the input waveform data condition is obtained through analysis; wherein the service quality classification and recognition model adopts a ResNet architecture, referring to Fig. 9 , which is pre-trained as follows:

[0077] S901, for a GSM-R waveform data sample set, marking the category of the service quality problem, and dividing the marked sample set to obtain a training set, a validation set, and a test set;

[0078] S902, using the training set to train the artificial intelligence model, combining the categories of the service quality issues, performing mapping learning on the artificial intelligence model from GSM-R waveform data to the categories of the service quality issues, and using the validation set and the test set to perform validation and testing, to obtain a service quality classification and recognition model; wherein the service quality classification and recognition model takes the GSM-R waveform data as input and takes the categories of the service quality issues as output.

[0079] The service quality classification and recognition model adopts the ResNet architecture of ResNet18, which includes 16 convolutional layers and 2 fully connected layers;

[0080] Among them, the convolution layer is used to extract multi-level feature representations from the input image, and the fully connected layer is used to perform classification or regression based on the extracted features;

[0081] ResNet18 is provided with residual blocks, each of which includes 2 convolutional layers and 1 skip connection, and the skip connection allows information to be directly transferred from one layer to the subsequent layer in the network.

[0082] In one embodiment, the present invention proposes an image recognition method based on deep learning for the GSM-R service quality problem. In order to enhance the generalization ability and robustness of the model, the present invention adopts a series of image preprocessing techniques. First, the input image is randomly cropped, which is an effective means of data generalization, allowing the model to learn image features of different positions and sizes. After random cropping, the cropped image is further resized to 224×224 pixels, ensuring that the model has a consistent input format when processing images of different sizes.

[0083] In order to further improve the robustness of the model, a random horizontal flipping strategy is introduced. During the training process, the image is horizontally flipped with a certain probability, which helps the model learn the direction-independent features in the image. Through this data enhancement technology, the model can show higher adaptability when dealing with various complex and changing scenes.

[0084] In the final stage of image preprocessing, the present invention normalizes the image. This step converts the pixel values ​​of the image to a distribution with a mean of 0 and a standard deviation of 1, which helps the model converge faster during training. Normalization not only improves the training efficiency of the model, but also helps the model learn a more robust feature representation. This lays the foundation for realizing automated and intelligent fault detection in railway communication systems.

[0085] ResNet18 is a deep convolutional neural network architecture. Fig.10 As shown in the figure, it is suitable for processing image recognition and other visual tasks. Its core design concept is to introduce the residual block. Through this innovative structure, the ResNet series of networks successfully solves the gradient disappearance and performance degradation problems that occur during deep network training. Each residual block consists of two convolutional layers and a skip connection. The skip connection allows information to be directly transferred from one layer to the subsequent layer in the network, thereby achieving rapid information flow.

[0086] exist Fig.10In the network, layer Name refers to the name of each layer in the network, which is usually used to identify and distinguish different layers in the network, such as "conv1", "conv2", etc. Output Size refers to the size of the feature map output by the network layer, usually including width and height. Conv refers to the convolutional layer, which is the basic unit for extracting features in deep learning. It performs weighted summation on the input data through a sliding window (convolution kernel) to generate a new feature map. Stride refers to the step size of the convolution kernel during the convolution operation. Stride determines the size of the output feature map. A larger step size will result in a smaller size of the output feature map. Max Pool refers to the maximum pooling layer, which reduces the spatial dimension of the feature map by selecting the maximum value from the local area of ​​the input feature map, which helps to extract important features and reduce the amount of calculation. Average Pool refers to the average pooling layer, which is similar to the maximum pooling, but it reduces the spatial dimension of the feature map by calculating the average value of the local area of ​​the input feature map. 1000-d FC: refers to the 1000-dimensional fully connected layer, which is used in convolutional neural networks to map the extracted features to the final output categories. For example, the number of output categories in the ImageNet classification task is 1000. Softmax is an activation function that is used to convert the raw scores output by the neural network into a probability distribution in multi-classification problems. The softmax function ensures that the sum of the probabilities of all output categories is 1. FLOPs is the number of floating point operations, which is an indicator of the computational complexity of the model, indicating the total number of floating point operations that the model needs to perform during the forward propagation process. The higher the FLOPs, the higher the computational cost of the model is usually.

[0087] ResNet18, such as Fig.11 As shown in the figure, it contains a total of 18 layers, of which 16 are convolutional layers responsible for extracting multi-level feature representations from the input image; 2 are fully connected layers used for classification or regression based on the extracted features. The residual block type of ResNet18 is BasicBlock.

[0088] exist Fig.11In the network structure of the exemplary ResNet variant, the input image of the input layer x has a size of 3×64×64, 3 channels, and the size of each channel is 64×64 pixels. Convolutional layer (Conv), there are multiple 3×3 convolutional layers, and the number of convolution kernels (C) and stride (S) are annotated in each layer. For example, the first layer is a 3×3 convolution, 64 convolution kernels, and the stride is 1. Max pooling layer (Max Pool), after the convolution layer, there is a 3×3 max pooling layer with a stride of 2, which is usually used to reduce the size of the feature map. Residual block (Residual Block), each block contains two 3×3 convolutional layers, and the number of convolution kernels is annotated in each block. The residual block is the core of the ResNet network, which allows the network to learn residual mapping, thereby solving the gradient disappearance problem in deep network training. Skip connection (Skip Connection), the solid arrow indicates the direct transfer of feature maps, which is part of the skip connection, allowing information in the network to bypass some layers directly. Global Average Pooling Layer (GAP), at the end of the network, there is a global average pooling layer, which compresses each channel of the feature map into a single value. Fully Connected Layer (FullConnection), the global average pooling layer is followed by a fully connected layer with an output size of 512, which is usually used to map features to the final classification result. Stage (Stage), different stages are distinguished by different colors, and each stage contains a certain number of residual blocks. These stages usually correspond to different depth levels of the network. Output layer y, indicates the final output category number of the network. The arrow indicates the direct transfer of feature maps, and the dotted arrow indicates that the feature maps are downsampled after the 1×1 convolution layer.

[0089] In deep neural networks, the gradient vanishing problem is a common problem. The gradient vanishing problem refers to the fact that during the back propagation process, the gradient will be multiplied by the weight matrix layer by layer, causing the gradient to shrink and eventually disappear. Skip connection can avoid the problem of gradient vanishing. Each residual block contains two convolutional layers and a skip connection. Skip connection allows information to be transmitted faster in the network. Specifically, after the input passes through the first convolutional layer and the first pooling layer, it will pass through a fully connected layer for feature mapping, and then after passing through the second convolutional layer and the second pooling layer, it will pass through a fully connected layer for feature mapping. The outputs of these two fully connected layers will be added to the original input to form two new outputs. The advantage of doing this is that it can increase the expressiveness and stability of the model. The cross-layer connection in ResNet can directly pass the input gradient to the subsequent layer, thereby avoiding the problem of gradient vanishing, which reflects the basic design idea of ​​the ResNet series of networks, that is, adding skip connections to the network to alleviate the problem of gradient vanishing in deep network training. ResNet18 can not only improve the performance of the model while maintaining the depth of the network, but also speed up the training process and improve the convergence speed of the model. like Fig.12 The figure shows a schematic diagram of the relationship of the skip connection of a specific embodiment of the present invention. The skip connection allows information in the network to be directly transmitted by bypassing some layers by directly transmitting feature maps. The residual block is a core component in the ResNet network. The residual block is designed to simplify the flow of information in the network, making it easier to train deep networks. Fig.12 In the example, x represents the input feature. Weight layer represents a layer with weights, usually a convolutional layer. In a residual block, there are usually two such layers, each followed by a relu activation function. The first relu stands for the rectified linear unit, a commonly used activation function that sets all negative values ​​to 0 and retains positive values. F(x) represents the residual function, a combination of two weighted layers and the relu activation function; the purpose of this function is to learn the residual of the input x. Identity means the same mapping, that is, passing the input x directly to the output without any transformation. F(x)+x means the output of the residual function is added to the input x; a key operation in the residual block, which allows the network to learn the difference (residual) between the input and output, rather than directly learning the output. The second relu means that after the residual is added, it is passed through a relu activation function to get the final output.

[0090] The training data of the present invention contains four folders, named dh, qhsb, qhyc, and yyzz, which represent dropped calls, switching failures, switching anomalies, and poor voice quality, respectively, and each contains 170 samples. The test data contains 75 images for each category. Based on ResNet, the intelligent classification prediction of GSM-R service quality issues is performed. The loss value (Loss) and accuracy value (Accuracy) corresponding to each Epoch of the iteration are shown in the figure. Fig.13A and Fig. 13B The highest accuracy is 84%.

[0091] In one embodiment, for S103, the acquired GSM-R waveform data and the service quality problems obtained through analysis are displayed through the railway facility monitoring platform.

[0092] For further reference, Fig.14 In order to collect and analyze disease cases and provide powerful technical support to operation and maintenance personnel, the present invention also proposes the following processing methods:

[0093] S1401, creating a problem case library, wherein the problem case library stores different types of service quality problems, and corresponding descriptions, causes, and solutions of the service quality problems;

[0094] S1402, when a service quality problem is obtained through analysis, the description, cause, and solution of the service quality problem are searched through the problem case library, and provided to the operation and maintenance personnel;

[0095] S1403: updating the problem case library according to the service quality problems output by the service quality classification and identification model, and using the problem case library as sample data for iterative training of the service quality classification and identification model to continuously learn and train the model.

[0096] Disease database case library (such as Fig.15 , 16 As an important part of the railway infrastructure detection and monitoring data platform, its content richness and accuracy are directly related to the safety and efficiency of railway operations. The introduction of the intelligent classification algorithm for GSM-R service quality issues provides rich and accurate sample case input materials, greatly enriching the sample case resources of the case library, making it more targeted and practical. These cases not only cover various service quality problems that may occur in the GSM-R system, but also include detailed descriptions of the problems, causes, solutions and other detailed information, providing valuable reference and reference for data analysts and testers.

[0097] In addition, the continuous optimization and upgrading of the intelligent classification algorithm for GSM-R service quality issues has also made the case library update faster. With the continuous emergence of new technologies and the continuous development of railway communication networks, new service quality issues are constantly emerging. The intelligent classification algorithm can track and identify these problems and incorporate them into the case library to ensure the timeliness and completeness of the case library.

[0098] In actual application scenarios, the GSM-R system mainly provides dispatching communication services such as railway traffic dispatching, realizes dispatching communication combining wired and wireless, and provides safe and reliable vehicle-to-ground information transmission services for high-speed train operation control systems. When there is a problem with the GSM-R service quality, it will affect the safety of train driving. The inspector reports the problem based on the waveform of the GSM-R service quality detection, uses the algorithm proposed in the present invention to extract the characteristics of different problem waveform images, identifies the problem image type, and establishes a sample library of GSM-R service quality problems. In view of the characteristics of the above-mentioned problem type images, the technical solution adopted by the present invention is to use a transfer learning method for the GSM-R service quality problem images to better generalize the problem image data set, increase the robustness of the model, normalize the image, and facilitate the convergence of the model training. Build a ResNet model, use the training parameters as initialization for full fine-tuning, and use the pre-training parameters as fixed feature extraction for classifier fine-tuning. At the beginning of each training iteration, the gradient of the parameters is cleared to ensure that the gradient update of each batch is independent, control whether the gradient is calculated, and perform back propagation and gradient update only in training mode. The learning rate scheduler is used to dynamically adjust the learning rate during the training process to improve the training effect. By comparing the model prediction results and the true labels, the accuracy of each epoch is calculated and stored to evaluate the performance of the model. The GSM-R service quality problem dataset is classified into four categories using a deep neural network, the problem images are marked, and a GSM-R service quality problem sample library is established.

[0099] It should be noted that, although the operations of the method of the present invention are described in a specific order in the above embodiments and the accompanying drawings, this does not require or imply that the operations must be performed in the specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0100] After introducing the method of the exemplary embodiment of the present invention, next, refer to Fig.17 A GSM-R service quality analysis and processing device based on a neural network according to an exemplary embodiment of the present invention is introduced.

[0101] The implementation of the GSM-R service quality analysis and processing device based on neural network can refer to the implementation of the above method, and the repeated parts will not be repeated. The terms "module" or "unit" used below can be a combination of software and / or hardware that implements the predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware is also possible and conceivable.

[0102] Based on the same inventive concept, the present invention also proposes a GSM-R service quality analysis and processing device based on a neural network, such as Fig.17 As shown, the device comprises:

[0103] The data acquisition module 1710 is used to acquire GSM-R waveform data;

[0104] The service quality analysis module 1720 is used to input the GSM-R waveform data into the service quality classification and recognition model of the GSM-R system, and obtain the service quality problem corresponding to the GSM-R system under the input waveform data condition through analysis; wherein, the service quality classification and recognition model adopts the ResNet architecture and is pre-trained in the following manner:

[0105] For the GSM-R waveform data sample set, the categories of service quality issues are marked, and the marked sample set is divided into a training set, a validation set, and a test set;

[0106] The artificial intelligence model is trained using the training set, and in combination with the category of the service quality problem, the artificial intelligence model is subjected to mapping learning from GSM-R waveform data to the category of the service quality problem, and the validation set and the test set are used for validation and testing to obtain a service quality classification and recognition model; wherein the service quality classification and recognition model takes the GSM-R waveform data as input and takes the category of the service quality problem as output;

[0107] The display module 1730 is used to display the acquired GSM-R waveform data and the service quality problems obtained through analysis through the railway facility monitoring platform.

[0108] In one embodiment, the categories of the service quality problems include at least: dropped calls, handover failures, handover anomalies, and poor voice quality.

[0109] In one embodiment, for the service quality problem of dropped calls, when the test interface prompts that the connection is lost and there is continuous voice quality reaching a set level, it is determined that the corresponding position is dropped; when the test interface prompts that the connection is lost and there is no voice quality problem, the line history data is played back, and if the same phenomenon occurs at the same position, it is determined that the corresponding position is dropped; if the same phenomenon does not occur, it is determined that the test system is abnormal;

[0110] For the service quality problem of switching failure, when the test interface prompts that the switching has failed, it is determined that the corresponding position has failed to switch;

[0111] For the service quality problem of abnormal handover, when no handover occurs between two base stations or multiple handovers occur, it is judged as abnormal handover;

[0112] For the service quality problem of poor voice quality, when the voice quality on the test interface is within the set level range and no call drop occurs, it is judged that the voice quality is poor.

[0113] In one embodiment, reference Fig.18 , the device further includes: a pre-processing module 1740;

[0114] Among them, the preprocessing module 1740 is used to perform image preprocessing on the GSM-R waveform data; among them, the GSM-R waveform data is randomly cropped, and the cropped image size is adjusted to the set pixel specification; during the training process, the image is horizontally flipped with a set probability; the image is normalized, and the pixel value of the image is converted to a distribution with a mean of 0 and a standard deviation of 1.

[0115] In one embodiment, the service quality classification and identification model adopts the ResNet architecture of ResNet18, which includes 16 convolutional layers and 2 fully connected layers;

[0116] Among them, the convolution layer is used to extract multi-level feature representations from the input image, and the fully connected layer is used to perform classification or regression based on the extracted features;

[0117] ResNet18 is provided with residual blocks, each of which includes 2 convolutional layers and 1 skip connection, and the skip connection allows information to be directly transferred from one layer to the subsequent layer in the network.

[0118] In one embodiment, reference Fig.18 , the device further includes: a case library module 1750;

[0119] The case library module 1750 is used to create a problem case library, which stores different types of service quality problems, as well as descriptions, causes, and solutions of the service quality problems;

[0120] When service quality problems are found through analysis, the description, cause and solution of the service quality problems are queried through the disease case library and provided to the operation and maintenance personnel;

[0121] The service quality analysis module 1720 is also used to update the problem case library according to the service quality problems output by the service quality classification and identification model, and use it as sample data for iterative training of the service quality classification and identification model to continuously learn and train the model.

[0122] It should be noted that although several modules of the GSM-R service quality analysis and processing device based on neural network are mentioned in the above detailed description, this division is only exemplary and not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more modules described above can be concretized in one module. Conversely, the features and functions of one module described above can be further divided into multiple modules for concretization.

[0123] Based on the above invention concept, Fig.19 As shown, the present invention also proposes a computer device 1900, including a memory 1910, a processor 1920, and a computer program 1930 stored in the memory 1910 and executable on the processor 1920, wherein the processor 1920 implements the aforementioned neural network-based GSM-R service quality analysis and processing method when executing the computer program 1930.

[0124] Based on the aforementioned inventive concept, the present invention proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the aforementioned neural network-based GSM-R service quality analysis and processing method is implemented.

[0125] Based on the aforementioned inventive concept, the present invention proposes a computer program product, which includes a computer program. When the computer program is executed by a processor, a GSM-R service quality analysis and processing method based on a neural network is implemented.

[0126] The neural network-based GSM-R service quality analysis and processing method and device proposed in the present invention analyzes a large number of GSM-R service quality problem pictures, sorts out the abnormal patterns of GSM-R service quality problem pictures, determines the characteristic laws of its abnormalities, and constructs the optimal algorithm flow for this characteristic law. The relevant parameters and thresholds in the present invention are determined based on the results of a large number of experiments with the best effects. Compared with the prior art, it has at least the following technical effects: based on the data set generalization method of transfer learning, it extracts the characteristics of GSM-R service quality problem pictures and enriches the multi-source heterogeneous data samples of data analysts; through the GSM-R image recognition algorithm based on the ResNet network, it assists detection personnel to realize automatic GSM-R service quality problem identification and improves detection efficiency. By enriching the GSM-R service quality problem sample library, it can provide valuable experimental materials for data analysts and developers and provide strong technical support for the identification of multi-category communication problems.

[0127] The acquisition, storage, use, and processing of data in the technical solution of this application comply with the relevant provisions of laws and regulations.

[0128] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, devices, 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.

[0129] The present invention is described with reference to flowcharts and / or block diagrams of methods 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 generate 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.

[0130] 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 1A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0131] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions 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.

[0132] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A GSM-R service quality analysis and processing method based on neural network, characterized in that: include: Get GSM-R waveform data; The GSM-R waveform data is input into the service quality classification and recognition model of the GSM-R system, and the service quality problem corresponding to the GSM-R system under the input waveform data condition is obtained through analysis; wherein the service quality classification and recognition model adopts a ResNet architecture and is pre-trained in the following manner: For the GSM-R waveform data sample set, the categories of service quality issues are marked, and the marked sample set is divided into a training set, a validation set, and a test set; The artificial intelligence model is trained using the training set, and in combination with the category of the service quality problem, the artificial intelligence model is subjected to mapping learning from GSM-R waveform data to the category of the service quality problem, and the validation set and the test set are used for validation and testing to obtain a service quality classification and recognition model; wherein the service quality classification and recognition model takes the GSM-R waveform data as input and takes the category of the service quality problem as output; The railway facility monitoring platform displays the acquired GSM-R waveform data and the service quality issues obtained through analysis.

2. The GSM-R service quality analysis and processing method according to claim 1, characterized in that: The categories of the service quality problems include at least: dropped calls, switching failures, switching anomalies, and poor voice quality.

3. The GSM-R service quality analysis and processing method according to claim 2, characterized in that: For the service quality problem of dropped calls, when the test interface prompts that the connection is lost and there is continuous voice quality reaching the set level, the corresponding location is determined to be dropped; When the test interface prompts that the connection is lost and there is no poor voice quality, replay the line history data. If the same phenomenon occurs at the same location, it is determined that the corresponding location has dropped calls; If the same phenomenon does not occur, it is determined that the test system is abnormal; For the service quality problem of switching failure, when the test interface prompts that the switching has failed, it is determined that the corresponding position has failed to switch; For the service quality problem of abnormal handover, when no handover occurs between two base stations or multiple handovers occur, it is judged as abnormal handover; For the service quality problem of poor voice quality, when the voice quality on the test interface is within the set level range and no call drop occurs, it is judged that the voice quality is poor.

4. The GSM-R service quality analysis and processing method according to claim 1, characterized in that: The method further includes: The GSM-R waveform data is subjected to image preprocessing; the GSM-R waveform data is subjected to random image cropping, and the cropped image size is adjusted to a set pixel specification; during the training process, the image is horizontally flipped with a set probability; the image is normalized, and the pixel value of the image is converted to a distribution with a mean of 0 and a standard deviation of 1.

5. The GSM-R service quality analysis and processing method according to claim 1, characterized in that: The service quality classification and recognition model adopts the ResNet architecture of ResNet18, which includes 16 convolutional layers and 2 fully connected layers; Among them, the convolution layer is used to extract multi-level feature representations from the input image, and the fully connected layer is used to perform classification or regression based on the extracted features; ResNet18 is provided with residual blocks, each of which includes 2 convolutional layers and 1 skip connection, and the skip connection allows information to be directly transferred from one layer to the subsequent layer in the network.

6. The GSM-R service quality analysis and processing method according to claim 1, characterized in that: The method further includes: Creating a problem case library, wherein the problem case library stores different types of service quality problems, and corresponding descriptions, causes, and solutions of the service quality problems; When service quality problems are found through analysis, the description, cause and solution of the service quality problems are queried through the disease case library and provided to the operation and maintenance personnel; The problem case library is updated according to the service quality problems output by the service quality classification and identification model, and is used as sample data for iterative training of the service quality classification and identification model to perform continuous learning and training on the model.

7. A GSM-R service quality analysis and processing device based on neural network, characterized in that: include: A data acquisition module, used to acquire GSM-R waveform data; The service quality analysis module is used to input the GSM-R waveform data into the service quality classification and recognition model of the GSM-R system, and obtain the service quality problem corresponding to the GSM-R system under the input waveform data condition through analysis; wherein, the service quality classification and recognition model adopts the ResNet architecture and is pre-trained in the following manner: For the GSM-R waveform data sample set, the categories of service quality issues are marked, and the marked sample set is divided into a training set, a validation set, and a test set; The artificial intelligence model is trained using the training set, and in combination with the category of the service quality problem, the artificial intelligence model is subjected to mapping learning from GSM-R waveform data to the category of the service quality problem, and the validation set and the test set are used for validation and testing to obtain a service quality classification and recognition model; wherein the service quality classification and recognition model takes the GSM-R waveform data as input and takes the category of the service quality problem as output; The display module is used to display the GSM-R waveform data obtained through the railway facility monitoring platform and the service quality problems obtained through analysis.

8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images