Railway turnout fault diagnosis method based on power data
Through deep learning model based on power data, the problem of accurate and low efficiency of railway switch fault diagnosis in the existing technology is solved, and high-precision fault type identification and prediction is achieved to ensure the safety and stability of railway transportation.
Patent Information
- Application Number
- CN202510635021.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-16
- Publication Date
- 2025-08-15
AI Technical Summary
In the prior art, railway switch fault diagnosis relies on manual inspection and microcomputer monitoring, and there are problems of poor accuracy and low efficiency, especially in complex structural switches, it is difficult to meet the diagnostic needs of high accuracy.
The railway switch fault diagnosis method based on power data is adopted. By obtaining the power data of the train passing through the railway switch and the displacement of the pointed rail and the basic rail, the depth cavity convolution length short-term memory network attention model is input after preprocessing, and the feature vector is corrected with the temperature data, the features are extracted and classified to achieve the diagnosis of fault types.
It improves the accuracy of the diagnosis of railway switch fault types, provides accurate prediction results, helps operation and maintenance personnel to plan maintenance strategies in advance, avoid equipment failures, and ensures the safe and stable operation of railway transportation.
Smart Images
Figure CN120492808A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of rail transportation technology, and in particular to a railway turnout fault diagnosis method based on power data. Background Art
[0002] Railway transportation plays a critical role in the modern transportation system. As a core component of railway tracks, the operational status of turnouts directly impacts transportation safety and efficiency. Traditional turnout fault diagnosis relies on manual inspections and regular maintenance, which suffers from low efficiency and poor accuracy. Therefore, timely detection and accurate diagnosis of turnout faults, along with early warning, are crucial for preventing major accidents and improving equipment utilization.
[0003] The current method commonly used to inspect turnouts is a combination of manual scheduled inspections and feedback from a computer-based monitoring system. Railway maintenance personnel observe and record the turnout operating current data collected by the computer-based monitoring system at fixed time points, compare this data with the data monitored during normal turnout switching, and identify turnout faults based on the similarity of the curves.
[0004] However, this approach requires staff to have extremely rich railway maintenance experience, and the various parameter indicators in the monitoring system are set by subject experts. The ambiguity of their experience also makes it impossible to maintain most standards for a long time. Therefore, the accuracy of fault diagnosis cannot be guaranteed and it is difficult to be effectively promoted. In addition, the complex structure of turnouts commonly used in railway speed-up areas has more stringent requirements on the accuracy of fault diagnosis, which also increases the difficulty of the daily work of railway staff.
[0005] Therefore, in order to improve the fault diagnosis accuracy of the turnout system, it is necessary to change the existing turnout fault diagnosis mode as soon as possible. Summary of the Invention
[0006] Based on this, it is necessary to provide a railway turnout fault diagnosis method based on power data to address the above technical problems.
[0007] The present invention adopts the following technical solutions: The present invention provides a railway turnout fault diagnosis method based on power data, comprising: Obtaining power data and the displacement between the point rail and the stock rail when the train passes through the railway switch and performing preprocessing to obtain standardized feature vector data; Obtaining temperature data of a time node corresponding to the standardized feature vector data; and correcting the standardized feature vector data according to the temperature data; Inputting the corrected normalized feature vector data into a trained deep dilated convolutional long short-term memory network attention model; the deep dilated convolutional long short-term memory network attention model includes a dilated convolution layer, a maximum pooling layer, a long short-term memory network layer, a fully connected layer, and an output layer connected in series; In the dilated convolution layer, the convolution kernel size is determined based on the characteristics of the corrected and standardized feature vector data to extract the power variation features at different time scales in the standardized feature vector. In the maximum pooling layer, the power variation features are downsampled and redundant features in the downsampled features are removed to obtain the key features. In the long short-term memory network layer, the dependencies of the key features in the time dimension are captured to obtain the operating status of the railway turnout. In the fully connected layer, according to the operating status of the railway turnout, the feature representation of the current state of the railway turnout equipment is determined through the activation function; In the output layer, according to the preset fault type, the feature representation of the current state of the railway turnout equipment is classified by the classifier to obtain the railway turnout fault type.
[0008] Preferably, obtaining the standardized feature vector data specifically includes: Detecting abnormal values in power data and displacement respectively, and deleting abnormal values in power data and displacement; Low-pass filtering was used to remove high-frequency noise interference in the power data, and linear interpolation was used to fill in the missing power data based on the power data of adjacent time points; The padded power data is normalized, and the spectrum characteristics of the power data are extracted using short-time Fourier transform, and the displacement is derived to obtain the displacement change characteristics; The frequency spectrum feature and the displacement change feature are spliced to obtain standardized feature vector time series data.
[0009] Preferably, the standardized feature vector data is corrected according to the temperature data, specifically comprising: Substituting the temperature data into the normalized characteristic vector data into a temperature compensation mathematical model to obtain corrected normalized characteristic vector data; The temperature compensation mathematical model is: ; Where, After correction Normalized feature vector data at time, for Normalized feature vector data at time, is the temperature coefficient, for The temperature data corresponding to the normalized eigenvector data at each moment, is the nominal temperature of railway turnout equipment.
[0010] Preferably, the convolution kernel size is determined according to the characteristics of the normalized feature vector data, specifically including: If the time scale of the corrected normalized feature vector data is per minute, the convolution kernel size is determined to be 3*3; If the time scale of the corrected normalized feature vector data is hourly, the convolution kernel size is determined to be 5*5; If the time scale of the corrected normalized feature vector data is every 4 hours or every day, the convolution kernel size is determined to be 9*9.
[0011] Preferably, the method further comprises: In the output layer, based on the feature representation of the current state of the railway turnout equipment, the failure time node and remaining life of the railway turnout equipment at a future moment are predicted.
[0012] Preferably, the prediction of the time point of potential failure and the remaining life of the railway turnout equipment specifically includes: Based on the safety standards of railway turnout equipment performance indicators, a linear regression model is used to obtain the change trend of equipment performance indicators over time; According to the changing trend of equipment performance indicators over time, the failure time node and remaining life of railway turnout equipment at future moments are obtained.
[0013] Preferably, classifying the characteristic representation of the current state of the railway turnout equipment by a classifier to determine the type of railway turnout fault specifically includes: The feature representation of the current state of the railway turnout equipment is classified by a classifier to obtain the failure probability of the railway turnout equipment under each failure type; The fault type with a fault probability greater than the corresponding fault threshold is determined as the railway turnout fault type.
[0014] The present invention also provides a railway turnout fault diagnosis device based on power data, which is characterized by comprising: The data acquisition module is used to obtain the power data and the displacement between the rail and the stock rail when the train passes through the railway switch and perform preprocessing to obtain standardized feature vector data; A temperature compensation module is used to correct the standardized feature vector data according to the temperature data of the time node corresponding to the standardized feature vector data to obtain the final standardized feature vector data; The fault prediction module inputs the final standardized feature vector data into the trained deep atrous convolutional long short-term memory network attention model to obtain the fault type diagnosis results at the future moment, the time node of potential failure, and the remaining life of the railway turnout equipment; The operation and maintenance decision module is used to help staff generate operation and maintenance decision recommendations based on the fault type diagnosis results, the time point of potential faults, the remaining life of railway turnout equipment, the historical operation and maintenance records of railway turnouts, and the current operation resource allocation; The user interaction and visualization module is used to present the equipment operating status, fault type diagnosis results, potential failure time nodes, remaining life of railway turnout equipment and operation and maintenance decision recommendations in real time.
[0015] The present invention also provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the computer program implements the above-mentioned railway turnout fault diagnosis method based on power data.
[0016] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the above-mentioned railway turnout fault diagnosis method based on power data is implemented.
[0017] At least one of the above technical solutions adopted by the present invention can achieve the following beneficial effects: In a railway turnout fault diagnosis method based on power data provided by the present invention, based on the standardized feature vector data obtained after preprocessing the power data and the displacement of the point rail and the base rail and the temperature data of the corresponding time node, the standardized feature vector data is corrected using a temperature compensation model through the dependency relationship between the temperature data and the standardized feature vector data, thereby avoiding the impact of temperature changes on the data and the fault diagnosis results; through the constructed deep void convolution long short-term memory network attention model, the complex patterns and rules of the railway turnout power data and displacement data are mined to obtain the characteristics of the current state of the railway turnout, and the feature representation is classified according to the preset fault type, thereby effectively improving the accuracy of railway turnout fault type diagnosis, and the accurate diagnosis and prediction results lay a solid scientific foundation for operation and maintenance decision-making, so that operation and maintenance personnel can plan maintenance strategies in advance, effectively avoid sudden equipment failures, and stabilize the safe and stable operation of railway transportation. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0019] Figure 1 A schematic flow chart of a railway turnout fault diagnosis method based on power data provided by the present invention; Figure 2A schematic diagram of the data acquisition and preprocessing process of a railway turnout fault diagnosis method based on power data provided by the present invention; Figure 3 A detailed diagram of the deep learning model architecture of a railway turnout fault diagnosis method based on power data provided by the present invention; Figure 4 The basic structure of a deep learning neural network for a railway turnout fault diagnosis method based on power data provided by the present invention; Figure 5 A schematic diagram of a fault diagnosis and prediction process of a railway turnout fault diagnosis method based on power data provided by the present invention; Figure 6 A schematic diagram of a railway turnout fault diagnosis device based on power data provided by the present invention; Figure 7 A schematic diagram of a computer device for implementing a railway turnout fault diagnosis method based on power data provided by the present invention. DETAILED DESCRIPTION
[0020] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments of the present invention and corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in the specification, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.
[0021] The following describes in detail the technical solutions provided by various embodiments of the present application in conjunction with the accompanying drawings.
[0022] Figure 1 The figure is a flow chart of a railway turnout fault diagnosis method based on power data in the present invention, which specifically includes the following steps: S101: Obtain power data and displacement of the point rail and stock rail when a train passes through a railway switch and perform preprocessing to obtain standardized feature vector data.
[0023] See also Figure 2, which is a schematic diagram of the data acquisition and preprocessing process. Sensors are installed on the turnout mechanical components to collect the displacement of the point rail and the stock rail and the power data of the turnout machine; the values of the displacement and power data are detected. If the variance of the value with the mean exceeds 3 times the standard deviation, the data is judged as an outlier and eliminated; low-pass filtering is used to remove high-frequency noise interference in the multi-source data, and linear interpolation is used to fill the missing data based on the data of adjacent time points; the multi-source data after missing value filling are normalized, and the short-time Fourier transform is used to extract the spectral characteristics of the multi-source data, and the displacement is derived to obtain the velocity information and acceleration information of the displacement, that is, the displacement change characteristics; the spectral characteristics and the displacement change characteristics are spliced to obtain standardized feature vector time series data.
[0024] Specifically, the speed, acceleration and braking force data of the train when passing through the railway switch are collected at the train operation control system (TCMS), with a data collection frequency of not less than once per second; sensors are installed on the key mechanical components of the switch to collect the displacement and power data of the point rail and the base rail. The power data includes: the degree of contact, the operating current of the switch machine, the operating time and the locking force data. The measurement accuracy of the displacement reaches the millimeter level, and the measurement accuracy of the operating current reaches ±0.1A; environmental monitoring equipment is installed at the location of the railway switch to collect environmental data at a frequency of not less than once every five minutes. The environmental data includes temperature, humidity, wind speed and rainfall; railway-specific industrial Ethernet or 4G / 5G is used. The wireless communication network transmits the collected multi-source data to the data storage and processing center, and adopts data encryption and verification mechanisms during the data transmission process to ensure the integrity and security of the data; the data obtained from the multi-source data collection method is processed for outliers. For the switch machine operating current data, when the deviation of the current value at a certain moment from the mean of the data sequence exceeds three times the standard deviation, the current value is marked as an outlier and removed from the data sequence; for missing data in the data sequence, linear interpolation or mean filling method is used to supplement it.
[0025] Specifically, the collected data is normalized according to the different characteristics of its physical quantities. For data with physical dimensions, such as operating current and displacement, the maximum and minimum normalization formula is used. , maps the data to the interval [0, 1], where x is the original data, is the minimum value in the data sequence, is the maximum value in the data sequence; for data with a normal distribution tendency, such as temperature data, the mean variance normalization formula is used Transform the data into a standard normal distribution with mean 0 and variance 1, where is the mean of the data and is the standard deviation of the data.
[0026] Specifically, combining railway turnout expertise with deep learning feature extraction technology, spectral features are extracted from the action current data through Fourier transform to detect anomalies in harmonic components; derivative operations are performed on the displacement data to obtain displacement velocity and acceleration information, and the extracted multiple features are combined into a feature vector as input data for the subsequent deep learning model.
[0027] S102: Acquire temperature data of a time node corresponding to the standardized feature vector data; and modify the standardized feature vector data according to the temperature data.
[0028] Optionally, the temperature data is substituted into the normalized characteristic vector data into a temperature compensation mathematical model to obtain corrected normalized characteristic vector data; The temperature compensation mathematical model is: ; Where, After correction Normalized feature vector data at time, for Normalized feature vector data at time, is the temperature coefficient, for The temperature data corresponding to the normalized eigenvector data at each moment, is the nominal temperature of railway turnout equipment.
[0029] S103: Input the corrected normalized feature vector data into the trained deep dilated convolutional long short-term memory network attention model; the deep dilated convolutional long short-term memory network attention model includes a dilated convolution layer, a maximum pooling layer, a long short-term memory network layer, a fully connected layer and an output layer connected in series.
[0030] Optionally, in the dilated convolution layer, the convolution kernel size is determined according to the characteristics of the corrected standardized feature vector data to extract the power change features at different time scales in the standardized feature vector; in the maximum pooling layer, the power change features are downsampled, and the redundant features in the downsampled features are removed to obtain the key features; in the long short-term memory network layer, the dependency of the key features in the time dimension is captured to obtain the operating status of the railway turnout; in the fully connected layer, according to the operating status of the railway turnout, the feature representation of the current state of the railway turnout equipment is determined through the activation function; in the output layer, according to the preset fault type, the feature representation of the current state of the railway turnout equipment is classified by the classifier to determine the railway turnout fault type.
[0031] Specifically, see Figure 3, which is a detailed diagram of the deep learning model architecture, constructs a deep neural network, which includes a convolutional neural network (CNN) part and a long short-term memory network (LSTM) part. The CNN part contains multiple convolutional layers, and its first convolutional layer uses 32 3×3 convolution kernels with a step size of 1, adopts the ReLU activation function, and is followed by a 2×2 maximum pooling layer to extract the spatial features of the data; the LSTM part sets 128 hidden units according to the length and characteristics of the time series data to capture the dependency of the switch operation status in the time dimension.
[0032] Specifically, the feature vectors obtained by the data preprocessing method are divided into training set, validation set and test set with a ratio of 7:2:1. The Adam optimizer is used to train the model with an initial learning rate of 0.001. L2 regularization technology is used during training to prevent overfitting, and the early stopping method is used. When the loss of the validation set no longer decreases in multiple consecutive training cycles, the training is stopped and the best performing model is saved. The performance of the trained model is evaluated based on the test set, and indicators such as accuracy, recall rate and F1 score are calculated. The hyperparameters of the model are adjusted according to the evaluation results. The hyperparameters include but are not limited to the number of convolution kernels, the number of hidden units in the LSTM part, and the learning rate to optimize the performance and generalization ability of the model.
[0033] Specifically, see Figure 4 The network structure diagram of the deep dilated convolutional long short-term memory network attention model is shown below. The dilated convolution layer parameters are set, such as a 3×3 kernel size, 64 kernels, a stride of 1, and a dilation rate of 2. The ReLU activation function is used, and a max pooling layer is added for downsampling. When processing power data, the dilated convolution layer effectively extracts power variation characteristics at different time scales. The LSTM layer has 128 memory cells and 2 layers, determined based on the length and complexity of the time series. Gating parameters are also set. The dataset is split into training, validation, and test sets in a 7:2:1 ratio. The model is trained on the training set using the Adam optimization algorithm with an initial learning rate of 0.001. An attention mechanism is introduced during training, and L2 regularization is used to prevent overfitting. The validation set is used for regular evaluation, and hyperparameters are adjusted based on metrics such as accuracy and loss. Training is terminated when performance on the validation set no longer improves.
[0034] Optionally, the convolution kernel size is determined based on the characteristics of the normalized feature vector data, specifically including: If the time scale of the corrected standardized feature vector data is every minute, the convolution kernel size is determined to be 3*3; if the time scale of the corrected standardized feature vector data is every hour, the convolution kernel size is determined to be 5*5; if the time scale of the corrected standardized feature vector data is every 4 hours or every day, the convolution kernel size is determined to be 9*9.
[0035] Optionally, the method further includes: at the output layer, predicting the failure time node and the remaining life of the railway turnout equipment at a future moment based on the characteristic representation of the current state of the railway turnout equipment.
[0036] Optionally, predicting the time node of potential failure and the remaining life of the railway turnout equipment specifically includes: based on the safety standards of the railway turnout equipment performance indicators, using a linear regression model to obtain the change trend of the equipment performance indicators over time; according to the change trend of the equipment performance indicators over time, obtaining the failure time node and remaining life of the railway turnout equipment at a future moment.
[0037] Optionally, the characteristic representation of the current state of the railway turnout equipment is classified by a classifier to determine the railway turnout fault type, specifically including: classifying the characteristic representation of the current state of the railway turnout equipment by a classifier to obtain the failure probability of the railway turnout equipment under each fault type; and determining the fault type with a failure probability greater than the corresponding fault threshold as the railway turnout fault type.
[0038] Specifically, a linear regression model is used to predict the time node of potential failure and the remaining life of railway turnout equipment, including: using a linear regression model to predict the changing trend of performance indicators of railway equipment over time, combined with a preset threshold of the wear limit value of the train wheelset, to predict the time node of potential failure and the remaining life of railway turnout equipment in advance.
[0039] Specifically, see Figure 5 , which is a schematic diagram of the fault diagnosis and prediction process. The trained deep learning model obtained by the deep learning model construction and training method is output through the softmax classifier to classify the railway turnout faults according to pre-set fault categories. The pre-set fault categories include point rail jamming, switch machine failure, poor adhesion, etc. When the probability of a certain fault category output by the classifier exceeds 0.8, it is determined that the railway turnout has this fault category; using the regression model, based on historical data and current railway turnout status data, the remaining service life of the key components of the turnout or the time when the next failure may occur is predicted, and fault warning information is issued in advance based on the pre-set threshold. The threshold is set according to the equipment specifications, historical operation and maintenance data and safety standards of the railway turnout.
[0040] Specifically, the preprocessed data is fed into a trained model. The DLeNN-Attention model uses a support vector machine to classify the status and output a fault probability distribution. For example, if the model outputs a probability of a certain fault type exceeding 0.8, the turnout is considered to have experienced that type of fault. For potential faults, linear regression is used to predict the trend of key performance indicators. Pre-set thresholds (e.g., a wheelset wear limit of 5mm) are used to warn of potential faults. Data is then accumulated for model optimization.
[0041] Specifically, the model's performance is evaluated using a test set, with metrics such as accuracy, precision, and recall calculated. For example, when evaluating 1,000 test samples, the accuracy reached 98%, the precision reached 97%, and the recall reached 96%. If model performance does not meet expectations, model hyperparameters are adjusted, such as reducing the learning rate of the DLeNN-Attention model to 5e-4 and increasing the number of iterations to 1,200. The model structure is improved, such as adding a convolutional layer to the feature extractor of the DANN transfer learning model and retraining. New data is collected regularly (e.g., monthly). When the amount of new data reaches 500 or more, incremental learning techniques are used to fine-tune the model online. For example, a gradient-based incremental learning algorithm is used to mix new data with historical data in a certain ratio, gradually updating model parameters to adapt the model to changes in equipment operating conditions, such as performance degradation caused by equipment aging or the impact of increased ambient humidity on turnout operation, thereby continuously improving fault prediction and diagnosis capabilities.
[0042] The above is a method for diagnosing railway turnout faults based on power data provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding device for diagnosing railway turnout faults based on power data, such as Figure 6 shown.
[0043] Figure 6 A schematic diagram of a device for diagnosing railway turnout faults based on power data provided by the present invention includes: The data acquisition module 601 is used to obtain power data and the displacement between the rail and the stock rail when the train passes through the railway switch and perform preprocessing to obtain standardized feature vector data; The temperature compensation module 602 is used to correct the standardized feature vector data according to the temperature data of the time node corresponding to the standardized feature vector data to obtain the final standardized feature vector data; The fault prediction module 603 inputs the final normalized feature vector data into the trained deep atrous convolutional long short-term memory network attention model to obtain the fault type diagnosis results at the future moment, the time node of the potential fault, and the remaining life of the railway turnout equipment; The operation and maintenance decision module 604 is used to help staff generate operation and maintenance decision suggestions based on the fault type diagnosis results, the time point of potential faults, the remaining life of the railway turnout equipment, the historical operation and maintenance records of the railway turnout, and the current operation resource allocation; The user interaction and visualization module 605 is used to present the equipment operation status, fault type diagnosis results, potential fault time nodes, remaining life of railway turnout equipment and operation and maintenance decision suggestions in real time.
[0044] Specifically, the user interaction and visualization module presents equipment operating status, fault diagnosis results, prediction information, and maintenance decision recommendations to operators in an intuitive and easy-to-understand visual format. Customized dashboards display real-time values and trend curves of key operating parameters (such as train speed and track geometry). Graphs (bar charts show the frequency of different fault types, line charts display historical trends of equipment performance indicators, and heat maps display fault distribution probability) and 3D models (simulating the structure and real-time operating conditions of key train components, the force distribution of train bogies, and the status of track fasteners) enable operators to quickly and comprehensively understand the overall picture of rail transit equipment. Through the interactive interface, operators can view detailed data and historical records, confirm or adjust maintenance decisions, and effectively transmit information, improving maintenance response speed and operational convenience.
[0045] Regarding the specific definition of a railway turnout fault diagnosis device based on power data, please refer to the definition of a railway turnout fault diagnosis method based on power data above, which will not be repeated here. The various modules in the above-mentioned railway turnout fault diagnosis device based on power data can be implemented in whole or in part by software, hardware, and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
[0046] The present invention also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 1 A railway turnout fault diagnosis method based on power data is provided.
[0047] The present invention also provides Figure 7 The structural diagram of the computer equipment shown in FIG. Figure 7 As mentioned above, at the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 XX method provided.
[0048] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described method embodiments. Any reference to memory, storage, database, or other media used in the embodiments provided herein may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).
[0049] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present invention.
[0050] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present invention.
Claims
1. A railway turnout fault diagnosis method based on power data, characterized in that: include: Obtaining power data and the displacement between the point rail and the stock rail when the train passes through the railway switch and performing preprocessing to obtain standardized feature vector data; Obtaining temperature data of a time node corresponding to the standardized feature vector data; and correcting the standardized feature vector data according to the temperature data; Inputting the corrected normalized feature vector data into a trained deep dilated convolutional long short-term memory network attention model; the deep dilated convolutional long short-term memory network attention model includes a dilated convolution layer, a maximum pooling layer, a long short-term memory network layer, a fully connected layer, and an output layer connected in series; In the dilated convolution layer, the convolution kernel size is determined based on the characteristics of the corrected and standardized feature vector data to extract the power variation features at different time scales in the standardized feature vector. In the maximum pooling layer, the power variation features are downsampled and redundant features in the downsampled features are removed to obtain the key features. In the long short-term memory network layer, the dependencies of the key features in the time dimension are captured to obtain the operating status of the railway turnout. In the fully connected layer, according to the operating status of the railway turnout, the feature representation of the current state of the railway turnout equipment is determined through the activation function; In the output layer, according to the preset fault type, the feature representation of the current state of the railway turnout equipment is classified by the classifier to determine the railway turnout fault type.
2. A railway turnout fault diagnosis method based on power data according to claim 1, characterized in that: The obtaining of the standardized feature vector data specifically includes: Detecting abnormal values in power data and displacement respectively, and deleting abnormal values in power data and displacement; Low-pass filtering was used to remove high-frequency noise interference in the power data, and linear interpolation was used to fill in the missing power data based on the power data of adjacent time points; The padded power data is normalized, and the spectrum characteristics of the power data are extracted using short-time Fourier transform, and the displacement is derived to obtain the displacement change characteristics; The frequency spectrum feature and the displacement change feature are spliced to obtain standardized feature vector time series data.
3. The railway turnout fault diagnosis method based on power data according to claim 1, characterized in that: The correcting of the standardized feature vector data according to the temperature data specifically includes: Substituting the temperature data into the normalized characteristic vector data into a temperature compensation mathematical model to obtain corrected normalized characteristic vector data; The temperature compensation mathematical model is: ； Where, After correction Normalized feature vector data at time, for Normalized feature vector data at time, is the temperature coefficient, for The temperature data corresponding to the normalized eigenvector data at each moment, is the nominal temperature of railway turnout equipment.
4. The railway turnout fault diagnosis method based on power data according to claim 1, characterized in that: Determining the convolution kernel size based on the characteristics of the standardized feature vector data specifically includes: If the time scale of the corrected normalized feature vector data is per minute, the convolution kernel size is determined to be 3*3; If the time scale of the corrected normalized feature vector data is hourly, the convolution kernel size is determined to be 5*5; If the time scale of the corrected normalized feature vector data is every 4 hours or every day, the convolution kernel size is determined to be 9*9.
5. The railway turnout fault diagnosis method based on power data according to claim 1, characterized in that: The method further comprises: In the output layer, based on the feature representation of the current state of the railway turnout equipment, the failure time node and remaining life of the railway turnout equipment at a future moment are predicted.
6. A railway turnout fault diagnosis method based on power data according to claim 5, characterized in that: The predicted time point of potential failure and the remaining life of railway turnout equipment specifically include: Based on the safety standards of railway turnout equipment performance indicators, a linear regression model is used to obtain the change trend of equipment performance indicators over time; According to the changing trend of equipment performance indicators over time, the failure time node and remaining life of railway turnout equipment at future moments are obtained.
7. The railway turnout fault diagnosis method based on power data according to claim 1, characterized in that: The classifying the characteristic representation of the current state of the railway turnout equipment by the classifier to determine the railway turnout fault type specifically includes: The feature representation of the current state of the railway turnout equipment is classified by a classifier to obtain the failure probability of the railway turnout equipment under each failure type; The fault type with a fault probability greater than the corresponding fault threshold is determined as the railway turnout fault type.
Citation Information
Cited By
Heavy fork electrical system fault diagnosis and intelligent repair method
CN120952759A