Short-time rapid fault detection method for overhead line system based on event-driven neural network
By applying event-driven neural networks in contact network fault detection, the rapid identification and positioning of short-term and rapid faults of contact networks is achieved, the problem of inefficiency of traditional detection methods is solved, and the response speed and accuracy of detection are improved.
Patent Information
- Application Number
- CN202411591328.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-08
- Publication Date
- 2025-02-28
AI Technical Summary
Traditional contact network fault detection methods are inefficient and difficult to deal with short-term and rapid failures of contact networks in real time, affecting the safety of train operations.
The short-term rapid fault detection method of contact network based on event-driven neural network (EDNN) is adopted to collect contact network data in real time, and use event detection modules and fault classification modules to achieve rapid fault identification and location.
It significantly shortens the response time of fault detection, improves the response speed and accuracy of detection, reduces system resource consumption, and enhances the safety and stability of the contact network system.
Smart Images
Figure CN119538146A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of high-speed railway contact network, and in particular relates to a short-term rapid fault detection method for high-speed railway contact network based on deep learning. Background Art
[0002] The overhead contact system is an important component of electrified railways and urban rail transit, and is mainly responsible for providing continuous power supply for trains. With the continuous increase in train speed and frequency, the demand for fault detection and maintenance of the overhead contact network is also increasing. However, traditional fault detection methods mainly rely on manual inspections and regular maintenance, which is not only inefficient, but also has great deficiencies in real-time and accuracy. In addition, the faults of the overhead contact network are sudden and short-term. Traditional detection methods are difficult to deal with such rapid and unpredictable faults, which can easily cause detection delays and affect the safety of train operation.
[0003] In recent years, with the development of artificial intelligence and machine learning technologies, intelligent fault detection methods have gradually become a research hotspot. In particular, event-driven neural networks (EDNNs) have become an ideal choice for solving short-term and rapid fault detection in contact networks due to their high efficiency and accuracy in handling emergencies. EDNNs can quickly detect abnormal events based on the real-time operating data of the contact network system by monitoring changes in parameters such as current, voltage, and tension, and immediately start the neural network to judge and classify faults after detecting the event. This method not only shortens the reaction time of fault detection, but also enables real-time alarm and rapid response, greatly improving the safety and stability of the contact network.
[0004] Unlike traditional neural networks, event-driven neural networks use an asynchronous activation mechanism, which triggers network operations only when abnormal events are detected, greatly reducing the computational burden of the system. This gives EDNN a significant advantage in dealing with scenarios such as contact networks that require high timeliness and large amounts of data. Through this technology, the fault point can be located at the first time, reducing the impact of the fault on the railway system and providing timely maintenance suggestions to maintenance personnel. Summary of the invention
[0005] The present invention proposes a method for detecting short-time and rapid faults of a contact network based on an event-driven neural network, aiming to improve the response speed and accuracy of fault detection of the contact network.
[0006] As an important part of the railway system, the fault of the contact network directly affects the safe operation of the train. Therefore, the present invention realizes real-time monitoring and rapid fault identification of the contact network system by introducing the event-driven neural network (EDNN) technology. The method first collects the real-time operation data of the contact network through sensors, including key parameters such as current, voltage, displacement, tension, etc., and pre-processes these data, such as denoising and filtering operations, to ensure data quality. Then, the event-driven neural network triggers the activation of the neural network by detecting abnormal events in the contact network, and uses the neural network to classify and locate faults. The event detection module monitors the key parameters of the contact network according to the set threshold. When the fluctuation of a certain parameter exceeds the threshold, the system will immediately trigger fault detection, which greatly improves the response speed of the detection. The fault classification module classifies the fault type through a convolutional neural network and outputs the corresponding detection results and alarm prompts. This method not only has a fast response speed, but also can dynamically adjust the sensitivity of fault detection according to real-time data, and has high flexibility and adaptability.
[0007] To achieve the above object, the technical solution adopted by the present invention is: a method for detecting short-term and rapid faults of contact networks based on event-driven neural networks, comprising the steps of:
[0008] S100, collecting real-time operation data of the overhead line system, the data including parameters such as current, voltage, displacement, tension, etc.;
[0009] S300, preprocessing the collected data, including denoising, filtering and normalization operations, to ensure the accuracy and consistency of the data;
[0010] S300, constructing an event-driven neural network (EDNN) model, wherein the model triggers activation and calculation of the neural network based on abnormal events in the contact network;
[0011] S400, the event-driven neural network includes an event detection module and a fault classification module, wherein: the event detection module detects abnormal events by monitoring the electrical and mechanical parameters of the contact network using the following formula:
[0012] ΔP=|P current -P prev
[0013] Among them, P current is the parameter value at the current time point, P prev is the parameter value at the previous time point. If ΔP exceeds the set threshold T event , then an abnormal event is triggered;
[0014] The fault classification module classifies the detected abnormal events. The classification process is based on the forward propagation calculation in the event-driven neural network. The feature extraction process of the convolution layer is:
[0015] Z conv =σ(W*X+b)
[0016] Where W is the convolution kernel weight, X is the input data, b is the bias term, σ is the activation function (such as ReLU), and the final output is the fault category C;
[0017] S500, through the event-driven mechanism, immediately activates the fault classification module when an abnormal event is detected, and performs rapid fault judgment and location;
[0018] S600: Output the fault detection result and generate corresponding alarm prompts to remind maintenance personnel to handle it in time.
[0019] The preprocessing step further includes a time series normalization operation of the data to ensure that the data in different time periods are comparable.
[0020] The event-driven neural network model is trained by an unsupervised learning method, and features are learned using normal and fault events in historical data to improve the fault detection accuracy of the model.
[0021] The event detection module sets a multi-parameter combined threshold trigger mechanism, and its trigger formula is:
[0022]
[0023] Among them, w i is the parameter weight, P i is the current value of each detection parameter, T is the preset trigger threshold, and the function f(I) is used to control the threshold change under different conditions.
[0024] The fault classification module extracts fault features through a deep learning architecture based on a convolutional neural network (CNN), including multi-layer convolution and pooling operations, to quickly identify and classify fault types. The loss function is the cross entropy loss, defined as:
[0025]
[0026] Where C is the number of fault categories, y i is the true label, is the fault classification probability output by the model.
[0027] It further includes matching the detected fault events with historical data to classify the fault types and perform trend analysis.
[0028] The output detection results include the specific type, location and occurrence time of the fault, and the detection information is presented through a visual interface.
[0029] It further includes generating maintenance suggestions based on the fault detection results and closing the alarm prompt after the system parameters return to normal.
[0030] The beneficial effects of adopting this technical solution are:
[0031] 1. Fast response and efficient detection: This technology is based on event-driven neural networks, which can instantly activate neural networks for analysis when abnormal events (such as sudden fluctuations in current, voltage, and displacement) are detected, greatly shortening the response time of fault detection. Compared with traditional periodic scanning monitoring systems, event-driven models are only activated when an abnormality is triggered, greatly improving detection efficiency.
[0032] 2. High-precision fault classification and location: By combining the deep learning architecture of convolutional neural network (CNN), this method can quickly extract fault features when an event occurs, and classify and accurately locate the fault. Multi-layer convolution operations can carefully analyze the operating parameters of the contact network, ensure the accuracy and robustness of fault detection, and reduce the probability of false detection and missed detection.
[0033] 3. Reduce system resource consumption: Traditional continuous monitoring methods require the system to continuously process a large amount of real-time data, which consumes a lot of resources. However, this technical solution uses an event-driven mechanism to start fault detection only when an abnormality occurs, which significantly reduces unnecessary computing burden and resource consumption, and improves the operating efficiency of the system.
[0034] 4. Robustness in complex environments: This method can still work stably in various complex operating environments. Regardless of different climatic conditions, load changes or other external interferences, the event-driven neural network can keenly capture the abnormal behavior of the contact network and ensure efficient fault detection.
[0035] 5. Automated and intelligent alarm function: This technical solution can not only detect faults in real time, but also automatically generate alarm prompts and maintenance suggestions based on the detection results, effectively reducing the need for manual intervention and improving the maintenance efficiency and safety of the contact network system.
[0036] In summary, this technical solution has significant advantages in contact network fault detection, such as rapid response, high-precision detection, low resource consumption and automatic alarm, which greatly improves the safety and operation and maintenance efficiency of the railway electrification system. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 The present invention is a schematic flow chart of a method for detecting short-time and rapid faults of a contact network based on an event-driven neural network. DETAILED DESCRIPTION
[0038] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described below with reference to the accompanying drawings.
[0039] In this embodiment, see Figure 1 As shown, the present invention proposes a method for detecting short-term and rapid faults of a contact network based on an event-driven neural network, comprising the steps of:
[0040] S100, collecting real-time operation data of the overhead line system, the data including parameters such as current, voltage, displacement, tension, etc.;
[0041] S300, preprocessing the collected data, including denoising, filtering and normalization operations, to ensure the accuracy and consistency of the data;
[0042] S300, constructing an event-driven neural network (EDNN) model, wherein the model triggers activation and calculation of the neural network based on abnormal events in the contact network;
[0043] S400, the event-driven neural network includes an event detection module and a fault classification module, wherein: the event detection module detects abnormal events by monitoring the electrical and mechanical parameters of the contact network using the following formula:
[0044] ΔP=|P current -P prev
[0045] Among them, P current is the parameter value at the current time point, P prev is the parameter value at the previous time point. If ΔP exceeds the set threshold T event , then an abnormal event is triggered;
[0046] The fault classification module classifies the detected abnormal events. The classification process is based on the forward propagation calculation in the event-driven neural network. The feature extraction process of the convolution layer is:
[0047] Z conv =σ(W*X+b)
[0048] Where W is the convolution kernel weight, X is the input data, b is the bias term, σ is the activation function (such as ReLU), and the final output is the fault category C;
[0049] S500, through the event-driven mechanism, immediately activates the fault classification module when an abnormal event is detected, and performs rapid fault judgment and location;
[0050] S600: Output the fault detection result and generate corresponding alarm prompts to remind maintenance personnel to handle it in time.
[0051] The preprocessing step further includes a time series normalization operation of the data to ensure that the data in different time periods are comparable.
[0052] The event-driven neural network model is trained by an unsupervised learning method, and features are learned using normal and fault events in historical data to improve the fault detection accuracy of the model.
[0053] The event detection module sets a multi-parameter combined threshold trigger mechanism, and its trigger formula is:
[0054]
[0055] Among them, w i is the parameter weight, P i is the current value of each detection parameter, T is the preset trigger threshold, and the function f(I) is used to control the threshold change under different conditions.
[0056] The fault classification module extracts fault features through a deep learning architecture based on a convolutional neural network (CNN), including multi-layer convolution and pooling operations, to quickly identify and classify fault types. The loss function is the cross entropy loss, defined as:
[0057]
[0058] Where C is the number of fault categories, y i is the true label, is the fault classification probability output by the model.
[0059] It further includes matching the detected fault events with historical data to classify the fault types and perform trend analysis.
[0060] The output detection results include the specific type, location and occurrence time of the fault, and the detection information is presented through a visual interface.
[0061] This includes generating maintenance recommendations based on fault detection results and shutting down alarm prompts after system parameters return to normal.
[0062] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.
Claims
1. A method for detecting short-term and rapid faults in overhead contact networks based on event-driven neural networks, characterized in that: The following steps are involved: S100, collecting real-time operation data of the overhead line system, the data including parameters such as current, voltage, displacement, tension, etc.; S300, preprocessing the collected data, including denoising, filtering and normalization operations, to ensure the accuracy and consistency of the data; S300, constructing an event-driven neural network (EDNN) model, wherein the model triggers activation and calculation of the neural network based on abnormal events in the contact network; S400, the event-driven neural network includes an event detection module and a fault classification module, wherein: the event detection module is used to monitor the electrical and mechanical parameters of the contact network and detect abnormal fluctuations or emergencies; the fault classification module learns the characteristic differences between the normal state and the fault state of the contact network through the neural network, and classifies the input abnormal events based on the preset fault threshold; The following formula is used to detect abnormal events: ΔP=|P current -P prev Among them, P current is the parameter value at the current time point, P prev is the parameter value at the previous time point. If ΔP exceeds the set threshold T event , then an abnormal event is triggered; The fault classification module classifies the detected abnormal events. The classification process is based on the forward propagation calculation in the event-driven neural network. The feature extraction process of the convolution layer is: Z conv =σ(W*X+b) Where W is the convolution kernel weight, X is the input data, b is the bias term, σ is the activation function (such as ReLU), and the final output is the fault category C; S500, through the event-driven mechanism, immediately activates the fault classification module when an abnormal event is detected, and performs rapid fault judgment and location; S600: Output the fault detection result and generate corresponding alarm prompts to remind maintenance personnel to handle it in time.
2. According to claim 1, a method for detecting short-term and rapid faults of a contact network based on an event-driven neural network is characterized in that: The preprocessing step further includes a time series normalization operation of the data to ensure that the data in different time periods are comparable.
3. According to claim 1, a method for detecting short-term and rapid faults of a contact network based on an event-driven neural network is characterized in that: The event-driven neural network model is trained by an unsupervised learning method, and features are learned using normal and fault events in historical data to improve the fault detection accuracy of the model.
4. According to claim 1, a method for detecting short-term and rapid faults of a contact network based on an event-driven neural network is characterized in that: The event detection module adopts a threshold trigger mechanism, which monitors the operating parameters of the contact network in real time and triggers event detection when one or more parameters exceed the set threshold.
5. The method for detecting short-term and rapid faults of a contact network based on an event-driven neural network according to claim 1, characterized in that: The fault classification module extracts fault features through a deep learning architecture based on a convolutional neural network (CNN), including multi-layer convolution and pooling operations, to quickly identify and classify fault types.
6. The method for detecting short-term and rapid faults of overhead contact network based on event-driven neural network according to claim 1, characterized in that: It further includes matching the detected fault events with historical data to classify the fault types and perform trend analysis.
7. The method for detecting short-term and rapid faults of a contact network based on an event-driven neural network according to claim 1, characterized in that: The output detection results include the specific type, location and occurrence time of the fault, and the detection information is presented through a visual interface.
8. A method for detecting short-term and rapid faults of a contact network based on an event-driven neural network as claimed in claim 1, characterized in that: It further includes generating maintenance suggestions based on the fault detection results and closing the alarm prompt after the system parameters return to normal.