High-voltage monitoring equipment for motor train unit and monitoring

Abstract

The invention discloses high-voltage monitoring equipment and monitoring and high-voltage side sensing modules for a motor train unit, and the equipment comprises a distributed optical fiber sensor, an optical current / voltage sensor, an acoustic-optical-electric joint bureau probe and a multi-dimensional environment sensing node, and is used for collecting voltage / current sensing, insulation state sensing, equipment state sensing and environment sensing data; and the vehicle-mounted edge calculation module is used for optical signal demodulation and electric signal preprocessing, cross-domain noise filtering based on mutual injection chaotic synchronization, multi-dimensional feature synchronous extraction and primary fusion diagnosis, data compression and encryption transmission, a cloud twinborn module, a high-voltage system digital twinborn body, a dynamic fault knowledge graph and a parameter self-evolution engine. Through comparison of simulation and actual measurement data, internal parameters are automatically adjusted to match a real equipment state, and based on this, life prediction is performed, and active maintenance early warning is triggered.

Description

Technical Field

[0001] This invention belongs to the field of high-speed train technology, specifically, it relates to a high-voltage monitoring device and monitoring system for high-speed trains. Background Technology

[0002] The high-voltage electrical system on the roof of high-speed trains is exposed to a complex and ever-changing operating environment for extended periods, and its insulation condition and equipment health directly affect train operation safety. Currently, mainstream high-voltage monitoring technologies in the industry are mainly based on traditional electrical sensing principles and discrete information system architectures, exhibiting systemic shortcomings in reliability, intelligence, and operation and maintenance models. These shortcomings manifest in the following three technical bottlenecks: I. Sensing Layer: Monitoring methods are singular and isolated, with poor environmental adaptability and insufficient reliability of basic data. Existing technologies largely rely on independently operating electrical sensors such as electromagnetic transformers and pulse current partial discharge sensors. These sensors themselves are both sources of electromagnetic interference and receivers. Under conditions of strong electromagnetic transients caused by train start-up and shutdown, overcurrent phase shifts, and lightning strikes, they are highly susceptible to signal saturation, distortion, or even momentary failure, leading to the loss of critical protection signals. II. Diagnostic Layer: Low level of information fusion, ambiguous fault location, and lack of intelligent decision-making ability. Existing systems mostly use a simple threshold over-limit alarm mode, which can only answer whether there is an anomaly, but cannot answer where the anomaly is or why. When multiple sensors generate alarms, due to the lack of digital modeling of the physical topology and signal propagation path of the high-voltage system, maintenance personnel find it difficult to quickly locate the core faulty equipment from discrete alarm points, let alone identify the root cause of the fault (such as whether it is caused by insulation deterioration, poor contact or external interference).

[0003] III. Operations and Maintenance Layer: Lagging status awareness, passive maintenance strategies, and lack of forward-looking management capabilities. Current maintenance models primarily rely on periodic inspections and reactive repairs, which are essentially time-based preventative maintenance or post-fault response. Because it's impossible to accurately monitor the microscopic aging state of equipment in real time (such as a gradual increase in contact resistance or slow deterioration of insulation materials), maintenance cycles are often either too conservative, leading to resource waste, or too risky, resulting in safety hazards. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a high-voltage monitoring device and monitoring system for high-speed trains.

[0005] To achieve the aforementioned objectives, the technical solution adopted by this invention includes: a high-voltage monitoring device for high-speed trains, comprising an onboard PHM system, including: The high-voltage side sensing module includes distributed fiber optic sensors, optical current / voltage sensors, a combined acoustic-optical-electrical probe, and multi-dimensional environmental sensing nodes. Multiple optical current / voltage sensors are included, with current sensors mounted around the main circuit breaker output terminals, grounding switch leads, and traction transformer primary cables. Voltage sensors are installed inside the composite insulators of the roof busbar. The distributed fiber optic sensors consist of distributed fiber optic gratings inscribed on the surfaces of circuit breakers and disconnectors, forming a distributed sensing system with a fiber optic grating array to measure micro-strain and temperature distribution. The multi-dimensional environmental sensing nodes include an online pollution monitor, temperature and humidity sensors, and a miniature weather station, installed between the insulator skirts near the pantograph on the roof or on the roof's high-voltage equipment area frame. This module employs the fiber optic composite overhead ground wire (OPGW) principle or an independent isolated power supply method. The power supply for the roof-mounted active equipment is located inside the vehicle, converting electrical energy into safe low voltage via an isolation transformer and a DC-DC converter, and then delivering it to each sensing node on the roof via a dedicated anti-electromagnetic interference cable or fiber optic power transmission system. The vehicle-mounted edge computing module is used for optical signal demodulation and electrical signal preprocessing, cross-domain noise filtering based on mutual injection chaotic synchronization, multi-dimensional feature synchronous extraction and primary fusion diagnosis, data compression and encrypted transmission. The cloud-based twin module includes a digital twin of the high-voltage system, a dynamic fault knowledge graph, and a parameter self-evolution engine. This invention's system is based on a heterogeneous computing platform, including an FPGA for high-speed demodulation and preprocessing of multiple optical signals; a GPU or dedicated AI acceleration chip for running lightweight GNN models and chaotic calculations; a security encryption module for vehicle-to-ground communication; and a description of a communication network module added before the vehicle-mounted edge computing module in the device section. For example: "The system constructs an all-fiber sensor access network. Various optical sensors are directly connected via optical fibers; non-optical sensors (such as UHF antennas) are connected to the fiber optic network through local photoelectric conversion modules. The network adopts deterministic Ethernet or lightweight TDMA (Time Division Multiple Access) protocols to ensure the real-time and deterministic transmission of multiple sensor data streams to the edge computing gateway."

[0006] Furthermore, the acoustic-optical-electric combined local probe integrates an ultra-high frequency antenna, an ultrasonic sensor, and an ultraviolet photosensitive element, and includes a high-precision synchronous clock generation and sampling trigger circuit to ensure that the ADC sampling of the three signals starts from the same absolute time scale, and the output is a parallel digital signal stream with a unified time scale.

[0007] A high-voltage monitoring method for high-speed trains includes the following steps: A1. Optical-electric hybrid multidimensional synchronous sensing and signal acquisition, acquiring voltage / current sensing, insulation status sensing, equipment status sensing and environmental sensing data; Step A1 specifically includes the following steps: A1.1 Based on the crystal electro-optic effect, the relationship between the output light intensity change ΔI and the electric field strength E is: ΔI / I0 = k *V, where k is the comprehensive electro-optic coefficient, which is determined by the crystal material and the optical path structure; A1.2 The output of the acoustic-optical-electric combined local probe only completes high-precision hardware time synchronization and multi-channel signal parallel acquisition inside the probe, and the output is a parallel raw data stream with a unified time scale; A1.3 Multi-dimensional environmental sensing nodes sense strain and temperature distribution and collect data.

[0008] A2. Edge-side noise filtering and feature enhancement based on mutual injection chaotic synchronization; Step A2 specifically includes: A2.1 In the edge computing gateway, a corresponding chaotic oscillator model (such as an improved Duffing oscillator) is established for each monitoring signal to be processed, and its dynamic equation is: Where, F cos(ωt): built-in driving force; γ is the signal injection coefficient, a dimensional conversion coefficient whose function is to transform the amplitude scale of the physical sensor signal S_a(t) to match the magnitude of the system's built-in driving force F cos(ωt). This coefficient needs to be calibrated according to the sensor sensitivity and the expected signal amplitude; S_a(t): partial discharge ultrasonic signal; k: oscillator damping coefficient, a dimensionless parameter greater than 0, controlling the dissipation characteristics of the system. Its specific value is determined through system dynamic characteristic calibration experiments; x: oscillator state variable, -x +x 3 : Nonlinear restoring force term; F: Amplitude of the built-in periodic driving force; ω: Angular frequency of the built-in periodic driving force. A2.2. Two chaotic oscillators driven by sensor signals from the same monitoring point but based on different principles are coupled through feedback of the difference in state variables (x1, x2). The coupling term is: C 12 = σ(x2 - x1), add to their respective equations, when the mutual injection coupling strength σ (σ is a dimensionless parameter greater than 0. Its "critical value" is the key threshold for two chaotic oscillators to enter a synchronized state from asynchrony, which is related to the specific oscillator parameters (k, F, ω). The feasible range needs to be determined in advance through theoretical calculation (such as the master stability function method) combined with simulation) is greater than the critical coupling strength and the two signals are from the same source (originating from the same physical event), the two chaotic oscillators will reach synchronization, and their output x1(t) ≈ x2(t); A2.3 Calculate the root mean square (RMS) (Δx) of the difference between the outputs of the two synchronized chaotic oscillators. If RMS (Δx) is below a threshold, the acquired S_a(t) and S_u(t) are considered highly correlated and valid signals, and their synchronized trajectory is used as the enhanced feature. If RMS (Δx) is high, one or both signals are considered severely contaminated by noise, and an algorithm based on the Lyapunov exponential spectrum change of the chaotic oscillator is initiated to separate the signal components. When RMS (Δx) is very high, the signals are considered out of sync. The system switches to a single-channel chaotic oscillator array detection mode: the contaminated signal S_a(t) is simultaneously injected into multiple chaotic oscillators with slightly different parameters (e.g., ω values). By observing and comparing which oscillator first undergoes a chaotic-to-periodic state transition, the potential characteristic frequency of the submerged valid signal is estimated, thus achieving preliminary separation of signal and noise. This method has excellent extraction capabilities for weak fault signals with non-stationary background noise.

[0009] A3. Fault location and diagnosis based on spatiotemporal correlation constraints and dynamic knowledge graphs, solving the problems of fuzzy fault location and single diagnostic dimensions; Step A3 specifically includes: Spatiotemporal correlation graph construction: The topology of the high-voltage system, the location of sensor points, and the physical connection relationships are abstracted into an attribute graph; the nodes in the graph represent devices or sensors, and the edges represent connection relationships or signal transmission paths.

[0010] Dynamic knowledge graph injection: The fault knowledge graph downloaded from the cloud twin layer (e.g., the "insulator flashover" event associated with features such as "increased leakage current", "partial discharge mode is surface discharge", and "ambient humidity > 85%)" is injected into the above attribute graph as semantic constraints; Collaborative reasoning and localization: When multiple sensors trigger an early warning, the edge fusion gateway initiates graph neural network model inference: using the features extracted by each sensor (such as graph spectra and waveforms) as node feature inputs. Graph neural network models propagate and aggregate these feature information along the edges of the graph. Ultimately, each node outputs a fault probability score. The system not only outputs the faulty device node with the highest score, but also traces back to provide the most critical chain of evidence that led to the judgment (for example, the basis for determining "disconnecting switch A contact overheating" is: abnormal temperature of the fiber grating above it > temperature difference of adjacent phase B > partial discharge signal appearing near the switch opening and closing phases > historical graphs showing a similar pattern as contact oxidation).

[0011] Step A3 also includes: The knowledge graph differential incremental update mechanism generates a differential update package for the corresponding graph substructures (entities, relations, feature rules) after verifying a new fault mode in the cloud. During vehicle-to-ground communication, the gateway verifies and receives the update package, completes merging and consistency verification in the sandbox version of the knowledge graph, and atomically activates the new version through a dual-storage area switching method after successful verification. Simultaneously, it is explicitly stated that the knowledge graph exists in the in-vehicle GNN in the form of embedded vectors and attention mechanism weights.

[0012] A4. Digital twin-driven parameter self-evolution and health trend prediction solve the problems of poor adaptability and unpredictable maintenance.

[0013] Step A4 specifically includes: A4.1 Cloud-based Full-Physics Twin: Constructing a multi-physics dynamic simulation model of a high-pressure system encompassing electrical, magnetic, thermal, mechanical, and fluid dynamics aspects. Its core formulas for thermal simulation are based on extended heat transfer equations, but real-time contamination parameters are introduced as boundary conditions. ξ is the equivalent salt density / ash density parameter obtained in real time from environmental sensors, k is the nonlinear thermal conductivity related to temperature and pollution level, T: temperature field; k(T, ξ): nonlinear thermal conductivity; ▽T: temperature gradient; Q(I, R): internal heat source generation rate; ρ: material density; C_p: constant pressure specific heat capacity; : Rate of change of temperature over time; A4.2, Evolution of Edge-Cloud Collaboration Parameters: The edge device sends real-time operating conditions (current I, speed, ambient temperature and humidity) and processed features to the cloud. The cloud twin runs under the same input conditions and generates simulation features at corresponding times (such as simulated temperature distribution and simulated partial discharge spectrum). Calculate the deviation vector ΔF between the simulated features and the measured features; The sensitivity matrix S is correlated with the adjustable parameters (such as contact resistance R_c, insulation aging factor α) in the twin. The sensitivity matrix S is obtained online through the adjoint equation method or perturbation method. For the target parameter P_j, its sensitivity is... This refers to the partial derivative of the i-th feature with respect to the j-th parameter. In each parameter evolution iteration, the twin observes the change ΔF_i of the simulated feature F_sim through a small perturbation ΔP_j, thereby approximately calculating S_{ij} ≈ ΔF_i / ΔP_j. Then, using a regularized least squares optimization algorithm, these parameters are fine-tuned in reverse to ensure that the twin continuously approximates the state of the real system. The optimization problem is: min ||S·ΔP - ΔF|| 2 + λ||ΔP|| 2ΔP is the parameter adjustment amount, and λ is the regularization coefficient to prevent overfitting. It is used to prevent the parameter update amount ΔP from being too large or the solution from being unstable due to noise in the observed data ΔF or the ill-conditioned nature of the model sensitivity matrix S. It is usually selected by the L-curve method or cross-validation method. A4.3 Health Prediction: The evolved twin can perform accelerated stress simulation. For example, given a future operating plan, it can predict the aging curve of critical insulation components and trigger predictive maintenance warnings when the predicted remaining functional life is lower than a safety threshold.

[0014] A5. The edge gateway periodically (e.g., daily) synchronizes and verifies the diagnostic results and evolved local model parameters with the cloud. The cloud utilizes data from the entire fleet to perform fault mode mining and model generalization optimization across vehicles and lines, generating updated general diagnostic models and knowledge graph differential packages, which are then distributed to the edge gateways of each train through a secure channel, completing the evolution from single-train intelligence to swarm intelligence.

[0015] A computer-readable storage medium having a computer program stored thereon, characterized in that: when the program is executed by a processor, it performs steps for high-voltage monitoring of a high-speed train.

[0016] A mobile terminal includes a mobile terminal body and a controller. The controller includes a memory, a processor, and a computer program stored in the memory and executable on the memory. When the processor executes the program, it implements the steps of a high-voltage monitoring method for high-speed trains.

[0017] Compared with the prior art, the advantages of the present invention include: (1) The present invention provides a high-voltage monitoring device and monitoring system for high-speed trains: Unlike simple sensor stacking or multi-source information fusion, this solution uses optical fiber for unified transmission at the physical layer and a nonlinear dynamic method of mutual injection chaotic synchronization as a jury judgment mechanism for signal validity at the information layer. This mechanism can fundamentally identify and suppress incoherent noise, solve the pain point of monitoring failure under specific interference, and represents a qualitative leap in the perception layer; (2) This invention provides a high-voltage monitoring device and monitoring system for high-speed trains, which combines the system's physical topology (spatiotemporal correlation graph) with expert experience and historical data (dynamic knowledge graph), and performs reasoning through a lightweight graph neural network model that can run on the vehicle's edge. It can not only pinpoint where the problem is, but also explain why it was determined to be faulty through evidence chain tracing, achieving an evolution in interpretability from a black box to a gray box in the diagnostic process, greatly improving positioning accuracy and operational reliability. (3) The high-voltage monitoring equipment and monitoring system provided by this invention for high-speed trains is a cloud-based digital twin that is not a static model, but a living model whose parameters can evolve online, automatically, and continuously. By comparing the deviations between actual measurements and simulations, it optimizes its own physical parameters (such as contact resistance and aging factor) in reverse, so that the model always tracks the actual degradation state of individual trains. The life prediction based on this evolved model has a high degree of individualization and accuracy, and is the core of predictive maintenance. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of a high-voltage monitoring device for high-speed trains and the monitoring process according to the present invention. Figure 2 This is a schematic diagram of the high-voltage system layout for high-speed trains. Detailed Implementation

[0020] In view of the shortcomings of the prior art, the inventors of this invention, through long-term research and extensive practice, have proposed the technical solution of this invention. The technical solution, its implementation process, and principles will be further explained below with reference to the accompanying drawings and specific implementation examples in the embodiments of this application.

[0021] It should be noted that the embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, the present invention covers any substitutions, modifications, equivalent methods and solutions made on the spirit, principles and scope of the present invention as defined by the claims. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] In the description of this application, the terms "first," "second," "third," and similar words do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms "a" or "one," and similar words, do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "comprising" or "including," and similar words, mean that the elements or objects preceding "comprising" or "including" encompass the elements or objects listed following "comprising" or "including," and their equivalents, but do not exclude other elements or objects. The terms "connected" or "linked," and similar words, are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0023] In the description of this application, the terms "center," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used solely for the convenience of describing this application and for simplification, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. Furthermore, when using positional terms such as "both sides," "outer side," and "upper and lower," it should be understood that they are used only for ease of understanding and description, taking into account that the structure may be oriented to other positions.

[0024] In the description of this application, unless otherwise expressly specified and limited, the technical or scientific terms used shall have the ordinary meaning understood by a person with ordinary skills in the art to which this application pertains. Terms such as “installation,” “connection,” and “joining” shall be interpreted broadly, for example, as fixed connection, detachable connection, mating connection, or integral connection. For a person skilled in the art, the specific meaning of the above terms in this application can be understood according to the specific circumstances.

[0025] The present invention aims to introduce and explain the high-voltage monitoring equipment for high-speed trains, the structural composition of the monitoring equipment, and the cooperation relationship between the various components. Unless otherwise specified, the dimensions, materials, and manufacturing processes of the components of the high-voltage monitoring equipment suitable for high-speed trains in the present invention can be selected according to specific circumstances, and no special limitations or explanations are made here.

[0026] Furthermore, to provide the public with a better understanding of the present invention, certain specific details are described in detail in the following description of the invention. However, those skilled in the art will fully understand the invention even without these detailed descriptions.

[0027] Please see Figure 2The main monitoring targets of the high-voltage system of the EMU (taking 25kV AC as an example) are the high-voltage equipment on the roof and the high-voltage cabinet inside the train. Figure 2 The typical installation layout of various sensors on the high-voltage side of the EMU is clearly shown.

[0028] Example 1 Please see Figure 1 A high-voltage monitoring device for high-speed trains, including an onboard PHM system, comprising: The high-voltage side sensing module includes distributed fiber optic sensors, optical current / voltage sensors, a combined acoustic-optical-electrical probe, and multi-dimensional environmental sensing nodes. Multiple optical current / voltage sensors are included, with current sensors mounted around the main circuit breaker outgoing lines, grounding switch leads, and traction transformer primary cables. Voltage sensors are installed inside the composite insulators of the roof busbar. The distributed fiber optic sensors consist of distributed fiber optic gratings inscribed on the surfaces of circuit breakers and disconnectors, forming a distributed sensing system. Special adhesive is used to tightly adhere the system to the roof high-voltage busbar housing, the operating mechanism housings of the main circuit breaker and disconnector, and the roots of key supporting insulators, measuring micro-strain and temperature distribution. The optical fiber serves as both the sensing element and the transmission medium. By detecting changes in the intensity, frequency, or phase of backscattered light in the optical fiber, the temperature and micro-strain distribution along the line are continuously measured. The multi-dimensional environmental sensing nodes include an online pollution monitor, temperature and humidity sensors, and a miniature weather station. This module employs either the fiber optic composite overhead ground wire (OPGW) principle or an independent isolated power supply method. The power supply for the active equipment on the roof is located inside the vehicle. The electrical energy is converted into a safe low voltage through an isolation transformer and a DC-DC converter, and then sent to each sensing node on the roof through a dedicated anti-electromagnetic interference cable or fiber optic power transmission system. The vehicle-mounted edge computing module is used for optical signal demodulation and electrical signal preprocessing, cross-domain noise filtering based on mutual injection chaotic synchronization, multi-dimensional feature synchronous extraction and primary fusion diagnosis, data compression and encrypted transmission. The cloud-based twin module includes a digital twin of the high-voltage system, a dynamic fault knowledge graph, and a parameter self-evolution engine. This invention's system is based on a heterogeneous computing platform, including an FPGA for high-speed demodulation and preprocessing of multiple optical signals; a GPU or dedicated AI acceleration chip for running lightweight GNN models and chaotic calculations; a security encryption module for vehicle-to-ground communication; and a description of a communication network module added before the vehicle-mounted edge computing module in the device section. For example: "The system constructs an all-fiber sensor access network. Various optical sensors are directly connected via optical fibers; non-optical sensors (such as UHF antennas) are connected to the fiber optic network through local photoelectric conversion modules. The network adopts deterministic Ethernet or lightweight TDMA (Time Division Multiple Access) protocols to ensure the real-time and deterministic transmission of multiple sensor data streams to the edge computing gateway."

[0029] Furthermore, the combined acoustic-optical-electric local probe integrates an ultra-high frequency antenna, an ultrasonic sensor, and an ultraviolet photosensitive element, and includes a high-precision synchronous clock generation and sampling trigger circuit to ensure that the ADC sampling of the three signals starts from the same absolute time scale, and the output is a parallel digital signal stream with a unified time scale; the combined acoustic-optical-electric local probe is installed on the grounded metal shell of the monitored equipment (such as current transformers and circuit breakers), or clipped to the equipment grounding wire through the "Rogowski coil + capacitive coupling" method.

[0030] A high-voltage monitoring method for high-speed trains includes the following steps: A1. Optical-electric hybrid multidimensional synchronous sensing and signal acquisition, acquiring voltage / current sensing, insulation status sensing, equipment status sensing and environmental sensing data; Step A1 specifically includes the following steps: A1.1 Based on the crystal electro-optic effect, the relationship between the output light intensity change ΔI and the electric field strength E is: ΔI / I0 = k *V, where k is the comprehensive electro-optic coefficient, which is determined by the crystal material and the optical path structure; A1.2 The output of the acoustic-optical-electric combined local probe only completes high-precision hardware time synchronization and multi-channel signal parallel acquisition inside the probe, and the output is a parallel raw data stream with a unified time scale; A1.3 Multi-dimensional environmental sensing nodes sense strain and temperature distribution and collect data.

[0031] A2. Edge-side noise filtering and feature enhancement based on mutual injection chaotic synchronization; Step A2 specifically includes: A2.1 In the edge computing gateway, a corresponding chaotic oscillator model (such as an improved Duffing oscillator) is established for each monitoring signal to be processed, and its dynamic equation is: Where, F cos(ωt): built-in driving force; γ is the signal injection coefficient, a dimensional conversion coefficient whose function is to transform the amplitude scale of the physical sensor signal S_a(t) to match the magnitude of the system's built-in driving force F cos(ωt). This coefficient needs to be calibrated according to the sensor sensitivity and the expected signal amplitude; S_a(t): partial discharge ultrasonic signal; k: oscillator damping coefficient, a dimensionless parameter greater than 0, controlling the dissipation characteristics of the system. Its specific value is determined through system dynamic characteristic calibration experiments; x: oscillator state variable, -x +x 3 : Nonlinear restoring force term; F: Amplitude of the built-in periodic driving force; ω: Angular frequency of the built-in periodic driving force. A2.2. Two chaotic oscillators driven by sensor signals from the same monitoring point but based on different principles are coupled through feedback of the difference in state variables (x1, x2). The coupling term is: C12 = σ(x2 - x1), add to their respective equations, when the mutual injection coupling strength σ (σ is a dimensionless parameter greater than 0. Its "critical value" is the key threshold for two chaotic oscillators to enter a synchronized state from asynchrony, which is related to the specific oscillator parameters (k, F, ω). The feasible range needs to be determined in advance through theoretical calculation (such as the master stability function method) combined with simulation) is greater than the critical coupling strength and the two signals are from the same source (originating from the same physical event), the two chaotic oscillators will reach synchronization, and their output x1(t) ≈ x2(t); A2.3 Calculate the root mean square (RMS) (Δx) of the difference between the outputs of the two synchronized chaotic oscillators. If RMS (Δx) is below a threshold, the acquired S_a(t) and S_u(t) are considered highly correlated and valid signals, and their synchronized trajectory is used as the enhanced feature. If RMS (Δx) is high, one or both signals are considered severely contaminated by noise, and an algorithm based on the Lyapunov exponential spectrum change of the chaotic oscillator is initiated to separate the signal components. When RMS (Δx) is very high, the signals are considered out of sync. The system switches to a single-channel chaotic oscillator array detection mode: the contaminated signal S_a(t) is simultaneously injected into multiple chaotic oscillators with slightly different parameters (e.g., ω values). By observing and comparing which oscillator first undergoes a chaotic-to-periodic state transition, the potential characteristic frequency of the submerged valid signal is estimated, thus achieving preliminary separation of signal and noise. This method has excellent extraction capabilities for weak fault signals with non-stationary background noise.

[0032] A3. Fault location and diagnosis based on spatiotemporal correlation constraints and dynamic knowledge graphs, solving the problems of fuzzy fault location and single diagnostic dimensions; Step A3 specifically includes: Spatiotemporal correlation graph construction: The topology of the high-voltage system, the location of sensor points, and the physical connection relationships are abstracted into an attribute graph; the nodes in the graph represent devices or sensors, and the edges represent connection relationships or signal transmission paths.

[0033] Dynamic knowledge graph injection: The fault knowledge graph downloaded from the cloud twin layer (e.g., the "insulator flashover" event associated with features such as "increased leakage current", "partial discharge mode is surface discharge", and "ambient humidity > 85%)" is injected into the above attribute graph as semantic constraints; Collaborative reasoning and localization: When multiple sensors trigger an early warning, the edge fusion gateway initiates graph neural network model inference: using the features extracted by each sensor (such as graph spectra and waveforms) as node feature inputs. Graph neural network models propagate and aggregate these feature information along the edges of the graph. Ultimately, each node outputs a fault probability score. The system not only outputs the faulty device node with the highest score, but also traces back to provide the most critical chain of evidence that led to the judgment (for example, the basis for determining "disconnecting switch A contact overheating" is: abnormal temperature of the fiber grating above it > temperature difference of adjacent phase B > partial discharge signal appearing near the switch opening and closing phases > historical graphs showing a similar pattern as contact oxidation).

[0034] Step A3 also includes: The knowledge graph differential incremental update mechanism generates a differential update package for the corresponding graph substructures (entities, relations, feature rules) after verifying a new fault mode in the cloud. During vehicle-to-ground communication, the gateway verifies and receives the update package, completes merging and consistency verification in the sandbox version of the knowledge graph, and atomically activates the new version through a dual-storage area switching method after successful verification. Simultaneously, it is explicitly stated that the knowledge graph exists in the in-vehicle GNN in the form of embedded vectors and attention mechanism weights.

[0035] A4. Digital twin-driven parameter self-evolution and health trend prediction solve the problems of poor adaptability and unpredictable maintenance.

[0036] Step A4 specifically includes: A4.1 Cloud-based Full-Physics Twin: Constructing a multi-physics dynamic simulation model of a high-pressure system encompassing electrical, magnetic, thermal, mechanical, and fluid dynamics aspects. Its core formulas for thermal simulation are based on extended heat transfer equations, but real-time contamination parameters are introduced as boundary conditions. ξ is the equivalent salt density / ash density parameter obtained in real time from environmental sensors, k is the nonlinear thermal conductivity related to temperature and pollution level, T: temperature field; k(T, ξ): nonlinear thermal conductivity; ▽T: temperature gradient; Q(I, R): internal heat source generation rate; ρ: material density; C_p: constant pressure specific heat capacity; : Rate of change of temperature over time; A4.2, Evolution of Edge-Cloud Collaboration Parameters: The edge device sends real-time operating conditions (current I, speed, ambient temperature and humidity) and processed features to the cloud. The cloud twin runs under the same input conditions and generates simulation features at corresponding times (such as simulated temperature distribution and simulated partial discharge spectrum). Calculate the deviation vector ΔF between the simulated features and the measured features; The sensitivity matrix S is correlated with the adjustable parameters (such as contact resistance R_c, insulation aging factor α) in the twin. The sensitivity matrix S is obtained online through the adjoint equation method or perturbation method. For the target parameter P_j, its sensitivity is... This refers to the partial derivative of the i-th feature with respect to the j-th parameter. In each parameter evolution iteration, the twin observes the change ΔF_i of the simulated feature F_sim through a small perturbation ΔP_j, thereby approximately calculating S_{ij} ≈ ΔF_i / ΔP_j. Then, using a regularized least squares optimization algorithm, these parameters are fine-tuned in reverse to ensure that the twin continuously approximates the state of the real system. The optimization problem is: min ||S·ΔP - ΔF|| 2 + λ||ΔP|| 2 ΔP is the parameter adjustment amount, and λ is the regularization coefficient to prevent overfitting. It is used to prevent the parameter update amount ΔP from being too large or the solution from being unstable due to noise in the observed data ΔF or the ill-conditioned nature of the model sensitivity matrix S. It is usually selected by the L-curve method or cross-validation method. A4.3 Health Prediction: The evolved twin can perform accelerated stress simulation. For example, given a future operating plan, it can predict the aging curve of critical insulation components and trigger predictive maintenance warnings when the predicted remaining functional life is lower than a safety threshold.

[0037] A5. The edge gateway periodically (e.g., daily) synchronizes and verifies the diagnostic results and evolved local model parameters with the cloud. The cloud utilizes data from the entire fleet to perform fault mode mining and model generalization optimization across vehicles and lines, generating updated general diagnostic models and knowledge graph differential packages, which are then distributed to the edge gateways of each train through a secure channel, completing the evolution from single-train intelligence to swarm intelligence.

[0038] A computer-readable storage medium having a computer program stored thereon, characterized in that: when the program is executed by a processor, it performs steps for high-voltage monitoring of a high-speed train.

[0039] A mobile terminal includes a mobile terminal body and a controller. The controller includes a memory, a processor, and a computer program stored in the memory and executable on the memory. When the processor executes the program, it implements the steps of a high-voltage monitoring method for high-speed trains.

[0040] Working principle: Step 1: All sensors send data to the edge gateway through an all-fiber network. Mutual injection chaotic synchronization technology is used to automatically distinguish between real fault signals and noise, ensuring that the features processed in subsequent steps are highly reliable.

[0041] The second step: The system integrates the physical connection relationship of the equipment (spatiotemporal correlation graph) with the fault logic knowledge (dynamic knowledge graph), and performs intelligent reasoning through the vehicle-mounted graph neural network model (GNN). This not only locates the faulty equipment, but also provides the basis for judgment (chain of evidence).

[0042] Step 3: Run the cloud-based digital twin simulation and compare it with the measured data. This is achieved using a self-evolutionary algorithm (the core of which is solving for min ||S·ΔP - ΔF||). 2 It dynamically corrects its own parameters (such as contact resistance) to make them consistent with the actual condition of the train, thereby accurately predicting the life of components based on this model.

[0043] Step 4: The cloud uses data from the entire fleet to continuously optimize the global diagnostic model and knowledge base, and automatically distributes the optimization results to each train in the form of "patch packages" (differential increments), enabling the entire fleet to have "collective intelligence" for continuous learning, forming a closed loop.

[0044] The system of this invention is an intelligent system that "builds from individual precision health management to collective wisdom evolution", realizing a fundamental transformation from passive response to proactive prediction and maintenance.

[0045] It should be understood that the above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. It should not be considered that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, several simple deductions or substitutions can be made without departing from the concept of the present invention. All equivalent changes or modifications made in accordance with the spirit and essence of the present invention should be covered within the protection scope of the present invention.

Claims

1. A high pressure monitoring device for a train set, comprising an on-board PHM system, characterized in that: Also includes: The high-voltage side sensing module includes distributed fiber optic sensors, optical current / voltage sensors, a combined acoustic-optical-electrical probe, and multi-dimensional environmental sensing nodes. Multiple optical current / voltage sensors are provided, with current sensors mounted around the main circuit breaker outgoing lines, grounding switch down conductors, and traction transformer primary cables, while voltage sensors are installed inside the composite insulators of the roof busbar. The distributed fiber optic sensors include distributed fiber optic gratings inscribed on the surfaces of circuit breakers and disconnectors, forming a distributed sensing system with a fiber optic grating array to measure micro-strain and temperature distribution. The multi-dimensional environmental sensing nodes include an online pollution monitor, temperature and humidity sensors, and a miniature weather station. The vehicle-mounted edge computing module is used for optical signal demodulation and electrical signal preprocessing, cross-domain noise filtering based on mutual injection chaotic synchronization, multi-dimensional feature synchronous extraction and primary fusion diagnosis, data compression and encrypted transmission. The cloud-based twin module includes a digital twin of the high-voltage system, a dynamic fault knowledge graph, and a parameter self-evolution engine.

2. The high-voltage monitoring device for high-speed trains according to claim 1, characterized in that: The combined acoustic-optical-electric local probe integrates an ultra-high frequency antenna, an ultrasonic sensor, and an ultraviolet photosensitive element, and includes a high-precision synchronous clock generation and sampling trigger circuit to ensure that the ADC sampling of the three signals starts from the same absolute time scale, and the output is a parallel digital signal stream with a unified time scale.

3. A method for monitoring the high voltage of a train set, by means of a device as claimed in any one of the claims 1-2, characterized in that: Includes the following steps: A1. Optical-electric hybrid multidimensional synchronous sensing and signal acquisition, acquiring voltage / current sensing, insulation status sensing, equipment status sensing and environmental sensing data; A2. Edge-side noise filtering and feature enhancement based on mutual injection chaotic synchronization; A3. Fault location and diagnosis based on spatiotemporal correlation constraints and dynamic knowledge graph; A4. Digital twin-driven parameter self-evolution and health trend prediction; A5. The edge gateway periodically synchronizes and verifies the diagnostic results and the evolved local model parameters with the cloud. The cloud uses data from the entire fleet to perform fault mode mining and model generalization optimization across vehicles and lines, generating updated general diagnostic models and knowledge graph differential packages, which are then distributed to the edge gateways of each train through a secure channel, completing the evolution from single-train intelligence to collective intelligence.

4. The high voltage monitoring method for a motor train unit according to claim 3, wherein: Step A1 specifically includes the following steps: A1.1 Based on the electro-optic effect of crystals, the relationship between the change in output light intensity ΔI and the electric field strength E is: ΔI / I0 = k * V, where k is the comprehensive electro-optic coefficient; A1.2 The acoustic-optical-electrical joint local probe only completes high-precision hardware time synchronization and multi-channel signal parallel acquisition inside the probe, and outputs a parallel raw data stream with a unified time scale; chaotic synchronization processing is performed in the gateway, using the raw streams that are strictly time-aligned but feature-independent to perform deep, adaptive correlation verification and noise filtering, thereby determining whether they originate from the same physical event. A1.3 Multi-dimensional environmental sensing nodes sense strain and temperature distribution and collect data.

5. The high voltage monitoring method for a motor train unit according to claim 4, wherein: Step A2 specifically includes: A2.1 In the edge computing gateway, a corresponding chaotic oscillator model is established for each monitoring signal to be processed, and its dynamic equation is: Where, F cos(ωt): built-in driving force, γ is the signal injection coefficient; S_a(t): partial discharge ultrasonic signal, k: oscillator damping coefficient; x: oscillator state variable, -x + x 3 : Nonlinear restoring force term; F: Amplitude of the built-in periodic driving force; ω: Angular frequency of the built-in periodic driving force. A2.

2. Two chaotic oscillators driven by sensor signals from the same monitoring point but based on different principles are coupled through feedback of the difference in state variables (x1, x2). The coupling term is: C 12 = σ(x2 - x1), add to their respective equations, when the mutual injection coupling strength σ is greater than the critical coupling strength and the two signals are from the same source, the two chaotic oscillators will reach synchronization and their output x1(t)≈ x2(t); A2.3 Calculate the root mean square value (RMS(Δx)) of the difference between the outputs of the two synchronized chaotic oscillators. If RMS(Δx) is lower than the threshold, it is determined that the acquired S_a(t) and S_u(t) are highly correlated and are valid signals, and their synchronized trajectory is used as the enhanced feature. If RMS(Δx) is very high, it is determined that one or both signals are severely contaminated by noise, and the algorithm based on the Lyapunov exponential spectrum change of the chaotic oscillator is started to separate the signal components.

6. The high voltage monitoring method for a motor train unit according to claim 5, wherein: Step A3 specifically includes: Spatiotemporal correlation graph construction: The topology of the high-voltage system, the location of sensor points, and the physical connection relationships are abstracted into an attribute graph; Dynamic knowledge graph injection: The fault knowledge graph downloaded from the cloud twin module is used as a semantic constraint and injected into the above attribute graph; Collaborative reasoning and localization: When multiple sensors trigger an early warning, the edge fusion gateway starts graph neural network model reasoning, taking the features extracted by each sensor as node feature input. The graph neural network model passes and aggregates these feature information along the edges of the graph, and each node outputs a fault probability score.

7. The high-voltage monitoring method for high-speed trains according to claim 3, characterized in that: Step A3 also includes: The knowledge graph differential incremental update mechanism generates a differential update package for the graph substructure corresponding to the new fault mode after the cloud verifies the new fault mode. During vehicle-to-ground communication, the gateway verifies and receives the update package, completes merging and consistency verification in the sandbox version of this knowledge graph, and activates the new version atomically through dual storage area switching after verification. At the same time, it is clear that the knowledge graph exists in the graph neural network model in the form of embedded vectors and attention mechanism weights.

8. The high voltage monitoring method for a motor train unit according to claim 6, wherein: Step A4 specifically includes: A4.1 Cloud Twin Module: Constructs a multi-physics dynamic simulation model of a high-pressure system covering electrical, magnetic, thermal, mechanical, and fluid mechanics aspects. Its core formulas for thermal simulation are based on extended heat transfer equations, but real-time pollution parameters are introduced as boundary conditions. , where ξ is the equivalent salt density / ash density parameter obtained in real time from the environmental sensor, k is the nonlinear thermal conductivity related to temperature and pollution level, T: temperature field; k(T, ξ): nonlinear thermal conductivity; ▽ T: Temperature gradient; Q(I, R): Internal heat source generation rate; ρ: Material density; C_p: Specific heat capacity under constant pressure; : Rate of change of temperature over time; A4.2, Evolution of Edge-Cloud Collaboration Parameters: The edge device sends real-time operating conditions and processed features to the cloud; The cloud-based twin module operates under the same input conditions, generating simulation features at corresponding times. Calculate the deviation vector ΔF between the simulated features and the measured features; The sensitivity matrix S of adjustable parameters in the cloud-based digital twin module is correlated with ΔF. This sensitivity matrix S is obtained online using the adjoint equation method or perturbation method. A regularized least squares optimization algorithm is then used to fine-tune these parameters in reverse, ensuring the digital twin continuously approximates the state of the real system. The optimization problem is: min ||S·ΔP - ΔF|| 2 + λ||ΔP|| 2 , where ΔP is the parameter adjustment amount, and λ is the regularization coefficient to prevent overfitting; A4.3 Health Prediction: The evolved digital twin can perform accelerated stress simulation.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by the processor, it performs the steps of the method as described in any one of claims 3-8.

10. A mobile terminal comprising a mobile terminal body and a controller, characterized by: The controller includes a memory, a processor, and a computer program stored in and executable on the memory, wherein the processor executes the program to implement the steps of the method as described in any one of claims 3-8.

Images