A Parallel Fault Location Method and System for Valve Hall Equipment Based on Deep Fusion of Multispectral Features

Abstract

This invention relates to the field of power system equipment condition monitoring and fault diagnosis technology, specifically disclosing a parallel fault location method and system for valve hall equipment using deep fusion of multispectral features. By establishing a fault evolution time-series model library, collecting and analyzing instantaneous weak signals in multispectral data streams, generating a spatiotemporally labeled event fragment pool, and comparing it with the model library to generate trigger events, fault evolution hypotheses are generated. This activates a predictive enhanced monitoring mode, directionally captures subsequent evolution signals, constructs a complete fault evidence chain, and finally generates a precisely located fault diagnosis report. This invention achieves advanced fault warning and high-sensitivity fault location, significantly improving the foresight and reliability of fault diagnosis, while reducing false alarm rates and system operating power consumption. It also constructs an adaptive learning closed-loop system that can maintain and continuously improve the accuracy of fault prediction and location over the long term.

Description

Technical Field

[0001] This invention belongs to the field of power system equipment condition monitoring and fault diagnosis technology, and relates to a parallel fault location method and system for valve hall equipment with deep fusion of multispectral features. Background Technology

[0002] As the core of a high-voltage direct current (HVDC) transmission system, the stability of the internal equipment in the valve hall directly affects the safety of the entire power grid. The current key challenge lies in effectively identifying the early, subtle signs of equipment failure, thereby enabling predictive maintenance. These early signs often manifest as extremely brief physical phenomena with low signal strength, such as localized micro-discharges or slight abnormal temperature rises. They are usually submerged in the background noise of a strong electromagnetic environment, making them difficult to detect and characterize effectively using traditional monitoring methods.

[0003] Currently, the solutions commonly used in the industry mainly rely on a variety of independent monitoring systems based on high-threshold alarms. For example, infrared thermal imagers are used to monitor equipment overheating, ultraviolet imagers are used to detect corona discharge and surface discharge, and visible light cameras are used for routine inspections. These systems operate independently, monitoring a single physical phenomenon. When the detected signal strength exceeds a preset, high safety threshold, an alarm is triggered, prompting maintenance personnel to intervene.

[0004] However, this traditional approach has significant drawbacks. First, it is essentially a passive response mechanism, only detectable when a fault develops to a certain extent and generates a sufficiently strong signal, thus missing the optimal opportunity to address the fault in its early stages. Second, the data from different monitoring systems are isolated, making it impossible to correlate and analyze weak signals across different physical dimensions. Consequently, it cannot identify early fault chains composed of multiple successive weak events, resulting in severely insufficient diagnostic sensitivity. Finally, achieving high-sensitivity, all-weather monitoring inevitably generates massive amounts of data, placing enormous pressure on data transmission and processing, leading to high costs and low efficiency. Summary of the Invention

[0005] In view of this, in order to solve the problems mentioned in the background technology, a parallel fault location method and system for valve hall equipment based on deep fusion of multispectral features is proposed.

[0006] The objective of this invention can be achieved through the following technical solution: The first aspect of this invention provides a parallel fault location method for valve hall equipment based on deep fusion of multispectral features, including: S1, establishing an evolution time series model library: establishing a fault evolution time series model library.

[0007] S2, Event Fragment Pool Generation: Acquire multispectral data streams and identify transient weak signals from them to generate a spatiotemporally labeled event fragment pool.

[0008] S3. Trigger Event Matching and Generation: The event fragment pool with spatiotemporal tags is compared with the fault evolution timing model library to match and generate trigger events that conform to the initial stage of evolution.

[0009] S4. Fault Evolution Hypothesis Generation: Based on the fault evolution path information contained in the triggering events that conform to the initial stage of evolution, generate fault evolution hypotheses to be verified.

[0010] S5. Predictive Enhanced Monitoring Activation: Activate the predictive enhanced monitoring mode based on the fault evolution hypothesis to be verified.

[0011] S6. Targeted capture of subsequent evolution signals: In predictive enhanced monitoring mode, targeted capture of subsequent evolution signals.

[0012] S7. Fault Evidence Chain Construction: Verify the spatiotemporal correlation between subsequent evolution signals and triggering events that conform to the initial stage of evolution in order to construct a complete fault evidence chain.

[0013] S8. Fault Diagnosis Report Generation: Based on a complete chain of fault evidence, generate and output a fault diagnosis report with precise location.

[0014] The second aspect of the present invention provides a parallel fault location system for valve hall equipment with deep fusion of multispectral features, including: an evolution time series model library establishment module for establishing a fault evolution time series model library.

[0015] The event fragment pool generation module acquires multispectral data streams and identifies transient weak signals from them to generate a spatiotemporally labeled event fragment pool.

[0016] The trigger event matching and generation module compares the spatiotemporally marked event fragment pool with the fault evolution timing model library to match and generate trigger events that conform to the initial stage of evolution.

[0017] The fault evolution hypothesis generation module generates fault evolution hypotheses to be verified based on the fault evolution path information contained in the triggering events that conform to the initial stage of evolution.

[0018] The predictive enhanced monitoring activation module activates the predictive enhanced monitoring mode based on the fault evolution hypothesis to be verified.

[0019] The subsequent evolution signal directional acquisition module, in predictive enhanced monitoring mode, directionally acquires subsequent evolution signals.

[0020] The fault evidence chain construction module verifies the spatiotemporal correlation between subsequent evolution signals and triggering events that conform to the initial stage of evolution, in order to construct a complete fault evidence chain.

[0021] The fault diagnosis report generation module generates and outputs a precisely located fault diagnosis report based on a complete fault evidence chain.

[0022] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) The present invention establishes a fault evolution time sequence model and actively searches for instantaneous weak signals that match the initial stage of the model, thereby realizing early warning of equipment faults. It moves the fault diagnosis threshold from the macroscopic stage that depends on significant physical phenomena to the nascent stage where only weak signals are associated, greatly improving the foresight and sensitivity of fault diagnosis, and gaining a valuable time window for preventive maintenance, thereby effectively avoiding unplanned downtime and major equipment damage.

[0023] (2) By constructing a complete chain of evidence from the triggering event to the subsequent evolutionary signals, this invention significantly improves the reliability of diagnostic results and reduces the false alarm rate. A single weak signal is only considered as a clue to be verified. The fault is only confirmed when one or more signals predicted by the model are successfully captured under strict spatiotemporal constraints. This diagnostic paradigm based on logical causal verification can effectively filter out random noise interference and ensure that every warning has a high degree of credibility.

[0024] (3) This invention achieves intelligent and efficient utilization of monitoring resources by adopting a predictive enhanced monitoring mode. The system does not perform indiscriminate continuous high-sensitivity scanning of the entire monitoring area, but only dynamically and accurately focuses high-precision detection resources on a very small spatiotemporal range after identifying potential fault initiation points. This strategy of "general survey in peacetime and detailed survey in wartime" greatly reduces the overall data processing burden and operating power consumption of the system without sacrificing the ability to capture key signals, and solves the inherent contradiction between high sensitivity and low cost.

[0025] (4) This invention constructs an adaptive learning closed-loop system by back-optimizing the model library using a verified fault evidence chain after diagnosis. Each successful fault location provides real data feedback from the physical world to the fault evolution model, making its description of fault evolution laws increasingly accurate. This ability to continuously improve itself enables the entire diagnostic system to adapt to factors such as equipment aging and changes in operating conditions, maintaining and continuously improving the accuracy of its fault prediction and location over the long term. Attached Figure Description

[0026] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a schematic diagram of the method steps of the present invention.

[0028] Figure 2 This is a schematic diagram of the system structure connection of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Please see Figure 1 The first aspect of the present invention provides a parallel fault location method for valve hall equipment by deep fusion of multispectral features, including: S1, establishing an evolution time series model library: establishing a fault evolution time series model library.

[0031] In a specific embodiment of the present invention, the specific steps for establishing the fault evolution time series model library include: defining the multi-stage physical process of a potential fault type as structured data containing triggering events and subsequent events.

[0032] Set the allowable time window and spatial location deviation between events for structured data.

[0033] All structured data are collected to form a fault evolution time series model library.

[0034] It's important to note that establishing a fault evolution time-series model library requires collaboration between power equipment experts and data analysis engineers. The first step is a comprehensive review of all typical equipment within the valve hall, such as converter valves, bushings, and insulators, listing all known and theoretically possible potential fault types for each type. The second step involves in-depth analysis of the complete physical process of each potential fault type, from its initial stage to its full development. This analysis breaks down a fault into multiple successive stages, each corresponding to a physical phenomenon observable on a specific spectrum. For example, a flashover fault on an insulator surface might be defined as follows: first, a weak ultraviolet light signal, characteristic of corona discharge; then, in the same area, an infrared sensor detects a slight increase in local temperature; finally, if the fault continues to develop, a visible light sensor captures a noticeable arc. The third step is to convert this multi-stage physical process into structured data that can be processed by a computer. This step is crucial; operators need to define precise parameters for each stage of the event. For the initial triggering event, its spectral type and the minimum signal strength threshold for triggering the alarm need to be clearly defined. For subsequent events, their spectral type and a reasonable range of physical parameters, such as the range of temperature increase, need to be defined. Simultaneously, the spatiotemporal correlation between events needs to be defined; that is, subsequent events must occur within an allowable time window after the occurrence of the preceding event, and the spatial deviation between the subsequent event and the preceding event's location cannot exceed a preset value. The fourth step involves storing a complete structured data record generated for a potential fault type as an independent fault evolution path in the database. The above steps are repeated to create corresponding fault evolution paths for all identified potential fault types. Finally, all these fault evolution paths are aggregated to form a comprehensive database, which is the final output fault evolution time-series model library.

[0035] The fault evolution time-series model library is a dedicated database for storing the development patterns of equipment faults. Its data structure is a collection of multiple fault evolution paths, serving as a preset template for subsequent real-time fault matching. Potential fault types refer to specific fault modes that may occur in the equipment, summarized from equipment design principles, operational experience, and historical data, such as partial discharge on insulator surfaces or overheating of valve control modules. Multi-stage physical processes refer to dividing a complete fault, from its precursors to its full exposure, into multiple sequential and observable independent stages according to the evolution of physical phenomena. Structured data is a standardized data format that encapsulates all information about a multi-stage physical process within a single data record. This record contains clearly defined data fields, such as event spectral type, intensity value, and time window. Its parameter settings are based on statistical analysis of multiple laboratory simulations and numerous field cases. Spectral type refers to different bands of electromagnetic waves; in this method, it specifically refers to visible light, infrared light, and ultraviolet light. An intensity threshold is a specific signal strength criterion set for a triggering event. Only when the signal strength received by the sensor exceeds the intensity threshold is it considered a valid triggering event. Subsequent events refer to signals from other stages that occur sequentially after the triggering event, according to the fault development sequence. A physical parameter range is a valid data interval set for subsequent events. For example, in a subsequent infrared event, the temperature rise must fall between 5 and 15 degrees Celsius to be considered valid. An allowable time window defines the maximum and minimum permissible time interval between two adjacent events, used to confirm the temporal correlation between events. Spatial location deviation defines the maximum permissible spatial distance between the locations of two adjacent events, used to confirm the spatial correlation between events.

[0036] For example, regarding the potential fault type "surface flashover" of an insulator in a valve hall, technicians first define its multi-stage physical process as "an initial ultraviolet signal triggered by corona discharge, followed by an infrared signal generated by localized heating." This process is then transformed into structured data, which defines the spectral type of the triggering event as ultraviolet spectrum with an intensity threshold of 120 photon counts; the spectral type of the subsequent event as infrared spectrum with physical parameters ranging from a temperature increase of 6 to 12 degrees Celsius; and the allowable time window between the two events as 20 to 400 milliseconds, with a spatial deviation within 4 centimeters. This complete structured data constitutes a fault evolution path. Storing this fault evolution path, along with all other fault evolution paths created for other potential fault types such as "valve control module overheating," forms the final fault evolution timing model library.

[0037] S2, Event Fragment Pool Generation: Acquire multispectral data streams and identify transient weak signals from them to generate a spatiotemporally labeled event fragment pool.

[0038] In a specific embodiment of the present invention, the specific steps of acquiring a multispectral data stream and identifying transient weak signals therefrom to generate a spatiotemporally marked event fragment pool include: acquiring a multispectral data stream that is precisely aligned in time and space through an integrated multispectral imaging system with a built-in high-frequency pulse light source.

[0039] Instantaneous weak signals in the multispectral data stream with signal strength below the conventional independent alarm threshold are identified as event fragments.

[0040] Add nanosecond-level timestamps and precise coordinates in three-dimensional space to event fragments to generate a spatiotemporally marked pool of event fragments.

[0041] It should be noted that the process of acquiring and generating a spatiotemporally labeled event fragment pool is first executed through an integrated multispectral imaging system installed within the valve hall, providing comprehensive coverage of the key equipment. This system continuously and synchronously observes the equipment. To achieve precise alignment of data between different spectral sensors, the system integrates a high-frequency pulsed light source, such as an ultraviolet laser diode flashing at a frequency of 10 MHz. Each light pulse emitted by this source serves as an internal synchronization reference signal, distributed to the visible, infrared, and ultraviolet sensors within the system. Image acquisition by each sensor is strictly triggered by the rising edge of this pulse signal, ensuring that at any given moment, the images captured by the three sensors correspond to the exact same instant, achieving nanosecond-level temporal alignment. Simultaneously, during the installation phase, the system underwent rigorous spatial calibration, mapping the pixel coordinate systems of all sensors to the three-dimensional physical space coordinate system of the valve hall equipment, ensuring that any point on the image corresponds to a unique spatial location, achieving millimeter-level spatial alignment. Next, the system's built-in real-time image processor continuously analyzes the multispectral data streams from the three sensors. The processor performs real-time assessments of the signal strength of each pixel. Conventional independent alarm thresholds are preset; for example, the threshold for the infrared channel is a temperature increase of 30 degrees Celsius. The processor's core logic is to search for transient, weak signals whose signal strength is far below this threshold but significantly higher than the background noise. When a signal, such as a point on the infrared channel, experiences a temperature fluctuation of 2 degrees Celsius lasting for several milliseconds, this fluctuation is identified as an event fragment. Finally, once an event fragment is identified, the system immediately packages it and adds two key pieces of information. The first is its nanosecond-level timestamp of the moment of its generation, which comes directly from the system's master clock driving the high-frequency pulse light source. The second is its precise coordinates in three-dimensional space, which are calculated from its pixel position on the sensor using a calibration model. This complete data packet, containing the original signal data, timestamp, and spatial coordinates, is immediately pushed into a dynamically updated data queue, which is the final spatiotemporally tagged event fragment pool.

[0042] The integrated multispectral imaging system is a composite monitoring device that integrates visible light, infrared, and ultraviolet sensors into a single mechanical structure and control system. Its function is to simultaneously acquire image information of the same monitored target under different spectra. A high-frequency pulsed light source is a signal generator built into the imaging system. It generates light pulses at an extremely high and stable frequency, typically set above 10 MHz to meet nanosecond-level synchronization accuracy requirements. The internal synchronization reference is a unified clock signal generated by the high-frequency pulsed light source, used to trigger synchronous acquisition by all sensors. The multispectral data stream is a sequence of unprocessed, continuous raw image data generated in parallel by the visible light, infrared, and ultraviolet sensors. The conventional independent alarm threshold is a preset, relatively high signal strength threshold, set according to the clearly defined fault indication signal standards in industry safety regulations. It is used to distinguish the weak signals of interest in this method from traditionally strong fault signals. A transient weak signal is a signal fluctuation with an extremely short duration and an intensity just exceeding the background noise but far below the conventional independent alarm threshold.

[0043] For example, continuing from the previous example, the integrated multispectral imaging system is continuously monitoring the insulator. At nanosecond 123,456,789, its built-in high-frequency pulsed light source emits a synchronization pulse. At this exact moment, a weak, transient signal with an intensity of 120 photons is captured at pixel (257, 638) of the ultraviolet sensor array. Since this value is below the preset independent alarm threshold of 500 photons, the system identifies it as an event fragment. The system immediately appends a nanosecond-level timestamp to this event fragment, i.e., 123,456,789.000 nanoseconds, and calculates the precise coordinates of the pixel in three-dimensional space using a spatial calibration model: X = 5.21 meters, Y = 4.11 meters, Z = 3.58 meters. This event fragment, containing the ultraviolet signal intensity, timestamp, and spatial coordinates, is immediately generated and pushed into a spatiotemporally tagged event fragment pool for further processing.

[0044] S3. Trigger Event Matching and Generation: The event fragment pool with spatiotemporal tags is compared with the fault evolution timing model library to match and generate trigger events that conform to the initial stage of evolution.

[0045] In a specific embodiment of the present invention, the specific steps of matching and generating triggering events that conform to the initial stage of evolution include: taking the event fragment pool with spatiotemporal marking as real-time input and continuously comparing it with the starting events of all fault evolution paths in the fault evolution time series model library.

[0046] When the spatial location, spectral type, and signal characteristics of an event fragment in the pool match the definition of the starting event of a fault evolution path, the event fragment is locked.

[0047] It should be noted that the process of matching and identifying triggering events that conform to the initial stage of evolution is executed continuously in a dedicated data processing module. This module uses the spatiotemporally labeled event fragment pool generated in the previous step as its dynamic real-time input data source, while simultaneously loading and maintaining static access to the fault evolution time-series model library established in the first step. The processing flow begins by retrieving the latest event fragments one by one from the spatiotemporally labeled event fragment pool in chronological order. For each retrieved event fragment, the module traverses all fault evolution paths stored in the fault evolution time-series model library. During the traversal, it rigorously compares the attributes of the current event fragment with the matching conditions of the starting event defined for each fault evolution path. This comparison process includes three dimensions of verification. First, it verifies whether the spectral type is consistent, that is, the spectral type of the event fragment must be exactly the same as the spectral type defined for the starting event in the model. Second, it verifies whether the signal characteristics meet the requirements, that is, the signal strength of the event fragment must be greater than or equal to the intensity threshold set for the starting event in the model. Third, the system verifies whether the spatial location matches; that is, the three-dimensional spatial coordinates of the event fragment must fall within the three-dimensional geometric model of the device or component associated with the fault evolution path. Only when all three conditions are met simultaneously will the system determine that the current event fragment has successfully matched the starting event of a fault evolution path. Once a match is successful, the system will immediately lock this event fragment from the spatiotemporally marked event fragment pool to prevent it from being processed repeatedly or discarded as ordinary noise.

[0048] The locked event fragments and their associated fault evolution path information are used to generate triggering events that conform to the initial stage of evolution.

[0049] It should be noted that, finally, the system merges all the information carried by the locked event fragment with the information of the complete fault evolution path that it successfully matched, and encapsulates them together into a new structured data object. This new object is the final generated trigger event that conforms to the initial stage of evolution.

[0050] For example, continuing from the previous example, the system retrieves the ultraviolet event fragment from the spatiotemporally labeled event fragment pool. Its spectral type is ultraviolet, its signal characteristic is an intensity of 120 photon counts, and its spatial coordinates are: X=5.21 m, Y=4.11 m, Z=3.58 m. The system then compares this event fragment with the starting events of all fault evolution paths in the fault evolution time-series model library. When comparing to the "surface flashover" fault evolution path, the system finds that its starting event definition requires an ultraviolet spectral type, an intensity threshold of 120 photon counts, and a spatial location on the target insulator. After verification, the spectral type and signal characteristics of the event fragment match the model definition, and its spatial location X=5.21 m, Y=4.11 m, Z=3.58 m also falls exactly within the geometric boundaries of the insulator in the 3D model. Since all conditions are met, the system locks the event fragment and combines it with the complete "surface flashover" fault evolution path information to generate a trigger event that conforms to the initial stage of evolution.

[0051] S4. Fault Evolution Hypothesis Generation: Based on the fault evolution path information contained in the triggering events that conform to the initial stage of evolution, generate fault evolution hypotheses to be verified.

[0052] In a specific embodiment of the present invention, the specific steps for generating the fault evolution hypothesis to be verified include: extracting predefined next-stage event features from the fault evolution path information, including the type of spectrum to be observed, the physical conditions to be met, and the effective time window.

[0053] It should be noted that the process of generating the fault evolution hypothesis to be verified based on the generated triggering event that conforms to the initial stage of evolution is as follows. First, the system receives this triggering event data packet and immediately parses it. The first step is to extract the complete fault evolution path information contained therein. The system will locate the next stage immediately following the current triggering event in this path information and read the predefined event characteristics of that stage. These characteristics include the spectral type of the subsequent signal, such as infrared; an expected range of physical parameters, such as a temperature rise between 6 and 12 degrees Celsius; and an expected time window, such as within 20 to 400 milliseconds after the triggering event occurs. The second step is for the system to extract the three-dimensional spatial coordinates carried by the original event fragments from the same triggering event data packet. These coordinates precisely mark the physical location of the initial event within the valve hall.

[0054] By combining the three-dimensional spatial coordinates of triggering events that conform to the initial stage of evolution, a structured instruction containing four elements is constructed, including the warning location, the type of spectrum to be observed, the physical conditions to be met, and the effective time window.

[0055] Structured instructions are encapsulated to generate fault evolution hypotheses to be verified.

[0056] It should be noted that the system combines the predictive features of the extracted future events with the precise spatial location of the current event extracted in this step to construct a structured instruction containing four core elements. These four elements are: warning location, whose value is the three-dimensional spatial coordinates of the triggering event; the type of spectrum to be observed, whose value is the spectrum of the next stage defined by the model; the physical conditions to be met, whose value is the range of physical parameters for the next stage defined by the model; and the effective time window, whose value is the allowed time period for the next stage event defined by the model. In the third step, the system encapsulates this newly created structured instruction containing four elements into a standardized data object. This data object is the final output of the fault evolution hypothesis to be verified. It is passed to the subsequent monitoring and control module as a temporary, target-specific observation task.

[0057] For example, continuing from the previous step, the system received a trigger event indicating that the insulator's "surface flashover" was in the initial stage of its evolution. The system first extracted the fault evolution path information associated with this event, determining the predefined characteristics of the next stage event: the observed spectral type is infrared, the expected physical parameter range is a temperature increase of 6 to 12 degrees Celsius, and the expected time window is 20 to 400 milliseconds after the trigger event. Next, the system extracted the three-dimensional spatial coordinates of the trigger event: X=5.21 meters, Y=4.11 meters, Z=3.58 meters. Based on these coordinates, a structured instruction was constructed containing four elements: "warning location X=5.21 meters, Y=4.11 meters, Z=3.58 meters, observed spectral type infrared, expected physical condition a temperature increase of 6 to 12 degrees Celsius, and effective time window 20 to 400 milliseconds after the trigger event." Finally, the system encapsulated this structured instruction to generate a fault evolution hypothesis to be verified, ready to guide the next monitoring action.

[0058] S5. Predictive Enhanced Monitoring Activation: Activate the predictive enhanced monitoring mode based on the fault evolution hypothesis to be verified.

[0059] In a specific embodiment of the present invention, the specific steps of activating the predictive enhanced monitoring mode include: parsing the fault evolution hypothesis to be verified, and locking the sensor corresponding to the specified warning location and the type of spectrum to be observed.

[0060] The operating parameters of a specific sensor may be temporarily and selectively increased for a very small local photosensitive area covering the warning location. These operating parameters include, but are not limited to, sampling frequency, detection sensitivity, or noise suppression threshold.

[0061] This enables a localized area of ​​a specific sensor to enter a hypersensitive observation state, forming a predictive enhanced monitoring mode.

[0062] It should be noted that the process of activating the predictive enhanced monitoring mode based on the generated fault evolution hypothesis to be verified is executed by the system's main control unit. First, the main control unit parses the received fault evolution hypothesis data packet to be verified, extracting two key pieces of information: "warning location" and "spectral type to be observed." Based on the spectral type to be observed, such as infrared, the system locks the infrared sensor in the integrated multispectral imaging system as the target. Simultaneously, the system uses a pre-calibrated spatial mapping relationship to accurately convert the three-dimensional physical coordinates of the warning location into a specific pixel region on the infrared sensor's photosensitive chip. This region is typically a very small pixel matrix, such as 5 by 5 pixels; this is the locked, minimal local photosensitive area. Next, the main control unit sends a dedicated configuration command to the driver program of this locked infrared sensor. The core of this command is to temporarily and selectively increase the operating parameters of the previously defined minimal local photosensitive area. The adjustment of operating parameters is determined based on the "physical conditions to be met" in the fault evolution hypothesis to be verified. For example, to capture a slight temperature rise, the command might require increasing the detection sensitivity of that area by two orders of magnitude, or to capture a rapidly changing temperature rise, the command might increase the sampling frequency of that area from the usual 50 frames per second to 500 frames per second. After receiving the command, the sensor's internal firmware adjusts the way it reads and processes the pixels in that local area, putting it into a hypersensitive observation state. In this state, only the designated local area operates with the enhanced parameters, while most of the rest of the sensor maintains its normal monitoring parameters. This achieves high-precision, high-sensitivity control of potential fault points without increasing the overall data burden of the system. The entire process of this specific sensor's local area entering the hypersensitive observation state completes the activation of the predictive enhanced monitoring mode.

[0063] The predictive enhanced monitoring mode is a dynamically adjusted, highly focused system operating state. Its function is to temporarily concentrate system resources within a very small spatiotemporal range based on a specific fault evolution hypothesis, in order to capture subsequent weak signals that are difficult to detect in conventional modes. The minimal local photosensitive area refers to a small pixel region precisely calculated and delineated on the photosensitive element of a specific sensor, which physically corresponds precisely to the warning location. The noise suppression threshold is an internal filtering standard used to distinguish valid signals from background electronic noise; these parameters are set based on the prediction of subsequent signal characteristics in the fault evolution hypothesis to be verified.

[0064] For example, continuing from the previous step, the system receives a hypothesis about the fault evolution generated for the insulator, specifying a warning location at X=5.21 m, Y=4.11 m, Z=3.58 m, and the observed spectrum type as infrared. The system first parses this hypothesis, locks onto the infrared sensor, and calculates that the warning location corresponds to a tiny local photosensitive area of ​​5 x 5 pixels centered at pixel (412,788) on the infrared sensor chip. Subsequently, the system sends a command to the infrared sensor, requesting that the detection sensitivity in its operating parameters be increased by 100 times and the sampling frequency increased from 50 Hz to 500 Hz, specifically for this area. After the infrared sensor executes the command, this 5 x 5 pixel area enters a hypersensitive observation state. Thus, the system successfully activates the predictive enhanced monitoring mode for this potential flashover fault.

[0065] S6. Targeted capture of subsequent evolution signals: In predictive enhanced monitoring mode, targeted capture of subsequent evolution signals.

[0066] In a specific embodiment of the present invention, the specific steps of directional acquisition of subsequent evolution signals include: within the effective time window set by the fault evolution hypothesis to be verified, using the local area of ​​the sensor that has entered the hypersensitive observation state to continuously collect data on the warning location with high sensitivity.

[0067] From the collected high-sensitivity data, we search for signals that match the physical conditions to be met in the fault evolution hypothesis.

[0068] If a signal that meets the criteria is successfully found, its complete data is captured and saved to generate subsequent evolution signals.

[0069] It should be noted that in the predictive enhanced monitoring mode, the process of directionally capturing subsequent evolution signals immediately follows the activation action of the previous step. The system first reads the set effective time window from the fault evolution hypothesis to be verified, for example, starting 20 milliseconds after the trigger event and ending at 400 milliseconds. The system starts a precise timer to ensure that subsequent data acquisition and analysis are strictly performed within this time window. During this period, the extremely small local photosensitive area of ​​the sensor that has previously entered hypersensitive observation state will continuously acquire data from the warning location with its enhanced operating parameters, such as a high frequency of 500 times per second. This generates a highly sensitive data stream with extremely high density, specifically depicting the physical state changes of this tiny area. The system's data processing unit receives and analyzes this data stream in real time. For each frame of acquired data, the processing unit extracts its key physical quantities, such as temperature values, and compares them instantly with the preset "physical conditions to be met" in the fault evolution hypothesis to be verified. This comparison is a continuous search process. If, at any point within the effective time window, the acquired data perfectly matches the required physical conditions—for example, a detected temperature rise of 8.5 degrees Celsius falling within the preset range of 6 to 12 degrees Celsius—the system determines that the target signal has been successfully detected. At this point, the system immediately performs an extraction and saving operation, copying the complete data segment containing the compliant signal, along with several preceding and following frames, from the real-time data stream and packaging it into a single data file. This file is the final generated subsequent evolution signal, containing direct observational evidence of the second stage of fault evolution. If no compliant signal is found by the end of the entire effective time window, the predictive enhancement monitoring mode automatically terminates, and the fault evolution hypothesis to be verified is considered invalid and discarded.

[0070] For example, continuing from the previous example, after activating the predictive enhanced monitoring mode for insulators, the system begins continuous high-sensitivity data acquisition of the warning location within an effective time window of 20 to 400 milliseconds after the occurrence of the ultraviolet-triggered event, utilizing a local area of ​​the infrared sensor in hypersensitive observation mode. At 150 milliseconds, the system detects a temperature rise signal of 8.5 degrees Celsius from the high-sensitivity data acquired from this area. This value perfectly matches the requirement of a 6 to 12 degree Celsius temperature rise in the fault evolution hypothesis to be verified. The system determines the search was successful, immediately captures and saves the complete data recording this temperature rise process, generating a subsequent evolution signal. This signal directly proves that the expected localized heating phenomenon did indeed occur at this location after the initial corona discharge.

[0071] S7. Fault Evidence Chain Construction: Verify the spatiotemporal correlation between subsequent evolution signals and triggering events that conform to the initial stage of evolution in order to construct a complete fault evidence chain.

[0072] In a specific embodiment of the present invention, the specific steps of verifying the spatiotemporal correlation between the subsequent evolution signal and the triggering event that conforms to the initial stage of evolution in order to construct a complete fault evidence chain include: calculating the actual spatial position deviation and time interval between the subsequent evolution signal and the triggering event that conforms to the initial stage of evolution.

[0073] Determine whether the actual spatial location deviation and time interval both fall within the preset allowable range of the fault evolution timing model library.

[0074] Once the judgment is passed, the triggering events that conform to the initial stage of evolution are associated with subsequent evolution signals in chronological order, thereby constructing a complete chain of fault evidence.

[0075] It should be noted that the process of confirming subsequent evolution signals and constructing a complete chain of fault evidence is executed by the system's logic verification module. Upon receiving the subsequent evolution signal captured in the previous step, the module immediately initiates a dual verification procedure. The first step is spatial verification: the module extracts the three-dimensional spatial coordinates of the subsequent evolution signal and compares them with the three-dimensional spatial coordinates of the original triggering event stored in the fault evolution hypothesis to be verified. An actual deviation value is obtained by calculating the straight-line distance between the two coordinate points. Then, the module reads the upper limit of the allowed spatial position deviation preset for this fault evolution path from the associated fault evolution timing model. The spatial verification passes only when the actual deviation value is less than or equal to this upper limit. The second step is temporal verification: the module extracts the timestamp of the subsequent evolution signal and calculates the time difference between it and the timestamp of the original triggering event. This time difference is compared with the allowed time window defined in the fault evolution timing model. The temporal verification passes only when this time difference falls exactly between the start and end times of the allowed time window. The system officially confirms the validity of the fault evolution hypothesis only after both spatial and temporal verifications have successfully passed. Following this, the system performs a correlation operation, arranging the triggering event data packet (the starting point) and subsequent evolutionary signal data packets (evidence of evolution) according to their respective timestamps, forming an ordered event sequence. Finally, the system combines this newly generated event sequence with the complete fault evolution timing model information upon which it is based, encapsulating it into a single, self-contained data record. This record contains both the temporally arranged, verified original signal evidence and the theoretical model used to explain the logical relationships between these pieces of evidence, thus forming a logically closed, irrefutable, and complete fault evidence chain, which is then presented as the final output.

[0076] For example, continuing from the previous example, the system received the subsequent evolution signal of the insulator, namely the infrared signal with a temperature rise of 8.5 degrees Celsius. The system first performs a verification, calculating that the spatial deviation between the location of the infrared signal and the location of the original ultraviolet triggering event is 1.5 cm, which is within the model's allowed deviation range of 4 cm; the spatial verification passes. Next, the system calculates that the infrared signal occurred 150 milliseconds later than the ultraviolet signal, which falls precisely within the model's defined allowable time window of 20 to 400 milliseconds; the temporal verification passes. Since both spatiotemporal conditions pass verification, the system confirms that the fault evolution hypothesis of "surface flashover" is valid. Subsequently, the system correlates the triggering event recording the ultraviolet signal and the subsequent evolution signal recording the infrared signal in chronological order. Finally, this correlation result is combined with the fault evolution timing model information of "surface flashover" to form a logically closed-loop, complete fault evidence chain, providing conclusive early evidence of an impending fault in the insulator.

[0077] S8. Fault Diagnosis Report Generation: Based on a complete chain of fault evidence, generate and output a fault diagnosis report with precise location.

[0078] In a specific embodiment of the present invention, the specific steps for generating and outputting a precisely located fault diagnosis report include: parsing the complete fault evidence chain to extract the fault type and the location of the fault occurrence.

[0079] It should be noted that the process of generating a precise fault diagnosis report in parallel, based on the constructed complete fault evidence chain, is handled by the system's top-level application module. First, upon receiving a complete fault evidence chain, the module immediately parses it. It reads the fault evolution time-series model information contained within, thereby directly extracting the confirmed fault type, such as "insulator surface flashover." Simultaneously, it extracts the three-dimensional spatial coordinates recorded in the triggering event and subsequent evolution signals to determine the precise location of the fault.

[0080] On the 3D visualization model of the valve hall equipment, the location of the fault is highlighted and marked, and all confirmed fault points can be presented in parallel.

[0081] It's important to note that the module then extracts core signal fragments from the raw data of these events, generating a series of visualized raw signal snapshots, such as a miniature image showing an ultraviolet spot and a temperature cloud map showing an infrared hotspot. Next, the module passes the extracted fault location coordinates to a 3D visualization model in the user interface. This 3D visualization model is a pre-built digital twin scene that perfectly matches the actual equipment layout of the valve hall. Upon receiving the coordinates, the visualization engine renders a prominent highlight mark, such as a flashing red warning icon, on the corresponding device and location in the model. A key feature of this method is that if the system confirms faults in multiple different locations within a short period—that is, generates multiple independent and complete chains of fault evidence—all these fault locations will be simultaneously highlighted on the 3D visualization model, achieving parallel presentation of all potential risk points and providing maintenance personnel with a global situational awareness.

[0082] An independent diagnostic report is automatically generated for all highlighted fault points. This report describes the entire process from the discovery of the triggering event to the subsequent enhanced capture of the evolving signal, forming a precisely located fault diagnosis report.

[0083] It's important to note that, finally, the system will automatically launch an independent report generation program for each highlighted fault point. This program will call a preset report template and fill it with all the information parsed from the complete fault evidence chain. The report will detail the entire fault diagnosis process, from how a weak triggering event was initially detected, to how the system generates hypotheses based on the model and activates the predictive enhancement monitoring mode, and finally, how subsequent evolutionary signals were successfully captured for verification. All original signal snapshots will also be attached to the report. This detailed and logically clear document is the final, precisely located fault diagnosis report.

[0084] For example, following the previous example, after receiving the complete fault evidence chain regarding the insulator, the system begins the report output process. First, the system parses the evidence chain, extracting the confirmed fault type as "surface flashover," with an occurrence location of X=5.21 m, Y=4.11 m, Z=3.58 m, and generates two raw signal snapshots: an ultraviolet signal and an infrared signal. Next, the system adds a flashing red marker to the insulator corresponding to coordinates X=5.21 m, Y=4.11 m, Z=3.58 m on the 3D visualization model of the valve hall equipment. At this point, if the system also detects an overheating fault in another converter valve module, two highlighted markers will appear simultaneously on the screen, achieving parallel presentation. Finally, the system automatically generates an independent diagnostic report for this insulator fault point. The report details how the system detected a UV triggering event with a count of 120 photons at 123456789.000 nanoseconds, and how it captured a subsequent evolution signal of a temperature rise of 8.5 degrees Celsius 150 milliseconds later through predictive enhanced monitoring mode, thus confirming the early fault of this surface flashover. This report is the final accurately located fault diagnosis report.

[0085] Reference Figure 2 The second aspect of the present invention provides a parallel fault location system for valve hall equipment with deep fusion of multispectral features, including: an evolutionary time series model library establishment module, an event fragment pool generation module, a trigger event matching generation module, a fault evolution hypothesis generation module, a predictive enhancement monitoring activation module, a subsequent evolution signal directional capture module, a fault evidence chain construction module, and a fault diagnosis report generation module.

[0086] The evolutionary time series model library establishment module and the event fragment pool generation module are both connected to the trigger event matching generation module. The trigger event matching generation module is connected to the fault evolution hypothesis generation module. The fault evolution hypothesis generation module is connected to the predictive enhancement monitoring activation module. The predictive enhancement monitoring activation module is connected to the subsequent evolution signal directional capture module. The trigger event matching generation module and the subsequent evolution signal directional capture module are both connected to the fault evidence chain construction module. The fault evidence chain construction module is connected to the fault diagnosis report generation module.

[0087] The evolution time series model library establishment module establishes a fault evolution time series model library.

[0088] The event fragment pool generation module acquires multispectral data streams and identifies transient weak signals from them to generate a spatiotemporally labeled event fragment pool.

[0089] The trigger event matching and generation module compares the spatiotemporally marked event fragment pool with the fault evolution timing model library to match and generate trigger events that conform to the initial stage of evolution.

[0090] The fault evolution hypothesis generation module generates fault evolution hypotheses to be verified based on the fault evolution path information contained in the triggering events that conform to the initial stage of evolution.

[0091] The predictive enhanced monitoring activation module activates the predictive enhanced monitoring mode based on the fault evolution hypothesis to be verified.

[0092] The subsequent evolution signal directional acquisition module, in predictive enhanced monitoring mode, directionally acquires subsequent evolution signals.

[0093] The fault evidence chain construction module verifies the spatiotemporal correlation between subsequent evolution signals and triggering events that conform to the initial stage of evolution, in order to construct a complete fault evidence chain.

[0094] The fault diagnosis report generation module generates and outputs a precisely located fault diagnosis report based on a complete fault evidence chain.

[0095] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A parallel fault location method for valve hall equipment based on deep fusion of multispectral features, characterized in that, include: S1. Establishment of Evolution Timing Model Library: Establish a fault evolution timing model library; S2, Event Fragment Pool Generation: Acquire multispectral data streams and identify transient weak signals from them to generate a spatiotemporally labeled event fragment pool; S3. Trigger event matching and generation: The event fragment pool with spatiotemporal tags is compared with the fault evolution time series model library to match and generate trigger events that conform to the initial stage of evolution. S4. Fault Evolution Hypothesis Generation: Based on the fault evolution path information contained in the triggering events that conform to the initial stage of evolution, generate fault evolution hypotheses to be verified. S5. Predictive Enhanced Monitoring Activation: Based on the fault evolution hypothesis to be verified, activate the predictive enhanced monitoring mode. S6. Targeted capture of subsequent evolution signals: Under the predictive enhanced monitoring mode, targeted capture of subsequent evolution signals; S7. Fault Evidence Chain Construction: Verify the spatiotemporal correlation between subsequent evolution signals and triggering events that conform to the initial stage of evolution in order to construct a complete fault evidence chain. S8. Fault Diagnosis Report Generation: Based on a complete chain of fault evidence, generate and output a fault diagnosis report with precise location.

2. The parallel fault location method for valve hall equipment based on deep fusion of multispectral features according to claim 1, characterized in that: The specific steps for establishing the fault evolution time series model library include: For potential fault types, their multi-stage physical processes are defined as structured data containing triggering events and subsequent events; Define the allowable time window and spatial location deviation between events for structured data; All structured data are collected to form a fault evolution time series model library.

3. The parallel fault location method for valve hall equipment based on deep fusion of multispectral features according to claim 1, characterized in that: The specific steps for acquiring multispectral data streams and identifying transient weak signals from them to generate a spatiotemporally labeled event fragment pool include: An integrated multispectral imaging system with a built-in high-frequency pulsed light source acquires a multispectral data stream that is precisely aligned in time and space. Identify transient weak signals in the multispectral data stream whose signal strength is below the conventional independent alarm threshold as event fragments; Add nanosecond-level timestamps and precise coordinates in three-dimensional space to event fragments to generate a spatiotemporally marked pool of event fragments.

4. The parallel fault location method for valve hall equipment based on deep fusion of multispectral features according to claim 1, characterized in that: The specific steps for matching and generating triggering events that conform to the initial stage of evolution include: The event fragment pool with spatiotemporal tags is used as real-time input and continuously compared with the starting events of all fault evolution paths in the fault evolution time series model library; When the spatial location, spectral type, and signal characteristics of an event fragment in the pool match the definition of the starting event of a fault evolution path, the event fragment is locked. The locked event fragments and their associated fault evolution path information are used to generate triggering events that conform to the initial stage of evolution.

5. The parallel fault location method for valve hall equipment based on deep fusion of multispectral features according to claim 1, characterized in that: The specific steps for generating the fault evolution hypothesis to be verified include: Extract the predefined next-stage event features from the fault evolution path information, including the type of spectrum to be observed, the physical conditions to be met, and the effective time window; By combining the three-dimensional spatial coordinates of triggering events that conform to the initial stage of evolution, a structured instruction containing four elements is constructed, including the warning location, the type of spectrum to be observed, the physical conditions to be met, and the effective time window. Structured instructions are encapsulated to generate fault evolution hypotheses to be verified.

6. The parallel fault location method for valve hall equipment based on deep fusion of multispectral features according to claim 1, characterized in that: The specific steps of the activation predictive enhancement monitoring mode include: Analyze the fault evolution hypothesis to be verified, and identify the sensor corresponding to the specified warning location and the type of spectrum to be observed; The operating parameters of a sensor may be temporarily and selectively increased only for a very small local photosensitive area covering the warning location. These operating parameters include, but are not limited to, sampling frequency, detection sensitivity, or noise suppression threshold. This enables a localized area of ​​a specific sensor to enter a hypersensitive observation state, forming a predictive enhanced monitoring mode.

7. The parallel fault location method for valve hall equipment based on deep fusion of multispectral features according to claim 5, characterized in that: The specific steps for directional capture of subsequent evolution signals include: Within the effective time window set by the fault evolution hypothesis to be verified, continuous high-sensitivity data acquisition is carried out on the early warning location using the local area of ​​the sensor that has entered the hypersensitive observation state. From the collected high-sensitivity data, search for signals that match the physical conditions to be met in the fault evolution hypothesis; If a signal that meets the criteria is successfully found, its complete data is captured and saved to generate subsequent evolution signals.

8. The parallel fault location method for valve hall equipment based on deep fusion of multispectral features according to claim 1, characterized in that: The specific steps for verifying the spatiotemporal correlation between subsequent evolution signals and triggering events that conform to the initial stage of evolution, in order to construct a complete chain of fault evidence, include: Calculate the actual spatial position deviation and time interval between the subsequent evolution signal and the triggering event that matches the initial stage of evolution; Determine whether the actual spatial location deviation and time interval both fall within the preset allowable range of the fault evolution timing model library; Once the judgment is passed, the triggering events that conform to the initial stage of evolution are associated with subsequent evolution signals in chronological order, thereby constructing a complete chain of fault evidence.

9. The parallel fault location method for valve hall equipment based on deep fusion of multispectral features according to claim 1, characterized in that: The specific steps for generating and outputting a precisely located fault diagnosis report include: Analyze the complete chain of fault evidence to extract the fault type and location; On the 3D visualization model of the valve hall equipment, the location of the fault is highlighted and marked, and all confirmed fault points are displayed in parallel. An independent diagnostic report is automatically generated for all highlighted fault points. This report describes the entire process from the discovery of the triggering event to the subsequent enhanced capture of the evolving signal, forming a precisely located fault diagnosis report.

10. A parallel fault location system for valve hall equipment based on deep fusion of multispectral features, characterized in that, include: The module for establishing an evolution time series model library is used to build a fault evolution time series model library. The event fragment pool generation module acquires multispectral data streams and identifies transient weak signals from them to generate a spatiotemporally labeled event fragment pool; The trigger event matching and generation module compares the spatiotemporally marked event fragment pool with the fault evolution timing model library to match and generate trigger events that conform to the initial stage of evolution. The fault evolution hypothesis generation module generates fault evolution hypotheses to be verified based on the fault evolution path information contained in the triggering events that conform to the initial stage of evolution. The predictive enhanced monitoring activation module activates the predictive enhanced monitoring mode based on the fault evolution hypothesis to be verified. The subsequent evolution signal directional acquisition module, in predictive enhanced monitoring mode, directionally acquires subsequent evolution signals; The fault evidence chain construction module verifies the spatiotemporal correlation between subsequent evolution signals and triggering events that conform to the initial stage of evolution, in order to construct a complete fault evidence chain. The fault diagnosis report generation module generates and outputs a precisely located fault diagnosis report based on a complete fault evidence chain.

Images