Ultra-high frequency partial discharge and temperature monitoring system used in power distribution network system

Abstract

This invention discloses an ultra-high frequency partial discharge and temperature monitoring system for use in power distribution network systems, belonging to the field of smart grid facility condition monitoring technology. The system includes: a signal acquisition module for acquiring mixed signals via a main ultra-high frequency channel and interference reference signals via a mirror channel; an inversion compensation module for establishing interference propagation relationships and progressively reducing interference to obtain candidate pulses; a thermal response construction module for constructing a partial discharge thermal excitation coding sequence and generating an event-triggered thermal response matrix; a thermal memory extraction module for constructing an exponentially decaying thermal memory kernel and extracting thermal memory spectral features; a mapping solution module for inverting heat source parameters and thermal resistance distribution based on a reversible Fourier neural operator model; and an assessment and early warning module for generating the spatial distribution of heat sources, analyzing the defect evolution sequence, and outputting condition assessment and early warning information. This invention combines mirror interference compensation and a reversible Fourier neural operator model to achieve internal defect location and early warning in power distribution facilities.

Description

Technical Field

[0001] This invention relates to the field of smart grid facility condition monitoring technology, and in particular to an ultra-high frequency partial discharge and temperature monitoring system used in power distribution network systems. Background Technology

[0002] Switchgear, cable joints, ring main units, and distribution terminal equipment are all critical facilities on the distribution side of smart grids. The insulation condition, temperature rise, and partial discharge status of these facilities directly affect the operational safety of the distribution network. With the continuous expansion of the distribution network and the sustained increase in equipment operating load, the operational reliability of key electrical equipment such as switchgear, cable joints, and ring main units has a significant impact on power system safety. Partial discharge, as an important early sign of insulation degradation, has become one of the important means of monitoring the condition of power equipment. Meanwhile, temperature monitoring can reflect poor conductor connections, increased contact resistance, and the development of thermal defects. Therefore, existing technologies typically use ultra-high frequency partial discharge detection or temperature monitoring for online assessment of the operating status of distribution equipment. Some solutions also attempt to combine the two to improve fault identification capabilities.

[0003] However, current technologies for detecting UHF partial discharge signals are still susceptible to the influence of complex electromagnetic environments. External wireless signals, corona interference, and switching operations can all introduce strong interference components, leading to insufficient accuracy in partial discharge signal identification. Simultaneously, temperature monitoring is typically treated as an independent parameter for trend analysis, lacking an effective correlation mechanism with partial discharge behavior. Most methods only rely on threshold judgment or simple data aggregation, failing to accurately reflect the actual impact of partial discharge on the internal thermal state of equipment. Existing joint monitoring methods largely fail to consider the temporal discreteness of partial discharge events and their delayed effect on temperature response, resulting in an insufficient characterization of the electro-thermal coupling relationship.

[0004] Existing technologies often struggle to effectively invert the location and heat propagation path of internal defects caused by partial discharge. They cannot infer the equivalent heat source and thermal resistance distribution inside the equipment from externally measured temperature changes. Consequently, it is difficult to achieve spatial location and evolution trend analysis of defects. This results in problems such as insufficient identification accuracy, limited anti-interference capability, and insufficient defect diagnosis depth in complex power distribution environments.

[0005] Therefore, how to provide an ultra-high frequency partial discharge and temperature monitoring system for use in power distribution network systems is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an ultra-high frequency partial discharge and temperature monitoring system for use in power distribution network systems. This invention achieves high-precision extraction of partial discharge signals through a self-calibrated mirror interference channel, analyzes the temperature response using partial discharge event-driven thermal memory phase-locked demodulation technology, and combines a reversible Fourier neural operator model to invert the distribution of internal heat sources and thermal resistance of the equipment, thereby accurately assessing the health status of the equipment and providing early warning of potential faults. It fully utilizes the electro-thermal coupling mechanism between partial discharge and temperature rise, and has the advantages of strong anti-interference ability, high diagnostic accuracy, accurate fault location, and long early warning time.

[0007] According to an embodiment of the present invention, a UHF partial discharge and temperature monitoring system used in a power distribution network system includes: The signal acquisition module acquires a mixed UHF signal containing partial discharge components and external electromagnetic interference components through the main UHF acquisition channel, acquires the corresponding external electromagnetic interference reference signal through the image interference acquisition channel, and performs time synchronization processing. The inversion compensation module performs inversion compensation on the mixed UHF signal based on the interference transmission relationship between the mirror interference acquisition channel and the main UHF acquisition channel to obtain the real partial discharge candidate pulse sequence. The thermal response construction module constructs a partial discharge thermal excitation coding sequence based on the real partial discharge candidate pulse sequence, performs time delay expansion processing on the temperature signals of multiple measuring points of the tested power distribution equipment, generates a time delay temperature rise matrix corresponding to each partial discharge event, and compensates the time delay temperature rise matrix to obtain the event-triggered thermal response matrix. The thermal memory extraction module constructs an exponentially decaying thermal memory kernel based on the partial discharge thermal excitation coding sequence, performs convolution operation on the partial discharge thermal excitation coding sequence and the exponentially decaying thermal memory kernel to generate a theoretical thermal response sequence, and uses the theoretical thermal response sequence as a phase-locked signal to perform correlation demodulation on the event-triggered thermal response matrix to extract thermal memory spectral features. The mapping solution module takes the partial discharge thermal excitation coding sequence, thermal memory spectrum features, event-triggered thermal response matrix and structural parameters of the tested power distribution equipment as input, constructs the forward mapping relationship of heat propagation, performs bidirectional mapping solution based on the reversible Fourier neural operator model, and outputs internal equivalent heat source parameters and thermal resistance distribution parameters. The assessment and early warning module generates status assessment results and fault warning information for the tested power distribution equipment based on the internal equivalent heat source parameters and thermal resistance distribution parameters.

[0008] Optionally, the acquisition of a mixed UHF signal containing partial discharge components and external electromagnetic interference components through the main UHF acquisition channel includes: UHF sensors are installed at preset detection locations in the insulation cavity, conductor connection points, or equipment casing of the power distribution equipment under test. The electromagnetic radiation signal generated by partial discharge is coupled and received with the external electromagnetic interference signal. The received signal is then processed by the UHF front-end conditioning circuit through bandpass filtering, low-noise amplification, and amplitude conditioning to output a mixed UHF electrical signal in a unified format.

[0009] Optionally, the step of acquiring the corresponding external electromagnetic interference reference signal through the mirror interference acquisition channel includes: A reference UHF sensor is installed at a distance from the main UHF acquisition channel. A metal shield is installed on the outside of the reference UHF sensor, and a coupling opening is provided on the side facing the external electromagnetic environment. The reference UHF sensor receives external electromagnetic interference signals, which are then processed by a bandpass filter circuit, an amplification circuit, and a delay correction circuit before outputting an external electromagnetic interference reference signal.

[0010] Optionally, the inversion compensation module includes: The mixed UHF signal output from the main UHF acquisition channel and the external electromagnetic interference reference signal output from the image interference acquisition channel are synchronously segmented. The two signals are divided into multiple corresponding signal segments according to a preset time window, and the arrival time difference, frequency band energy distribution and phase change information of each corresponding signal segment are extracted. Based on the arrival time difference, frequency band energy distribution and phase change information of each corresponding signal segment, a segmented image propagation profile from the image interference acquisition channel to the main UHF acquisition channel is established. Time delay correction, frequency band gain correction and phase compensation are performed on the external electromagnetic interference reference signal in each time window to obtain the image compensation signal corresponding to the interference component in the main UHF acquisition channel. The mixed UHF signal and the image compensation signal are inverted and compensated segment by segment. Within each time window, the interference components are reduced sequentially in the order of coarse compensation followed by fine compensation. The coarse compensation removes the main common-mode interference component, and the fine compensation locally corrects the remaining narrowband pulse interference and transient spike interference to obtain the initial clean signal. The initial clean signal is subjected to partial discharge fidelity constraint processing to retain the pulse component, rising edge abrupt component and short duration broadband pulse component associated with the power frequency phase, and suppress the residual interference component that appears synchronously with the image compensation signal to obtain the real partial discharge enhancement signal. The real partial discharge enhancement signal is subjected to pulse screening and adjacent pulse aggregation to remove pulse segments that do not meet the preset pulse width, phase continuity and energy continuity, and output the real partial discharge candidate pulse sequence.

[0011] Optionally, the thermal response building module includes: Read the real partial discharge candidate pulse sequence, merge the events according to the pulse occurrence time order, phase proximity relationship and energy continuity relationship, divide adjacent pulses into multiple partial discharge events, and record the start time, duration, number of pulses, cumulative energy and phase center of each partial discharge event; Based on the start time, duration, number of pulses, cumulative energy and phase center of each partial discharge event, a partial discharge thermal excitation coding sequence is constructed. Each partial discharge event is mapped to a thermal excitation coding segment at the corresponding time position to form a continuous partial discharge thermal excitation coding sequence. The temperature signals of each temperature measuring point of the tested power distribution equipment are collected synchronously. Taking the start time of each partial discharge event as the trigger reference, the temperature rise segments of each temperature measuring point are continuously extracted after the start time according to the preset time delay sampling interval. The temperature rise segments of each temperature measuring point corresponding to the same partial discharge event are expanded and arranged according to the measuring point order and time delay order to generate the time delay temperature rise matrix corresponding to the partial discharge event. The time-delay temperature rise matrix corresponding to each partial discharge event is subjected to hierarchical compensation processing. First, the overall temperature rise background is eliminated by using ambient temperature measurement data and the steady-state operating temperature baseline of the equipment. Then, the inertial temperature drift component is eliminated based on the short-term temperature change trend before the partial discharge event. Finally, the time-delay temperature rise matrix is ​​subjected to in-window differential compensation based on the temperature segment of the non-discharge pulse period adjacent to the partial discharge event to obtain the event-triggered thermal response matrix.

[0012] Optionally, the thermal memory extraction module includes: Read the partial discharge excitation code sequence and the event-triggered thermal response matrix. According to the duration interval, cumulative pulse number, cumulative pulse energy and the pre-event temperature reference of the corresponding measurement point of each partial discharge event, construct a thermal memory initial unit for each partial discharge event. Arrange the thermal memory initial units in the order of event occurrence to form an event thermal memory unit group. The event thermal memory unit group is subjected to segmented decay fitting processing. The preset short time period after the start time of each partial discharge event is taken as the thermal memory start segment, and the subsequent continuous time delay interval is taken as the thermal memory decay segment. According to the temperature rise decay rate, the order of temperature rise peak appearance and the temperature rise fall slope of the same partial discharge event at different measurement points, the corresponding decay start point, decay rate and decay duration interval are determined respectively, and an exponential decay type thermal memory sub-core corresponding to each partial discharge event is generated. The exponentially decaying thermal memory sub-cores are spliced ​​together by events. The exponentially decaying thermal memory sub-cores with an inter-event splicing process are spliced ​​together by the remaining decay amount of the previous partial discharge event and the initial thermal memory amount of the next partial discharge event. The exponentially decaying thermal memory sub-cores with an inter-event splicing process is spliced ​​together by events with an inter-event splicing process of ... The partial discharge excitation coding sequence and the exponentially decaying thermal memory kernel sequence are processed by event-by-event time-delay mapping to generate a theoretical thermal response sequence. The theoretical thermal response sequence is then expanded into a theoretical thermal response matrix according to the partial discharge event number, the measurement point number, and the time-delay sampling order. The theoretical thermal response matrix and the event-triggered thermal response matrix are matched on an event-by-event, measurement-point-by-time, and time-delay basis to extract the dominant thermal memory time constant, demodulation amplitude, phase lag, and cross-cycle consistency coefficient, thus forming the thermal memory spectrum characteristics.

[0013] Optionally, the mapping solution module includes: Read the partial discharge excitation coding sequence, thermal memory spectrum features and event-triggered thermal response matrix, collect the structural parameters of the power distribution equipment under test, encode the structural parameters according to conductor position, insulation layer thickness, shell boundary position, measurement point spatial position and material thermal diffusion partition, and perform unified temporal and spatial alignment with the partial discharge excitation coding sequence, thermal memory spectrum features and event-triggered thermal response matrix to form bidirectional mapping solution input data; Based on the bidirectional mapping solution of the input data, a forward thermal propagation mapping relationship is constructed. The partial discharge heat excitation code sequence is used as the heat source excitation input, and the conductor position, insulation layer thickness, shell boundary position, measurement point spatial position and material thermal diffusion partition in the structural parameters are used as propagation constraint input. The event-by-event temperature rise delay distribution in the event-triggered thermal response matrix is ​​used as the propagation response output, forming a thermal propagation forward mapping sample sequence from partial discharge heat excitation to measurement point temperature response. Construct a reversible Fourier neural operator model comprising an event heat source embedding layer, a partitioned reversible Fourier propagation layer, and a boundary parameter decoding layer, wherein: The event-based heat source embedding layer jointly encodes the partial discharge thermal excitation sequence, thermal memory spectrum features, and thermal diffusion partitions in the structural parameters to generate a per-event heat source characterization tensor. The partitioned reversible Fourier propagation layer establishes reversible Fourier propagation units according to the conductor region, insulation region and shell region respectively, sets boundary exchange channels between adjacent partitions, and performs frequency domain propagation and reverse reconstruction of the event-by-event heat source characterization tensor; The boundary parameter decoding layer maps the frequency domain propagation results to the event-triggered thermal response matrix and outputs the internal equivalent heat source parameters and thermal resistance distribution parameters. The invertible Fourier neural operator model is solved bidirectionally based on the forward mapping sample sequence of heat propagation. The forward solution generates the predicted temperature response based on the partial discharge heat excitation coding sequence and structural parameters. The inverse solution infers the internal equivalent heat source parameters and thermal resistance distribution parameters based on the event-triggered thermal response matrix, thermal memory spectrum characteristics and structural parameters. The predicted temperature response and the event-triggered thermal response matrix are corrected for each event, each measurement point and each time delay to obtain the bidirectional converged inversion result. The invertible Fourier neural operator model is solved by inputting the online-acquired partial discharge excitation coding sequence, thermal memory spectrum features, event-triggered thermal response matrix and structural parameters to complete the bidirectional mapping solution, and outputs the internal equivalent heat source parameters and thermal resistance distribution parameters.

[0014] Optionally, the assessment and early warning module includes: Read the internal equivalent heat source parameters and thermal resistance distribution parameters, and spatially map the equivalent heat source corresponding to each partial discharge event according to the heat source location, heat source intensity and heat source action depth to form an internal heat source distribution map of the equipment. Based on the thermal resistance variation results of each zone in the thermal resistance distribution parameters, the equivalent boundary range of the defect is matched with the conductor area, insulation area and shell area to obtain the spatial distribution area of ​​the defect inside the equipment, and the boundary expansion direction and expansion amplitude information are extracted. Based on the internal equivalent heat source parameters and thermal resistance distribution parameters, a sequence analysis of the changes in heat source location, heat source intensity, and boundary expansion trend of multiple partial discharge events is performed in chronological order to generate a defect evolution sequence. Based on the defect evolution sequence, the heat source intensity change range, boundary expansion speed and thermal resistance change degree are classified to form the equipment status level classification result. The equipment status level includes normal status, attention status, warning status and fault status. Based on the equipment status level classification results and defect evolution sequence, corresponding fault warning information is generated, and the fault warning information is associated with the internal heat source distribution map and the defect spatial distribution area for output.

[0015] The beneficial effects of this invention are: This invention successfully achieves online condition monitoring of key electrical equipment in power distribution networks by combining ultra-high frequency partial discharge detection and temperature monitoring technologies. Employing a self-calibrated mirror interference acquisition channel and a thermal memory phase-locked demodulation method, it not only effectively eliminates the influence of electromagnetic interference on partial discharge signals but also accurately correlates partial discharge events with temperature rise, providing more accurate data support for the condition assessment of power distribution equipment.

[0016] This invention overcomes the bottleneck of existing technologies that cannot accurately locate fault sources and their evolution trends by introducing a reversible Fourier neural operator model to invert the distribution of heat sources and thermal resistance inside equipment. Through forward mapping of heat propagation and bidirectional inversion calculation, the location of defects inside the equipment and their development trends can be accurately identified, thus providing a strong basis for equipment maintenance decisions.

[0017] This invention not only enhances the multi-parameter fusion diagnostic capabilities for partial discharge and temperature rise, but also improves the accuracy of fault identification and the timeliness of early warning. Compared with existing technologies, this invention has significant advantages in anti-interference capability, fault location capability, and early warning capability, which helps reduce the probability of equipment failure, extend equipment service life, and ensure the safe and stable operation of the power system. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the ultra-high frequency partial discharge and temperature monitoring system for use in a power distribution network system proposed in this invention; Figure 2 This is a schematic diagram of the forward mapping and backward solving of heat propagation based on the reversible Fourier neural operator model for the ultra-high frequency partial discharge and temperature monitoring system used in the power distribution network system proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figure 1 and Figure 2 UHF partial discharge and temperature monitoring systems used in power distribution network systems include: The signal acquisition module acquires a mixed UHF signal containing partial discharge components and external electromagnetic interference components through the main UHF acquisition channel, acquires the corresponding external electromagnetic interference reference signal through the image interference acquisition channel, and performs time synchronization processing. The inversion compensation module performs inversion compensation on the mixed UHF signal based on the interference transmission relationship between the mirror interference acquisition channel and the main UHF acquisition channel to obtain the real partial discharge candidate pulse sequence. The thermal response construction module constructs a partial discharge thermal excitation coding sequence based on the real partial discharge candidate pulse sequence, performs time delay expansion processing on the temperature signals of multiple measuring points of the tested power distribution equipment, generates a time delay temperature rise matrix corresponding to each partial discharge event, and compensates the time delay temperature rise matrix to obtain the event-triggered thermal response matrix. The thermal memory extraction module constructs an exponentially decaying thermal memory kernel based on the partial discharge thermal excitation coding sequence, performs convolution operation on the partial discharge thermal excitation coding sequence and the exponentially decaying thermal memory kernel to generate a theoretical thermal response sequence, and uses the theoretical thermal response sequence as a phase-locked signal to perform correlation demodulation on the event-triggered thermal response matrix to extract thermal memory spectral features. The mapping solution module takes the partial discharge thermal excitation coding sequence, thermal memory spectrum features, event-triggered thermal response matrix and structural parameters of the tested power distribution equipment as input, constructs the forward mapping relationship of heat propagation, performs bidirectional mapping solution based on the reversible Fourier neural operator model, and outputs internal equivalent heat source parameters and thermal resistance distribution parameters. The assessment and early warning module generates status assessment results and fault warning information for the tested power distribution equipment based on the internal equivalent heat source parameters and thermal resistance distribution parameters.

[0021] In this embodiment, the acquisition of a mixed UHF signal containing partial discharge components and external electromagnetic interference components through the main UHF acquisition channel includes: UHF sensors are installed at preset detection locations in the insulation cavity, conductor connection points, or equipment casing of the power distribution equipment under test. The electromagnetic radiation signal generated by partial discharge is coupled and received with the external electromagnetic interference signal. The received signal is then processed by the UHF front-end conditioning circuit through bandpass filtering, low-noise amplification, and amplitude conditioning to output a mixed UHF electrical signal in a unified format.

[0022] In this embodiment, the acquisition of the corresponding external electromagnetic interference reference signal through the mirror interference acquisition channel includes: A reference UHF sensor is installed at a distance from the main UHF acquisition channel. A metal shield is installed on the outside of the reference UHF sensor, and a coupling opening is provided on the side facing the external electromagnetic environment. The reference UHF sensor receives external electromagnetic interference signals, which are then processed by a bandpass filter circuit, an amplification circuit, and a delay correction circuit before outputting an external electromagnetic interference reference signal.

[0023] In this embodiment, the inversion compensation module includes: The mixed UHF signal output from the main UHF acquisition channel and the external electromagnetic interference reference signal output from the image interference acquisition channel are synchronously segmented. The two signals are divided into multiple corresponding signal segments according to a preset time window. The arrival time difference, frequency band energy distribution and phase change information of each corresponding signal segment are extracted. The preset time window is 20μs. Based on the arrival time difference, frequency band energy distribution, and phase change information of each corresponding signal segment, a segmented image propagation profile from the image interference acquisition channel to the main UHF acquisition channel is established. Time delay correction, frequency band gain correction, and phase compensation are performed on the external electromagnetic interference reference signal within each time window to obtain the image compensation signal corresponding to the interference component in the main UHF acquisition channel. Specifically, the segmented image propagation profile from the image interference acquisition channel to the main UHF acquisition channel is established as follows: Based on the arrival time difference between the main UHF acquisition channel signal segment and the external electromagnetic interference reference signal segment within the same time window, the interference propagation time offset corresponding to the time window is determined, and the propagation time offset is used as the time alignment reference to perform initial time alignment of the two signal segments. Based on the energy distribution of each corresponding signal segment within the preset frequency band, the energy ratio of the external electromagnetic interference reference signal segment in each frequency band is calculated and compared with the energy ratio of the corresponding frequency band in the main UHF acquisition channel to determine the gain adjustment coefficient corresponding to each frequency band and form a frequency band gain mapping relationship. Based on the phase change information of each corresponding signal segment, the phase change trend of the external electromagnetic interference reference signal segment in each frequency band is extracted, and the phase change trend of the corresponding frequency band in the main UHF acquisition channel is aligned and analyzed to determine the phase offset of each frequency band and form a phase compensation relationship. The propagation time offset, frequency band gain mapping relationship and phase compensation relationship are combined in the order of time window to construct a segmented mirror propagation profile covering the entire time series, which is used to describe the dynamic change characteristics of external electromagnetic interference propagating from the mirror interference acquisition channel to the main UHF acquisition channel. The mixed UHF signal and the image compensation signal are inverted and compensated segment by segment. Within each time window, interference components are reduced sequentially in the order of coarse compensation followed by fine compensation. Coarse compensation removes the main common-mode interference component, while fine compensation locally corrects the remaining narrowband pulse interference and transient spike interference to obtain the initial cleaned signal. Specifically, the mixed UHF signal and the image compensation signal are inverted and compensated segment by segment as follows: Within each time window, the time-aligned hybrid UHF signal segment and the corresponding mirror-compensated signal segment are subjected to amplitude normalization processing so that the two signals can be compared on the same amplitude scale. Based on the energy distribution of each frequency band in the mirror compensation signal band, the main frequency band component corresponding to the interference component in the main UHF acquisition channel is extracted, and frequency band separation processing is performed on the mixed UHF signal band within the corresponding frequency band range to obtain the frequency band signal containing the main interference component. The corresponding frequency band components in the mirror compensation signal band are mapped one by one to the hybrid ultra-high frequency signal band, and the amplitude cancellation processing is performed on the interference components in each frequency band to form a coarse compensation result. Based on the coarse compensation results, short-time local analysis processing is performed on the remaining signal to identify narrowband pulse interference with short duration and concentrated energy, as well as transient spike interference with abrupt amplitude changes. Local correction components in the mirror compensation signal segment are introduced at the corresponding time positions for compensation to form fine compensation results. The signal segments after coarse and fine compensation within each time window are spliced ​​together in chronological order to obtain the initial purified signal; The initial clean signal undergoes partial discharge fidelity constraint processing, retaining pulse components, rising edge abrupt changes, and short-duration broadband pulse components associated with the power frequency phase, while suppressing residual interference components that occur synchronously with the image compensation signal, thus obtaining a true partial discharge enhancement signal. Specifically, the partial discharge fidelity constraint processing on the initial clean signal involves: Within each time window, the initial purification signal is divided into phases according to the power frequency cycle. The occurrence time of each pulse in the signal is mapped to the power frequency phase interval. Pulse components with phase repeatability and stable phase distribution in adjacent power frequency cycles are selected as candidate partial discharge pulses. The rising edge feature is extracted from each pulse in the initial purification signal. The amplitude change rate and duration of the pulse at the beginning stage are calculated. Pulse components with rapid amplitude transition characteristics and rising edge duration within a preset short time range are retained. Short-time frequency band analysis is performed on the initial purification signal to identify short-duration pulse components with energy distribution in a wide frequency band, and pulse components that meet the broadband characteristics are selected based on the frequency band coverage and energy concentration. Residual signal components that appear synchronously with the mirror compensation signal in time location and exhibit repetitive distribution within multiple time windows are marked, and are identified as residual interference components and suppressed based on the characteristics of phase distribution discreteness, frequency band concentration and duration. The pulse components after phase filtering, rising edge filtering and frequency band filtering are fused and retained, and the signal segment after suppressing residual interference is reconstructed to obtain the real partial discharge enhancement signal. The real partial discharge enhancement signal is subjected to pulse screening and adjacent pulse aggregation to remove pulse segments that do not meet the preset pulse width, phase continuity and energy continuity, and output the real partial discharge candidate pulse sequence, where the preset pulse width is 1μs to 3μs.

[0024] In this embodiment, the thermal response construction module includes: Read the real partial discharge candidate pulse sequence, merge the events according to the pulse occurrence time order, phase proximity relationship and energy continuity relationship, divide adjacent pulses into multiple partial discharge events, and record the start time, duration, number of pulses, cumulative energy and phase center of each partial discharge event; Based on the start time, duration, number of pulses, cumulative energy and phase center of each partial discharge event, a partial discharge thermal excitation coding sequence is constructed. Each partial discharge event is mapped to a thermal excitation coding segment at the corresponding time position to form a continuous partial discharge thermal excitation coding sequence. The coding amplitude is determined by the cumulative energy and the number of pulses, the coding width is determined by the duration, and the coding order is consistent with the order of occurrence of partial discharge events. The temperature signals of each temperature measuring point of the tested power distribution equipment are collected synchronously. Taking the start time of each partial discharge event as the trigger reference, the temperature rise segments of each temperature measuring point are continuously extracted after the start time according to the preset time delay sampling interval. The temperature rise segments of each temperature measuring point corresponding to the same partial discharge event are expanded and arranged according to the measuring point order and time delay order to generate the time delay temperature rise matrix corresponding to the partial discharge event. The time-delay temperature rise matrix corresponding to each partial discharge event is subjected to hierarchical compensation processing. First, the overall temperature rise background is eliminated using ambient temperature measurement data and the steady-state operating temperature baseline of the equipment. Then, the inertial temperature drift component is eliminated based on the short-term temperature change trend before the partial discharge event. Finally, the time-delay temperature rise matrix is ​​subjected to in-window differential compensation based on the temperature segment of the non-discharge pulse period adjacent to the partial discharge event to obtain the event-triggered thermal response matrix. Specifically, the in-window differential compensation of the time-delay temperature rise matrix based on the temperature segment of the non-discharge pulse period adjacent to the partial discharge event is as follows: Before and after the time window corresponding to each partial discharge event, the adjacent time segment where no partial discharge pulse was detected was selected as the reference time period. The temperature change data of each temperature measuring point was extracted according to the time delay sampling interval consistent with the event time window to construct the reference temperature segment. The reference temperature segments are expanded and arranged according to the measurement point order and time delay order to form a reference temperature matrix consistent with the time delay temperature rise matrix structure. The reference temperature matrix is ​​then time-aligned. Perform point-by-point and time-by-time difference operations on the time-delay temperature rise matrix and the reference temperature matrix, and subtract the reference temperature value at the corresponding position from the time-delay temperature rise matrix to eliminate the background temperature change component caused by environmental fluctuations and non-partial discharge factors. The results after differential processing are subjected to continuity constraint processing to retain the temperature rise component that shows a continuous upward or slow decay trend after the partial discharge event is triggered, and the abnormal abrupt change points generated during the differential process are smoothed and corrected to obtain the temperature rise data after differential compensation within the same window.

[0025] In this embodiment, the thermal memory extraction module includes: Read the partial discharge thermal excitation code sequence and the event-triggered thermal response matrix. Based on the duration interval, cumulative pulse count, cumulative pulse energy, and the pre-event temperature baseline of each partial discharge event, construct an initial thermal memory unit for each partial discharge event. Arrange these initial thermal memory units in the order of event occurrence to form an event thermal memory unit group. Specifically, constructing an initial thermal memory unit for each partial discharge event is as follows: Read the start time, end time, duration, cumulative number of pulses and cumulative pulse energy of a single partial discharge event, and use the time interval corresponding to the partial discharge event as the event action interval of the initial unit of thermal memory; Extract temperature data from each temperature measuring point corresponding to a partial discharge event within a preset time before the event occurs, calculate the baseline temperature value and short-term temperature change trend of each temperature measuring point before the event occurs, and use the baseline temperature value and short-term temperature change trend as the temperature background information of the initial unit of thermal memory. The initial thermal excitation intensity corresponding to the partial discharge event is determined based on the cumulative number of pulses and the cumulative pulse energy. The thermal excitation effect width is determined based on the duration. Combining the reference temperature value of each temperature measuring point, the short-term temperature change trend, and the event effect range, a thermal memory initial unit containing the event start position, thermal excitation intensity, thermal excitation effect width, and measuring point temperature background information is established for the partial discharge event. The event thermal memory unit group is subjected to segmented decay fitting processing. A preset short time period after the start time of each partial discharge event is taken as the thermal memory start segment, and the subsequent continuous time delay interval is taken as the thermal memory decay segment. According to the rate of temperature rise decay, the order of temperature rise peak appearance, and the temperature rise fall slope of the same partial discharge event at different measurement points, the corresponding decay start point, decay rate, and decay duration interval are determined respectively, generating an exponential decay type thermal memory sub-core corresponding to each partial discharge event. Specifically, the generation of the exponential decay type thermal memory sub-core corresponding to each partial discharge event is as follows: In the initial unit of thermal memory corresponding to each partial discharge event, a preset short period of time after the start of the partial discharge event is taken as the thermal memory start segment. The temperature rise change data of each temperature measuring point in this time interval is extracted, and the time position when the temperature rise of each measuring point begins to rise significantly is determined as the reference position of the attenuation start point. Within the thermal memory decay period, the temperature rise data of each temperature measuring point is read in chronological order to identify the time position when the temperature rise of each measuring point reaches the peak value. The order of the peak values ​​of each measuring point is used to determine the order of the heat propagation path, and the position corresponding to the measuring point where the peak value appears earliest is used as the priority decay reference node. Based on the temperature rise and fall process after the peak value at each temperature measuring point, the temperature rise and fall trend over time are extracted. The decay rate of each measuring point is calculated based on the temperature rise and fall amplitude and the fall time. Combined with the difference in decay rate between different measuring points, the decay rate is calibrated by region to form a decay rate distribution corresponding to spatial location. Based on the time interval between the temperature rise at each measuring point and the temperature drop from the peak to near the reference temperature, the attenuation duration interval corresponding to each measuring point is determined. The attenuation duration intervals of different measuring points are then aligned in chronological order to form an attenuation time range covering the entire heat propagation process. By combining the attenuation start point, attenuation rate distribution, and attenuation duration interval, and mapping them with the thermal excitation intensity and action range of the corresponding partial discharge event, an exponentially decaying thermal memory subnucleus reflecting the gradual attenuation characteristics of temperature rise over time is constructed. The exponentially decaying thermal memory sub-cores are spliced ​​together by events. The exponentially decaying thermal memory sub-cores with an inter-event splicing process are spliced ​​together by the remaining decay amount of the previous partial discharge event and the initial thermal memory amount of the next partial discharge event. The exponentially decaying thermal memory sub-cores with an inter-event splicing process is spliced ​​together by events with an inter-event splicing process of ... The partial discharge thermal excitation encoding sequence and the exponentially decaying thermal memory kernel sequence are subjected to event-by-event time-delay mapping to generate a theoretical thermal response sequence. The theoretical thermal response sequence is then expanded into a theoretical thermal response matrix according to the partial discharge event number, measurement point number, and time-delay sampling order. Specifically, the generation of the theoretical thermal response sequence is as follows: According to the order of partial discharge events, read the thermal excitation code segment corresponding to each partial discharge event in the partial discharge thermal excitation code sequence one by one, and determine the starting position and effective range of the thermal excitation code segment on the time axis. Within the effective range corresponding to each partial discharge event, the corresponding exponentially decaying thermal memory sub-cores are mapped onto the same time axis according to the time alignment relationship, so that the decay start point of the thermal memory sub-core is aligned with the start time of the partial discharge event. Based on the amplitude and width of the thermal excitation coding segment, the amplitude scaling process is applied to the exponentially decaying thermal memory subcore to make the initial amplitude of the thermal memory subcore consistent with the thermal excitation intensity of the partial discharge event. When multiple partial discharge events overlap in time or have short intervals, the thermal memory sub-nuclei corresponding to adjacent partial discharge events are superimposed on the time axis, and the thermal responses within the overlapping time intervals are accumulated to form a multi-event coupled thermal response result. The thermal response results corresponding to each partial discharge event are spliced ​​and combined in chronological order to obtain a theoretical thermal response sequence covering the complete time series. The theoretical thermal response matrix and the event-triggered thermal response matrix are matched event-by-event, measurement-point-by-measurement, and time-delay-by-time. The dominant thermal memory time constant, demodulation amplitude, phase lag, and cross-cycle consistency coefficient are extracted to form the thermal memory spectrum characteristics. Specifically, the matching process between the theoretical thermal response matrix and the event-triggered thermal response matrix is ​​as follows: Within the time interval corresponding to each partial discharge event, the theoretical thermal response matrix and the event-triggered thermal response matrix are matched one-to-one according to the event sequence number, and the corresponding temperature rise sequence is extracted as the matching object within the same measurement point range. For each measurement point, the theoretical thermal response sequence and the actual thermal response sequence are time-aligned. Taking the start time of the partial discharge event as the reference, the alignment is performed point by point within a preset time delay range to determine the optimal alignment position between the two sequences. At the optimal alignment position, the theoretical thermal response sequence and the actual thermal response sequence are compared time-delay by time to calculate the amplitude difference and trend at each time delay position, and the corresponding demodulation amplitude is determined based on the amplitude difference. Based on the time offset between the theoretical thermal response sequence and the actual thermal response sequence at the peak position and during the fall-off process, the time offset between the two sequences is extracted and converted into a phase lag. Consistency analysis is performed on the matching results corresponding to multiple consecutive partial discharge events. The thermal response morphology, amplitude change and phase hysteresis change of each event at the same measurement point are compared. The matching features that maintain a stable trend in multiple event cycles are extracted, and the corresponding cross-cycle consistency coefficient is calculated. The demodulation amplitude, phase lag, and cross-cycle consistency coefficient obtained from each measurement point during the event-by-event and time-delay matching process are combined to form a thermal memory spectrum feature used to characterize the thermal response of partial discharge.

[0026] In this embodiment, the mapping solution module includes: Read the partial discharge excitation coding sequence, thermal memory spectrum features and event-triggered thermal response matrix, collect the structural parameters of the power distribution equipment under test, encode the structural parameters according to conductor position, insulation layer thickness, shell boundary position, measurement point spatial position and material thermal diffusion partition, and perform unified temporal and spatial alignment with the partial discharge excitation coding sequence, thermal memory spectrum features and event-triggered thermal response matrix to form bidirectional mapping solution input data; A forward thermal propagation mapping relationship is constructed based on the input data obtained through bidirectional mapping. The partial discharge heat excitation code sequence is used as the heat source excitation input, and the conductor position, insulation layer thickness, shell boundary position, measuring point spatial position, and material thermal diffusion partition in the structural parameters are used as propagation constraint inputs. The event-by-event temperature rise delay distribution in the event-triggered thermal response matrix is ​​used as the propagation response output, forming a sample sequence of forward thermal propagation mappings from partial discharge heat excitation to the measuring point temperature response. Specifically, the forward thermal propagation mapping relationship is constructed based on the input data obtained through bidirectional mapping as follows: The partial discharge thermal excitation coding sequence is expanded according to the order of partial discharge events. The thermal excitation intensity, time interval and time position corresponding to each partial discharge event are extracted to construct the thermal source excitation sequence and arrange them according to a unified time axis. Spatial coding is performed on the conductor position, insulation layer thickness, shell boundary position, measuring point spatial position, and material thermal diffusion partition in the structural parameters to establish spatial adjacency relationships and partition correspondence relationships between each structural parameter, forming a set of structural constraints describing the heat propagation path and boundary constraints; The event-triggered thermal response matrix is ​​expanded according to the partial discharge event number, measurement point number, and time delay sampling order. The temperature rise time delay distribution characteristics of each partial discharge event at different measurement points are extracted and aligned according to a unified time axis and spatial location to form a propagation response sequence. Using the heat source excitation sequence corresponding to the partial heat excitation excitation sequence as input, the spatial distribution relationship and heat diffusion partition described in the structural constraint set as propagation constraint conditions, and the propagation response sequence as output, the three are correlated one-to-one according to the event sequence, measurement point location and time delay to construct a mapping relationship that reflects the temperature response generated at the measurement point after the heat source excitation is propagated through the structure. The mapping relationship is continuously organized over multiple partial discharge events and multiple time windows to form a heat propagation forward mapping sample sequence that covers the complete time series and spatial distribution. Construct a reversible Fourier neural operator model comprising an event heat source embedding layer, a partitioned reversible Fourier propagation layer, and a boundary parameter decoding layer, wherein: The event-based heat source embedding layer jointly encodes the partial discharge thermal excitation encoding sequence, thermal memory spectrum features, and thermal diffusion partitions in the structural parameters to generate a per-event heat source representation tensor. Specifically, the generation of the per-event heat source representation tensor involves: Based on the partial discharge thermal excitation coding sequence, the thermal excitation intensity, time interval and time location corresponding to each partial discharge event are extracted, and the time dimension is expanded according to the event sequence to form a sequential thermal excitation feature sequence. Based on the thermal memory spectrum characteristics, the dominant thermal memory time constant, demodulation amplitude, phase lag and cross-cycle consistency coefficient corresponding to each partial discharge event are extracted and aligned with the event-by-event thermal excitation feature sequence in the time dimension to form the event-by-event thermal response feature sequence. Based on the thermal diffusion partitions and the spatial locations of the measurement points in the structural parameters, a spatial partition correspondence is established. The event-by-event thermal excitation feature sequence and the event-by-event thermal response feature sequence are mapped to each thermal diffusion partition. They are then combined according to the time dimension, spatial dimension, and feature dimension to generate an event-by-event heat source characterization tensor. The partitioned reversible Fourier propagation layer establishes reversible Fourier propagation units according to the conductor region, insulation region, and shell region, respectively. Boundary exchange channels are set between adjacent partitions to perform frequency domain propagation and inverse reconstruction of the event-by-event heat source characterization tensor. Specifically, the frequency domain propagation and inverse reconstruction of the event-by-event heat source characterization tensor are as follows: The event-by-event heat source characterization tensor is divided into spatial partitions corresponding to the conductor region, insulation region, and shell region. Frequency domain transformation is performed on the heat source characterization data in each partition to convert the heat propagation information in the time domain and spatial domain into frequency domain expression, thus obtaining the frequency domain characterization data corresponding to each partition. In the frequency domain, the frequency domain characterization data is propagated according to the thermal diffusion characteristics of each partition, so that the thermal source excitation extends along the spatial direction within each partition, and the frequency domain energy is transferred between adjacent partitions through the boundary exchange channel, thereby realizing thermal propagation coupling between different partitions. An inverse transform is performed on the frequency domain characterization data of each partition after frequency domain propagation processing to restore the frequency domain results to the time and spatial domains, resulting in reconstructed thermal response data that includes the propagation results within the partition and the coupling results across the partition. The boundary parameter decoding layer maps the frequency domain propagation results to the event-triggered thermal response matrix, outputting the internal equivalent heat source parameters and thermal resistance distribution parameters. Specifically, the mapping between the frequency domain propagation results and the event-triggered thermal response matrix in the boundary parameter decoding layer is as follows: The reconstructed thermal response data obtained by inverse transformation of the frequency domain propagation results are expanded according to the partial discharge event number, measurement point number and time delay sampling order, and aligned one by one with the event-triggered thermal response matrix in time dimension and spatial location to form corresponding matching data pairs. At each measurement point, the reconstructed thermal response data and the event-triggered thermal response matrix are compared time-delay by time. Based on the differences in amplitude distribution and trend of the two, the response deviation distribution corresponding to each measurement point is determined, and the response deviation distribution is mapped to the corresponding spatial partition boundary position. Based on the distribution of response deviations at the boundaries of each spatial partition, feature information related to changes in heat source intensity and thermal resistance is extracted. The feature information is then combined according to the spatial distribution relationship to output the corresponding internal equivalent heat source parameters and thermal resistance distribution parameters. A bidirectional mapping solution is performed on the reversible Fourier neural operator model based on the forward mapping sample sequence of heat propagation. The forward solution generates the predicted temperature response based on the partial discharge heat excitation encoding sequence and structural parameters. The inverse solution infers the internal equivalent heat source parameters and thermal resistance distribution parameters based on the event-triggered thermal response matrix, thermal memory spectrum characteristics, and structural parameters. The predicted temperature response and the event-triggered thermal response matrix are then corrected event-by-event, point-by-point, and time-delay-by-time, yielding the bidirectional converged inversion result. The forward solution generates the predicted temperature response based on the local heat release excitation coding sequence and structural parameters, specifically as follows: Based on the partial discharge excitation coding sequence, the thermal excitation intensity, time interval and time location corresponding to each partial discharge event are extracted. Combined with the conductor position, insulation layer thickness, shell boundary position and thermal diffusion partition information in the structural parameters, the correspondence between heat source input and spatial propagation constraints is constructed. The heat source input and spatial propagation constraints are input into the reversible Fourier neural operator model. The heat source excitation is processed by spatial diffusion propagation through the partitioned reversible Fourier propagation layer to generate the temperature change results of each measuring point under different time delays, forming a predicted temperature response sequence. The reverse solution deduces the internal equivalent heat source parameters and thermal resistance distribution parameters based on the event-triggered thermal response matrix, thermal memory spectrum characteristics, and structural parameters. Specifically: Based on the event-triggered thermal response matrix, the temperature rise time delay distribution characteristics of each partial discharge event at different measurement points are extracted. Combined with the dominant thermal memory time constant, demodulation amplitude, and phase lag in the thermal memory spectrum characteristics, the correspondence between temperature response and heat source characteristics is constructed. The temperature response characteristics and the spatial partitioning and boundary constraint information in the structural parameters are input into the reversible Fourier neural operator model. The heat propagation path is reversed through the reverse propagation process to determine the corresponding heat source intensity distribution and thermal resistance change at the boundary in each spatial partition. The internal equivalent heat source parameters and thermal resistance distribution parameters are output. The invertible Fourier neural operator model is solved by inputting the online-acquired partial discharge excitation coding sequence, thermal memory spectrum features, event-triggered thermal response matrix and structural parameters to complete the bidirectional mapping solution, and outputs the internal equivalent heat source parameters and thermal resistance distribution parameters.

[0027] In this embodiment, the assessment and early warning module includes: Read the internal equivalent heat source parameters and thermal resistance distribution parameters, and spatially map the equivalent heat sources corresponding to each partial discharge event according to the heat source location, heat source intensity, and heat source depth to form an internal heat source distribution map of the equipment. Specifically, the formation of the internal heat source distribution map is as follows: Based on the information of heat source location, heat source intensity and heat source action depth corresponding to each partial discharge event in the internal equivalent heat source parameters, the spatial distribution of each equivalent heat source in the conductor region, insulation region and shell region is determined, and the corresponding spatial coordinate relationship is established. The positions of each equivalent heat source in the spatial coordinates are mapped according to the spatial location of the measuring point and the heat diffusion partition in the structural parameters. The heat source intensity within the same spatial partition is summarized, and the different levels of positions are marked according to the depth of heat source action to form partitioned and layered heat source distribution data. The heat source intensities corresponding to each spatial partition and layer location are arranged according to a unified spatial grid. Each grid location is assigned a corresponding heat source intensity value, and the grids are combined according to the spatial location order to generate an internal heat source distribution map that reflects the heat source distribution in each area of ​​the equipment. Based on the thermal resistance variation results of each zone in the thermal resistance distribution parameters, the equivalent boundary range of the defect is matched with the conductor region, insulation region, and shell region to obtain the spatial distribution area of ​​the defect inside the equipment. The boundary expansion direction and expansion amplitude information are then extracted. Specifically, the spatial distribution area of ​​the defect inside the equipment is as follows: Based on the thermal resistance change results of each thermal diffusion zone in the thermal resistance distribution parameters, spatial zones with significantly increased thermal resistance relative to the steady-state baseline are screened out and marked as candidate abnormal regions. Candidate anomaly regions are matched according to the spatial division of conductor region, insulation region and shell region. Combined with the spatial adjacency relationship in structural parameters, adjacent and continuously distributed candidate anomaly regions are merged to form connected defect candidate regions. The boundaries of the candidate defect regions are identified according to their spatial location inside the equipment. The boundary range of each candidate defect region is determined. The heat source intensity distribution within the region is verified in conjunction with the heat source distribution map. Regions that simultaneously meet the characteristics of abnormal thermal resistance and concentrated heat source distribution are retained as the spatial distribution regions of defects inside the equipment. Based on the internal equivalent heat source parameters and thermal resistance distribution parameters, a sequential analysis of the changes in heat source location, heat source intensity, and boundary expansion trends of multiple partial discharge events is performed in chronological order to generate a defect evolution sequence. Specifically, the generated defect evolution sequence is as follows: According to the time sequence of partial discharge events, the internal equivalent heat source parameters and thermal resistance distribution parameters corresponding to each event are read, the heat source location, heat source intensity and defect boundary range information of each event are extracted, and the event occurrence time is used as the sequence index to form an initial time series data sequence. The changes in the position of the heat source between adjacent partial discharge events are compared, the displacement of the heat source in spatial coordinates is calculated, and the migration path and intensity change trend of the heat source in the time series are determined by combining the heat source intensity change corresponding to each event. The defect boundary range corresponding to each event is compared, and the changes in the boundary expansion direction and expansion magnitude are extracted. The heat source location change sequence, heat source intensity change sequence, and boundary expansion change sequence are combined in chronological order to synchronously correlate the three types of change information, forming a defect evolution sequence that reflects the evolution of the defect in spatial location, intensity, and boundary range over time. Based on the defect evolution sequence, the changes in heat source intensity, boundary expansion rate, and thermal resistance are classified to form a device status level classification result. The device status levels include normal state, concern state, warning state, and fault state. Specifically, the formation of the device status level classification result is as follows: Based on the changes in heat source intensity, boundary expansion rate, and thermal resistance at each time point in the defect evolution sequence, the three types of indicators are quantified and normalized according to a unified scale to obtain the set of indicator values ​​corresponding to each time point. Based on the preset grading threshold range, the set of indicator values ​​at each time node is judged in intervals. When the change range of heat source intensity, the rate of boundary expansion, and the degree of change of thermal resistance are all in the low change range, it is judged as normal. When a single indicator enters the medium change range, it is judged as a state of concern. When two or more indicators enter the high change range, it is judged as a state of warning. When all three indicators enter the high change range, it is judged as a state of fault. The status determination results of each time node are arranged in chronological order and corrected by combining the status continuity between adjacent time nodes to obtain the equipment status level classification result that reflects the change of equipment operating status over time. Based on the equipment status level classification results and defect evolution sequence, corresponding fault warning information is generated. This fault warning information is then correlated with the internal heat source distribution map and the spatial distribution area of ​​the defects and output accordingly. Specifically, the generated fault warning information includes: Based on the equipment status level classification results, extract the time nodes corresponding to the attention status, warning status and fault status, read the heat source location, heat source intensity and boundary expansion information of the corresponding time node in the defect evolution sequence, and determine the time location and abnormal characteristics of the warning trigger. By combining the changes in heat source intensity, boundary expansion rate, and thermal resistance at each time point in the defect evolution sequence, the abnormal development trend is determined, the defect development direction and development rate information is extracted, and the fault risk is classified and marked according to the development rate. By combining information such as early warning trigger time, defect spatial location, heat source intensity level, boundary expansion range, and development trend, a fault early warning information that includes time information, spatial information, and trend information is formed.

[0028] Example 1: To verify the feasibility of this invention in practice, it was applied to a city's power distribution network. A 10kV ring main unit that had been in operation for approximately 6 years was selected as the test object. This equipment had been operating under medium load conditions for a long time, with the daily load rate fluctuating between approximately 60% and 85%. The operating environment was located in a mixed commercial and residential area, surrounded by communication base stations, frequency converters, and electric vehicle charging facilities, resulting in a complex electromagnetic environment. In actual operation, traditional UHF partial discharge detection systems are easily affected by external electromagnetic interference in this environment, leading to false triggering and missed detections. Existing temperature monitoring systems mainly rely on threshold judgments, which are difficult to accurately reflect the impact of partial discharge on the internal thermal state and cannot effectively invert the defect location.

[0029] In practical applications, main UHF acquisition channels are deployed in the cable joint area, busbar connection area, and switch contact area of ​​the ring main unit, while a mirror interference acquisition channel is set up on the outside of the unit. The main channel is used to acquire mixed UHF signals containing partial discharge components and external electromagnetic interference components, while the mirror channel is used to acquire external electromagnetic interference reference signals. The two signals are synchronized using a unified clock.

[0030] During system operation, the hybrid UHF signal and the interfering reference signal are processed in segments according to fixed time windows. Within each time window, arrival time difference, frequency band energy distribution, and phase change information are extracted. Based on this information, the image interference propagation relationship is established, and time delay correction and frequency band compensation are performed on the interfering reference signal to obtain the image compensation signal. Subsequently, the image compensation signal and the hybrid UHF signal are inverted and compensated segment by segment. First, common-mode interference is removed, and then residual narrowband interference is locally corrected to obtain the purified partial discharge signal. Through pulse screening and aggregation processing, the true partial discharge candidate pulse sequence is obtained.

[0031] After obtaining the partial discharge pulse, pulses that are temporally adjacent and have continuous energy are grouped into the same partial discharge event, and the duration, number of pulses, and cumulative energy of the event are recorded. A partial discharge thermal excitation coding sequence is constructed based on these parameters. Temperature measurement points are deployed inside the equipment, and temperature data from each point is collected. Using the time of the partial discharge event as a reference, temperature rise data within the corresponding time window is extracted and expanded according to the measurement point and time delay to form a time-delay temperature rise matrix. Through ambient temperature compensation and short-term trend elimination processing, the event-triggered thermal response matrix is ​​obtained.

[0032] An exponentially decaying thermal memory kernel is constructed based on each partial discharge event. The partial discharge thermal excitation encoding sequence is mapped in the time domain to the thermal memory kernel to obtain the theoretical thermal response sequence. This sequence is then matched with the actual thermal response matrix to extract the thermal memory spectrum features, which can reflect the delay relationship and decay law between the partial discharge event and the temperature response.

[0033] In the inversion solution stage, the partial discharge thermal excitation encoding sequence, thermal memory spectrum characteristics, event-triggered thermal response matrix, and equipment structural parameters are input into the reversible Fourier neural operator model. The model is partitioned according to the heat propagation characteristics of the conductor region, insulation region, and shell region, and the inversion is achieved from the temperature response to the internal heat source and thermal resistance distribution through bidirectional mapping. Finally, the equivalent heat source location and thermal resistance variation distribution inside the equipment are output.

[0034] Table 1. Comprehensive Comparison of the Effects of Joint Monitoring of Partial Discharge and Temperature As shown in Table 1, the present invention exhibits more stable anti-interference capabilities in complex electromagnetic environments in terms of signal quality and partial discharge identification. The average signal-to-noise ratio increased from 11.8 dB to 16.2 dB, indicating that the external electromagnetic interference components were effectively weakened through the mirror interference acquisition channel and inversion compensation mechanism, making the partial discharge characteristics more prominent. The number of false interference triggers decreased from 12-18 times per day to 4-7 times, indicating that the system's ability to suppress non-partial discharge signals was significantly enhanced. On this basis, the effective partial discharge identification rate increased from 78.5% to 91.3%, while the pulse false rejection rate decreased from 9.2% to 4.8%, indicating that under the synergistic effect of interference suppression and fidelity constraints, not only was the identification accuracy improved, but the false loss of real partial discharge signals was also reduced, thus improving the overall reliability of partial discharge detection.

[0035] From the perspective of electro-thermal coupling analysis capabilities, this invention achieves a precise correlation between partial discharge events and temperature rise responses through partial discharge thermal excitation encoding and thermal memory phase-locked demodulation mechanisms. The temperature rise response matching accuracy improved from 72.6% to 88.7%, indicating that the system can more accurately identify temperature rise changes caused by partial discharge. The temperature rise response delay error decreased from 9.5 seconds to 3.6 seconds, demonstrating that thermal response modeling and time delay decomposition processing more closely approximate the actual heat propagation process. Regarding defect location, the location error decreased from 35–50 cm to 12–20 cm, and the defect identification success rate increased from 74.2% to 89.5%, indicating that through forward mapping and inversion solving of thermal propagation, the internal heat source and thermal resistance distribution of the equipment can be more accurately recovered, achieving refined location of defects.

[0036] From the perspectives of early warning capability and overall system performance, this invention improves fault early warning capability while ensuring real-time performance. The average early warning time has increased from 10-18 hours to 30-42 hours, indicating that the system can identify potential anomalies in the early stages of defects, providing more time for maintenance and operation. In terms of system performance, the single data processing time has decreased from 95ms to 82ms, indicating that the adopted processing flow and model structure have good computational efficiency; the continuous operation stability rate has increased from 97.8% to 99.1%, indicating that the system has high reliability and stability during long-term online operation. This invention achieves a relatively balanced and engineering-feasible improvement in anti-interference capability, analysis accuracy, and early warning timeliness.

[0037] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A UHF partial discharge and temperature monitoring system used in a power distribution network system, characterized in that, include: The signal acquisition module acquires a mixed UHF signal containing partial discharge components and external electromagnetic interference components through the main UHF acquisition channel, acquires the corresponding external electromagnetic interference reference signal through the image interference acquisition channel, and performs time synchronization processing. The inversion compensation module performs inversion compensation on the mixed UHF signal based on the interference transmission relationship between the mirror interference acquisition channel and the main UHF acquisition channel to obtain the real partial discharge candidate pulse sequence. The thermal response construction module constructs a partial discharge thermal excitation coding sequence based on the real partial discharge candidate pulse sequence, performs time delay expansion processing on the temperature signals of multiple measuring points of the tested power distribution equipment, generates a time delay temperature rise matrix corresponding to each partial discharge event, and compensates the time delay temperature rise matrix to obtain the event-triggered thermal response matrix. The thermal memory extraction module constructs an exponentially decaying thermal memory kernel based on the partial discharge thermal excitation coding sequence, performs convolution operation on the partial discharge thermal excitation coding sequence and the exponentially decaying thermal memory kernel to generate a theoretical thermal response sequence, and uses the theoretical thermal response sequence as a phase-locked signal to perform correlation demodulation on the event-triggered thermal response matrix to extract thermal memory spectral features. The mapping solution module takes the partial discharge thermal excitation coding sequence, thermal memory spectrum features, event-triggered thermal response matrix and structural parameters of the tested power distribution equipment as input, constructs the forward mapping relationship of heat propagation, performs bidirectional mapping solution based on the reversible Fourier neural operator model, and outputs internal equivalent heat source parameters and thermal resistance distribution parameters. The assessment and early warning module generates status assessment results and fault warning information for the tested power distribution equipment based on the internal equivalent heat source parameters and thermal resistance distribution parameters.

2. The ultra-high frequency partial discharge and temperature monitoring system used in a power distribution network system according to claim 1, characterized in that, The acquisition of a mixed UHF signal containing partial discharge components and external electromagnetic interference components through the main UHF acquisition channel includes: UHF sensors are installed at preset detection locations in the insulation cavity, conductor connection points, or equipment casing of the power distribution equipment under test. The electromagnetic radiation signal generated by partial discharge is coupled and received with the external electromagnetic interference signal. The received signal is then processed by the UHF front-end conditioning circuit through bandpass filtering, low-noise amplification, and amplitude conditioning to output a mixed UHF electrical signal in a unified format.

3. The ultra-high frequency partial discharge and temperature monitoring system used in a power distribution network system according to claim 1, characterized in that, The acquisition of the corresponding external electromagnetic interference reference signal through the mirror interference acquisition channel includes: A reference UHF sensor is installed at a distance from the main UHF acquisition channel. A metal shield is installed on the outside of the reference UHF sensor, and a coupling opening is provided on the side facing the external electromagnetic environment. The reference UHF sensor receives external electromagnetic interference signals, which are then processed by a bandpass filter circuit, an amplification circuit, and a delay correction circuit before outputting an external electromagnetic interference reference signal.

4. The ultra-high frequency partial discharge and temperature monitoring system used in a power distribution network system according to claim 1, characterized in that, The inversion compensation module includes: The mixed UHF signal output from the main UHF acquisition channel and the external electromagnetic interference reference signal output from the image interference acquisition channel are synchronously segmented. The two signals are divided into multiple corresponding signal segments according to a preset time window, and the arrival time difference, frequency band energy distribution and phase change information of each corresponding signal segment are extracted. Based on the arrival time difference, frequency band energy distribution and phase change information of each corresponding signal segment, a segmented image propagation profile from the image interference acquisition channel to the main UHF acquisition channel is established. Time delay correction, frequency band gain correction and phase compensation are performed on the external electromagnetic interference reference signal in each time window to obtain the image compensation signal corresponding to the interference component in the main UHF acquisition channel. The mixed UHF signal and the image compensation signal are inverted and compensated segment by segment. Within each time window, the interference components are reduced sequentially in the order of coarse compensation followed by fine compensation. The coarse compensation removes the main common-mode interference component, and the fine compensation locally corrects the remaining narrowband pulse interference and transient spike interference to obtain the initial clean signal. The initial clean signal is subjected to partial discharge fidelity constraint processing to retain the pulse component, rising edge abrupt component and short duration broadband pulse component associated with the power frequency phase, and suppress the residual interference component that appears synchronously with the image compensation signal to obtain the real partial discharge enhancement signal. The real partial discharge enhancement signal is subjected to pulse screening and adjacent pulse aggregation to remove pulse segments that do not meet the preset pulse width, phase continuity and energy continuity, and output the real partial discharge candidate pulse sequence.

5. The ultra-high frequency partial discharge and temperature monitoring system used in a power distribution network system according to claim 1, characterized in that, The thermal response construction module includes: Read the real partial discharge candidate pulse sequence, merge the events according to the pulse occurrence time order, phase proximity relationship and energy continuity relationship, divide adjacent pulses into multiple partial discharge events, and record the start time, duration, number of pulses, cumulative energy and phase center of each partial discharge event; Based on the start time, duration, number of pulses, cumulative energy and phase center of each partial discharge event, a partial discharge thermal excitation coding sequence is constructed. Each partial discharge event is mapped to a thermal excitation coding segment at the corresponding time position to form a continuous partial discharge thermal excitation coding sequence. The temperature signals of each temperature measuring point of the tested power distribution equipment are collected synchronously. Taking the start time of each partial discharge event as the trigger reference, the temperature rise segments of each temperature measuring point are continuously extracted after the start time according to the preset time delay sampling interval. The temperature rise segments of each temperature measuring point corresponding to the same partial discharge event are expanded and arranged according to the measuring point order and time delay order to generate the time delay temperature rise matrix corresponding to the partial discharge event. The time-delay temperature rise matrix corresponding to each partial discharge event is subjected to hierarchical compensation processing. First, the overall temperature rise background is eliminated by using ambient temperature measurement data and the steady-state operating temperature baseline of the equipment. Then, the inertial temperature drift component is eliminated based on the short-term temperature change trend before the partial discharge event. Finally, the time-delay temperature rise matrix is ​​subjected to in-window differential compensation based on the temperature segment of the non-discharge pulse period adjacent to the partial discharge event to obtain the event-triggered thermal response matrix.

6. The ultra-high frequency partial discharge and temperature monitoring system used in a power distribution network system according to claim 1, characterized in that, The thermal memory extraction module includes: Read the partial discharge excitation code sequence and the event-triggered thermal response matrix. According to the duration interval, cumulative pulse number, cumulative pulse energy and the pre-event temperature reference of the corresponding measurement point of each partial discharge event, construct a thermal memory initial unit for each partial discharge event. Arrange the thermal memory initial units in the order of event occurrence to form an event thermal memory unit group. The event thermal memory unit group is subjected to segmented decay fitting processing. The preset short time period after the start time of each partial discharge event is taken as the thermal memory start segment, and the subsequent continuous time delay interval is taken as the thermal memory decay segment. According to the temperature rise decay rate, the order of temperature rise peak appearance and the temperature rise fall slope of the same partial discharge event at different measurement points, the corresponding decay start point, decay rate and decay duration interval are determined respectively, and an exponential decay type thermal memory sub-core corresponding to each partial discharge event is generated. The exponentially decaying thermal memory sub-cores are spliced ​​together by events. The exponentially decaying thermal memory sub-cores with an inter-event splicing process are spliced ​​together by the remaining decay amount of the previous partial discharge event and the initial thermal memory amount of the next partial discharge event. The exponentially decaying thermal memory sub-cores with an inter-event splicing process is spliced ​​together by events with an inter-event splicing process of ... The partial discharge excitation coding sequence and the exponentially decaying thermal memory kernel sequence are processed by event-by-event time-delay mapping to generate a theoretical thermal response sequence. The theoretical thermal response sequence is then expanded into a theoretical thermal response matrix according to the partial discharge event number, the measurement point number, and the time-delay sampling order. The theoretical thermal response matrix and the event-triggered thermal response matrix are matched on an event-by-event, measurement-point-by-time, and time-delay basis to extract the dominant thermal memory time constant, demodulation amplitude, phase lag, and cross-cycle consistency coefficient, thus forming the thermal memory spectrum characteristics.

7. The ultra-high frequency partial discharge and temperature monitoring system used in a power distribution network system according to claim 1, characterized in that, The mapping solution module includes: Read the partial discharge excitation coding sequence, thermal memory spectrum features and event-triggered thermal response matrix, collect the structural parameters of the power distribution equipment under test, encode the structural parameters according to conductor position, insulation layer thickness, shell boundary position, measurement point spatial position and material thermal diffusion partition, and perform unified temporal and spatial alignment with the partial discharge excitation coding sequence, thermal memory spectrum features and event-triggered thermal response matrix to form bidirectional mapping solution input data; Based on the bidirectional mapping solution of the input data, a forward thermal propagation mapping relationship is constructed. The partial discharge heat excitation code sequence is used as the heat source excitation input, and the conductor position, insulation layer thickness, shell boundary position, measurement point spatial position and material thermal diffusion partition in the structural parameters are used as propagation constraint input. The event-by-event temperature rise delay distribution in the event-triggered thermal response matrix is ​​used as the propagation response output, forming a thermal propagation forward mapping sample sequence from partial discharge heat excitation to measurement point temperature response. Construct a reversible Fourier neural operator model comprising an event heat source embedding layer, a partitioned reversible Fourier propagation layer, and a boundary parameter decoding layer, wherein: The event-based heat source embedding layer jointly encodes the partial discharge thermal excitation sequence, thermal memory spectrum features, and thermal diffusion partitions in the structural parameters to generate a per-event heat source characterization tensor. The partitioned reversible Fourier propagation layer establishes reversible Fourier propagation units according to the conductor region, insulation region and shell region respectively, sets boundary exchange channels between adjacent partitions, and performs frequency domain propagation and reverse reconstruction of the event-by-event heat source characterization tensor; The boundary parameter decoding layer maps the frequency domain propagation results to the event-triggered thermal response matrix and outputs the internal equivalent heat source parameters and thermal resistance distribution parameters. The invertible Fourier neural operator model is solved bidirectionally based on the forward mapping sample sequence of heat propagation. The forward solution generates the predicted temperature response based on the partial discharge heat excitation coding sequence and structural parameters. The inverse solution infers the internal equivalent heat source parameters and thermal resistance distribution parameters based on the event-triggered thermal response matrix, thermal memory spectrum characteristics and structural parameters. The predicted temperature response and the event-triggered thermal response matrix are corrected for each event, each measurement point and each time delay to obtain the bidirectional converged inversion result. The invertible Fourier neural operator model is solved by inputting the online-acquired partial discharge excitation coding sequence, thermal memory spectrum features, event-triggered thermal response matrix and structural parameters to complete the bidirectional mapping solution, and outputs the internal equivalent heat source parameters and thermal resistance distribution parameters.

8. The ultra-high frequency partial discharge and temperature monitoring system used in a power distribution network system according to claim 1, characterized in that, The assessment and early warning module includes: Read the internal equivalent heat source parameters and thermal resistance distribution parameters, and spatially map the equivalent heat source corresponding to each partial discharge event according to the heat source location, heat source intensity and heat source action depth to form an internal heat source distribution map of the equipment. Based on the thermal resistance variation results of each zone in the thermal resistance distribution parameters, the equivalent boundary range of the defect is matched with the conductor area, insulation area and shell area to obtain the spatial distribution area of ​​the defect inside the equipment, and the boundary expansion direction and expansion amplitude information are extracted. Based on the internal equivalent heat source parameters and thermal resistance distribution parameters, a sequence analysis of the changes in heat source location, heat source intensity, and boundary expansion trend of multiple partial discharge events is performed in chronological order to generate a defect evolution sequence. Based on the defect evolution sequence, the heat source intensity change range, boundary expansion speed and thermal resistance change degree are classified to form the equipment status level classification result. The equipment status level includes normal status, attention status, warning status and fault status. Based on the equipment status level classification results and defect evolution sequence, corresponding fault warning information is generated, and the fault warning information is associated with the internal heat source distribution map and the defect spatial distribution area for output.

Images