Phototube high-voltage excitation type solar blind ultraviolet discharge detection method and system

Abstract

The invention relates to the technical field of photoelectric tube detection, in particular to a photoelectric tube high-voltage excitation type solar blind ultraviolet discharge detection method and system, and the method comprises the steps: obtaining original voltage sequence data collected by a high-speed transimpedance sampling circuit in real time; positioning feature points of each suspected pulse according to the original voltage sequence data, and extracting pulse waveform features; obtaining a physical confidence factor of each suspected pulse according to the falling edge attenuation characteristics; calculating effective discharge energy flow in unit time based on the physical confidence factor; calculating a discharge stability coefficient according to a coupling relationship between the time correlation entropy and the effective discharge energy flow; according to the method, a scheme from microscopic physical judgment, waveform form deconstruction and time law analysis to multi-dimensional trend prediction is constructed, so that the contradiction between the detection sensitivity and the quantification precision is solved; and adaptive tracking monitoring of the partial discharge process is realized.

Description

Technical Field

[0001] This invention relates to the field of phototube detection technology, and in particular to a method and system for detecting solar-blind ultraviolet discharge by high-voltage excitation of phototubes. Background Technology

[0002] In partial discharge monitoring of power systems, solar-blind ultraviolet (UV) detection boasts an extremely high signal-to-noise ratio due to its avoidance of solar spectral interference. Back-illuminated solar-blind UV phototubes integrate a high-voltage ionization structure within the tube, utilizing the strong electric field generated by high-voltage excitation to induce secondary ionization and avalanche of photogenerated electrons. This achieves a high signal gain at the physical level, allowing the system to capture weak discharge signals without the need for complex external multi-stage amplifier circuits.

[0003] However, under this high-voltage excitation mode, the internal physical processes of the phototube become complex. On the one hand, while the high-voltage, strong electric field amplifies the ultraviolet signal, it also induces random avalanche noise excited by heated electrons. This type of noise is highly similar to the real ultraviolet pulse in its time-domain morphology. On the other hand, as the intensity of partial discharge increases, the time density of photons reaching the photosensitive surface increases, leading to a nonlinear pulse stacking effect in the internal ionization avalanche process. Existing detection methods mostly use pulse counting or amplitude accumulation, which makes it difficult to distinguish between thermal noise and weak signals from a physical mechanism perspective. Furthermore, when pulse stacking is severe, it is impossible to accurately quantify the true discharge energy, resulting in a contradiction between detection sensitivity and quantization accuracy. Summary of the Invention

[0004] The main objective of this invention is to provide a method and system for detecting solar-blind ultraviolet discharge using a high-voltage excited phototube, aiming to solve the problems of "avalanche noise interference" under extremely weak signals and "pulse stacking quantization inaccuracy" under strong discharge. By exploring the physical attenuation characteristics of the avalanche process inside the ionization tube, a detection method based on ionization dynamics consistency verification and energy flow temporal evolution analysis is constructed to achieve deep extraction of discharge signals and accurate quantification of hazard level, while maintaining a minimalist hardware architecture.

[0005] This invention proposes a phototube high-voltage excitation method for detecting solar-blind ultraviolet discharge, comprising: The raw voltage sequence data is acquired in real time by a high-speed transimpedance sampling circuit. The raw voltage sequence data is generated by a back-illuminated solar-blind ultraviolet phototube with a built-in high-voltage ionization structure converting the received solar-blind ultraviolet light signal into an electrical pulse signal under the DC bias provided by the high-voltage excitation module and then sampling it through transimpedance. Based on the original voltage sequence data, the feature points of each suspected pulse are located, and the pulse waveform features are extracted. The pulse waveform features include peak points, rising edge slope distribution, and falling edge attenuation characteristics. The measured attenuation constant of each suspected pulse is calculated based on the falling edge attenuation characteristics, and the measured attenuation constant is compared with the pre-stored hardware inherent ionization quenching constant to obtain the physical confidence factor of each suspected pulse. Suspected pulses are screened based on the physical confidence factor to determine candidate valid pulses. Energy quantization is performed on each candidate valid pulse. Based on the rising edge slope distribution, energy deconstruction and correction are performed on candidate valid pulses with nonlinear stacking to obtain the effective discharge energy flow per unit time. The time distribution characteristics of adjacent candidate effective pulses are obtained, and the time distribution characteristics are subjected to regular quantitative analysis to obtain the time-related entropy value. Then, the discharge stability coefficient is calculated based on the coupling relationship between the time-related entropy value and the effective discharge energy flow. The current state vector is constructed based on the effective discharge energy flow, the physical confidence factor, and the discharge stability coefficient. The dynamic hazard index is calculated based on the characteristic distance of the current state vector in the preset insulation degradation space and the slope of the historical state vector.

[0006] Preferably, the step of locating the feature points of each suspected pulse based on the original voltage sequence data and extracting pulse waveform features, wherein the pulse waveform features include peak points, rising edge slope distribution, and falling edge attenuation characteristics, includes: The original voltage sequence data is subjected to waveform rate of change analysis. Based on the zero-crossing characteristics of the waveform rate of change, the start and end points of each suspected pulse are located, and the waveform range of each suspected pulse is determined according to the start and end points. Within each waveform interval, the peak point of the suspected pulse is located with subsampling point precision based on local extremum analysis, and the peak amplitude and peak time corresponding to the peak point are recorded. Based on the peak point and the waveform interval, the rising edge data before the peak point is extracted to obtain the rising edge slope distribution characteristics, and the falling edge data points within a preset time window after the peak point are extracted to construct a pulse decay sequence for characterizing pulse decay characteristics.

[0007] Preferably, the step of calculating the measured attenuation constant of each suspected pulse based on the falling edge attenuation characteristics, and comparing the measured attenuation constant with the pre-stored hardware-inherent ionization quenching constant to obtain the physical confidence factor of each suspected pulse includes: The pulse decay sequence was fitted with an exponential decay model to obtain the measured decay constant of the suspected pulse. Obtain the DC bias voltage applied to the phototube by the high-voltage excitation module at the current moment, and read the inherent ionization quenching constant of the phototube under the current DC bias voltage from the pre-stored parameters; Calculate the deviation between the measured attenuation constant and the inherent ionization quenching constant of the hardware; According to the preset deviation-confidence mapping relationship, the deviation value is mapped to a physical confidence factor between 0 and 1, wherein the mapping relationship satisfies the following: the smaller the deviation value, the larger the physical confidence factor; when the deviation value exceeds the preset range, the physical confidence factor tends to 0.

[0008] Preferably, the steps of screening suspected pulses based on the physical confidence factor to determine candidate valid pulses, performing energy quantization on each candidate valid pulse, and deconstructing and correcting the energy of candidate valid pulses with nonlinear stacking based on the rising edge slope distribution to obtain the effective discharge energy flow per unit time include: Possible pulses whose physical confidence factor is greater than a preset confidence threshold are selected as candidate valid pulses; For each candidate valid pulse, energy quantization is performed based on its waveform range, and the feedback resistance value of the pre-stored high-speed transimpedance sampling circuit and the internal ionization gain multiple of the phototube under the current DC bias are obtained. The voltage time integral value is then subjected to dimensional conversion and gain correction to obtain the apparent charge of the candidate valid pulse. Analyze the rising edge slope distribution of each candidate valid pulse, and detect whether there are waveform distortion features in the rising edge slope distribution that characterize the superposition of multiple avalanche processes; if so, determine that the candidate valid pulse has nonlinear stacking, and divide the corresponding pulse waveform into multiple sub-pulse units according to the waveform distortion features. Each of the sub-pulse units is subjected to energy quantization processing and corrected in combination with inherent hardware parameters to obtain the corresponding sub-unit charge. The charge quantity of the sub-unit is compared and verified with the apparent charge quantity of the corresponding candidate valid pulse to obtain the charge quantity of the sub-unit that has passed the verification. The corrected effective charge of the sub-unit is obtained by comparing the verified charge of the sub-unit with the physical confidence factor of the corresponding candidate effective pulse, and the effective discharge energy flow per unit time is obtained by accumulating them.

[0009] Preferably, the step of acquiring the time distribution characteristics of adjacent candidate effective pulses, performing regularity quantification analysis on the time distribution characteristics to obtain time-related entropy values, and then calculating the discharge stability coefficient based on the coupling relationship between the time-related entropy values ​​and the effective discharge energy flow includes: Based on the timestamps of the candidate valid pulses, construct a time interval sequence of adjacent pulses; The probability distribution statistics of the time interval sequence are performed, and the time correlation entropy value of the time interval sequence is calculated based on the information entropy theory. The time correlation entropy value is used to quantify the regularity of the pulse time distribution. The discharge stability coefficient is calculated based on the coupling relationship between the time-related entropy value and the effective discharge energy flow. The discharge stability coefficient is used to comprehensively characterize the energy intensity and temporal regularity of the discharge process.

[0010] Preferably, the step of constructing a current state vector based on the effective discharge energy flow, the physical confidence factor, and the discharge stability coefficient, and calculating the dynamic hazard index based on the characteristic distance of the current state vector in the preset insulation degradation space and the slope of the historical state vector includes: Calculate the mean of the physical confidence factor for all candidate valid pulses per unit time. The effective discharge energy flow, the mean of the physical confidence factor, and the discharge stability coefficient are normalized to construct the current state vector. Obtain a set of standard state vectors in a preset insulation degradation space, wherein the standard state vectors represent different risk levels; The instantaneous danger value of the current state is determined based on the similarity between the current state vector and each of the standard state vectors. The state vector sequences of the previous several time windows are read from the historical database, and the evolution trend of the state vector sequences is analyzed to obtain the rate of change of the danger level. The dynamic hazard index is calculated based on the instantaneous hazard value and the rate of change. The dynamic hazard index is used to reflect the current discharge risk and its development trend.

[0011] The present invention is further configured such that, after the step of calculating the dynamic hazard index, it further includes: Based on the relationship between the dynamic risk index and multiple preset risk thresholds, alarms of the corresponding level are triggered, and corresponding status information is uploaded.

[0012] The present invention is further configured such that, after the step of calculating the dynamic hazard index, it also includes executing an adaptive feedback adjustment process: The energy contribution ratio of low-confidence pulses is statistically analyzed in real time per unit time, which is used as the noise base level of the current time window. The low-confidence pulses refer to suspected pulses whose physical confidence factor is not greater than a preset confidence threshold. Based on the changing trends of the statistical characteristics of the physical confidence factor, the changing trends of the dynamic risk index, and the noise floor level, the type of the current abnormal state is comprehensively determined. Based on the type of the current abnormal state, a feedback adjustment command is generated and sent to the high-voltage excitation module to fine-tune the DC bias value of the phototube.

[0013] This invention also provides a phototube high-voltage excitation solar-blind ultraviolet discharge detection system, comprising: The ultraviolet detection unit adopts a back-illuminated solar-blind ultraviolet phototube with a built-in high-voltage ionization structure. Under the preset DC bias voltage provided by the high-voltage excitation module, the phototube converts the received solar-blind ultraviolet light signal into an avalanche pulse with inherent ionization quenching characteristics. The high-voltage excitation module, electrically connected to the ultraviolet detection unit, is used to provide an adjustable DC bias voltage to the built-in high-voltage ionization structure in order to control the internal ionization gain ratio. A high-speed transimpedance sampling circuit is directly connected to the output terminal of the ultraviolet detection unit to capture the avalanche electrical pulse and convert it into raw voltage sequence data; The edge computing core is connected to the high-speed transimpedance sampling circuit and has a built-in non-volatile memory. The memory pre-stores the hardware physical characteristic parameters of the phototube, including at least the inherent ionization quenching constant of the phototube.

[0014] The present invention is further configured such that the edge computing core is used to run the following logic modules in real time: The signal acquisition module is used to acquire the raw voltage sequence data collected in real time by the high-speed transimpedance sampling circuit; The data extraction module is used to locate the feature points of each suspected pulse based on the original voltage sequence data and extract the pulse waveform features; The pulse evaluation module is used to calculate the measured attenuation constant of each suspected pulse based on the falling edge attenuation characteristics, and compare the measured attenuation constant with the pre-stored hardware inherent ionization quenching constant to obtain the physical confidence factor of each suspected pulse. The screening and reconstruction module is used to screen suspected pulses according to the physical confidence factor, determine candidate valid pulses, perform energy quantization processing on each candidate valid pulse, and perform energy deconstruction and correction on the candidate valid pulses with nonlinear stacking according to the rising edge slope distribution to obtain the effective discharge energy flow per unit time. The stability analysis module is used to obtain the time distribution characteristics of adjacent candidate effective pulses, perform regularity quantification analysis on the time distribution characteristics to obtain the time-related entropy value, and then calculate the discharge stability coefficient based on the coupling relationship between the time-related entropy value and the effective discharge energy flow. The risk assessment module is used to construct a current state vector based on the effective discharge energy flow, the physical confidence factor, and the discharge stability coefficient, and to calculate a dynamic risk index based on the characteristic distance of the current state vector in the preset insulation degradation space and the slope of the historical state vector. The alarm communication module is used to trigger alarms of the corresponding level and upload corresponding status information based on the relationship between the dynamic risk index and multiple preset risk thresholds.

[0015] The feedback adjustment module is used to generate a feedback adjustment command based on the changing trends of the physical confidence factor and the dynamic risk index, and send it to the high-voltage excitation module to fine-tune the DC bias value of the phototube.

[0016] The beneficial effects of this invention are as follows: By constructing a scheme that integrates microscopic physical discrimination, waveform morphology deconstruction, time pattern analysis, and multidimensional trend prediction, this invention resolves the contradiction between detection sensitivity and quantification accuracy, and achieves adaptive tracking and monitoring of the partial discharge process. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of a method flow according to an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of the system structure according to an embodiment of this application.

[0019] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0021] like Figure 1 As shown, this application provides a phototube high-voltage excitation method for detecting solar-blind ultraviolet discharge, comprising: S1, acquire the raw voltage sequence data collected in real time by the high-speed transimpedance sampling circuit, wherein the raw voltage sequence data is generated by the back-illuminated solar-blind ultraviolet phototube with built-in high-voltage ionization structure converting the received solar-blind ultraviolet light signal into an electrical pulse signal and then sampling it through transimpedance under the DC bias provided by the high-voltage excitation module; S2, locate the peak point of each suspected pulse based on the original voltage sequence data, extract the rising edge data before the peak, obtain the rising edge slope distribution characteristics, and extract the falling edge data points within a preset time window after the peak based on the peak point to construct a pulse decay sequence; S3, calculate the measured attenuation constant of each suspected pulse according to the pulse attenuation sequence, and compare the measured attenuation constant with the pre-stored hardware inherent ionization quenching constant. Obtain the physical confidence factor of each suspected pulse based on the comparison deviation value. S4. Based on the physical confidence factor, suspected pulses exceeding the preset confidence threshold are screened to determine candidate valid pulses. Each candidate valid pulse is integrated over time to obtain its apparent charge. Based on the rising edge slope distribution characteristics, the candidate valid pulses with nonlinear stacking are deconstructed and corrected to obtain the effective discharge energy flow per unit time. S5, obtain the timestamps of adjacent candidate valid pulses, construct a pulse interval sequence, and perform information entropy analysis on the pulse interval sequence to obtain time-related entropy values. Then, calculate the discharge stability coefficient based on the coupling relationship between the time-related entropy values ​​and the effective discharge energy flow. S6. Construct a current state vector based on the effective discharge energy flow, the mean of the physical confidence factor, and the discharge stability coefficient. Calculate the dynamic hazard index based on the characteristic distance of the current state vector in the preset insulation degradation space and the slope of the historical state vector.

[0022] As described in steps S1-S6 above, the phototube high-voltage excitation solar-blind ultraviolet discharge detection method proposed in this invention addresses the technical problems in the existing solar-blind ultraviolet discharge detection field. Through a design that integrates raw voltage signal acquisition with dynamic risk index calculation, it achieves accurate detection of solar-blind ultraviolet discharge signals and dynamic risk assessment of power equipment insulation degradation. In existing technologies, while back-illuminated solar-blind ultraviolet phototubes with built-in high-voltage ionization structures provide extremely high detection sensitivity, they also present two interrelated technical challenges: first, the random avalanche noise induced by the high-voltage electric field is highly similar to the real ultraviolet pulse in its time-domain morphology, making it difficult for traditional pulse counting or amplitude accumulation methods to distinguish between the two from a physical mechanism perspective; second, the pulse stacking effect that occurs during discharge enhancement leads to inaccurate energy quantization, creating a contradiction between detection sensitivity and quantization accuracy. To solve these problems, this invention identifies and eliminates random noise at the physical mechanism level, deconstructs stacked pulses and restores the real discharge energy at the waveform morphology level, analyzes the regular characteristics of the discharge at the time distribution level, and finally integrates multi-dimensional information to achieve dynamic risk quantification, thereby solving the aforementioned technical problems.

[0023] In one embodiment of the present invention, the steps of locating the peak point of each suspected pulse based on the original voltage sequence data, extracting the rising edge data before the peak, obtaining the rising edge slope distribution characteristics, and extracting the falling edge data points within a preset time window after the peak based on the peak point to construct a pulse decay sequence include: S21, perform a first-order difference operation on the original voltage sequence data, identify the zero-crossing point in the sequence that changes from negative to positive as the pulse start point, and the zero-crossing point that changes from positive to negative as the pulse end point, and locate the waveform range of each suspected pulse based on the pulse start point and the pulse end point. S22, within each waveform interval, a local maximum point is located as the peak point of the suspected pulse by using a parabolic fitting algorithm, and the peak amplitude and peak time corresponding to the peak point are recorded; S23, based on the peak time Extracting from the waveform range to The falling edge data points within the time period, among which This is the offset at the start of the falling edge. The length to be truncated on the falling edge, and Construct the pulse decay sequence to be analyzed. ,in The number of sampling points within the time period.

[0024] As described in steps S21-S23 above, this scheme is designed based on the physical morphological characteristics of the pulse. Since real discharge pulses and random noise have different physical generation mechanisms, this difference in mechanism will inevitably leave identifiable feature traces on the time-domain waveform of the pulse. However, to extract these physical features from the original sampled data, it is necessary to solve the problem of how to accurately obtain the complete waveform information of each pulse from the original voltage sequence, including its start position, end position, peak time, and waveform data of specific stages. Moreover, this acquisition method must be accurate enough and cannot introduce additional errors or distortions. To address this fundamental problem, in a specific embodiment of the present invention, the original voltage sequence data is first subjected to a first-order differential operation. The algorithm utilizes the zero-crossing points in the first-order difference sequence that transition from negative to positive as the pulse start point and the zero-crossing points that transition from positive to negative as the pulse end point. Without the need for a preset amplitude threshold, it adaptively determines the complete waveform range of each suspected pulse. This design forms the basis for all subsequent analyses because both physical confidence assessment based on the attenuation constant and stacking detection based on the rising edge slope distribution need to be performed within an accurate waveform range. If the pulse start point is located too early, baseline noise will be included in the integration interval; if it is located too late, the pulse's leading edge information will be lost. Traditional fixed threshold methods are difficult to handle pulses of different amplitudes, while the rate of change zero-crossing detection used in this invention can adaptively follow the true boundary of the pulse.

[0025] After determining the waveform intervals, this invention further locates the peak point of the pulse within each interval using a parabolic fitting algorithm. The accuracy of the peak moment directly determines the accuracy of subsequent falling edge truncation. If there is a deviation in the peak moment, the truncated falling edge data will deviate from the true exponential decay stage, leading to systematic errors in subsequent physical confidence assessments based on the decay constant. This invention does not directly take the maximum value in the sampling sequence as the peak point, but instead uses multiple sampling points near the peak to perform parabolic fitting, thereby obtaining the peak moment with sub-sampling point accuracy. The physical basis of this design is that the waveform of an avalanche pulse near the peak can be approximated by a quadratic function, and the true peak position between sampling points can be estimated through fitting. For a system with a sampling rate of 50 MSPS, this design can improve the peak positioning accuracy and reduce the error in subsequent decay constant calculations.

[0026] After obtaining the precise peak moment, the present invention further extracts from the waveform interval... to The falling edge data points within a time period are used to construct a pulse decay sequence, providing high-quality input data for subsequent physical confidence assessment based on the ionization quenching constant. This design is based on the fact that during the avalanche ionization process inside the phototube, the output current decays exponentially during the avalanche quenching phase. The decay constant is determined by the quenching circuit and gas discharge physical characteristics inside the tube, and is a key physical feature distinguishing real ultraviolet signals from random noise. However, different stages of the pulse falling edge exhibit different physical characteristics: a nonlinear transition region exists near the peak due to space charge effects, where data does not conform to the exponential decay law; the pulse end is easily overwhelmed by noise. Therefore, by setting a starting offset... Skip the nonlinear region near the peak by setting the cutoff length. Ensure sufficient data points for linear regression fitting, while avoiding entering noise-dominated regions. and The specific value needs to be optimized based on the physical characteristics of the phototube. Generally speaking... A time of 0.5 μs to 1 μs is acceptable. A time interval of 3μs to 5μs can be used to ensure that the intercepted data sequence truly reflects the inherent ionization quenching characteristics of the phototube.

[0027] In one embodiment of the present invention, the step of calculating the measured attenuation constant of each suspected pulse based on the pulse attenuation sequence, comparing the measured attenuation constant with the pre-stored inherent ionization quenching constant of the hardware, and obtaining the physical confidence factor of each suspected pulse based on the comparison deviation value includes: S31, Perform a natural logarithmic transformation on each data point in the pulse decay sequence to obtain a logarithmic decay sequence. ; S32, with sampling time Using the logarithmic decay sequence as the dependent variable and the least squares method for linear regression fitting, the slope of the regression line is obtained, and the measured decay constant of the suspected pulse is calculated based on the slope of the regression line. S33, obtain the DC bias voltage applied to the phototube by the high voltage excitation module at the current moment, read the inherent ionization quenching constant of the phototube from the pre-stored parameter area of ​​the edge computing core, and calculate the absolute deviation between the two. S34, according to the preset deviation-confidence mapping function Calculate the physical confidence factor of the suspected pulse, wherein the mapping function satisfies: when When it approaches 0, tending to 1; when Exceeding the preset deviation threshold hour, It approaches 0.

[0028] As described in steps S31-S34 above, this invention achieves effective differentiation between real ultraviolet signals and random avalanche noise from a physical mechanism perspective. In the high-voltage excitation mode described in the background art, the high-voltage electric field inside the phototube amplifies the ultraviolet signal while also inducing random avalanche noise excited by heated electrons. This type of noise is highly similar to real ultraviolet pulses in its time-domain morphology. Traditional pulse counting or amplitude accumulation methods are difficult to distinguish between the two physically, inevitably introducing a large number of false alarms while pursuing high sensitivity in the detection system. Although existing technologies employ waveform matching or template comparison methods, these methods rely on pre-set template libraries, cannot adapt to waveform differences caused by different discharge types and different device batches, and have high computational complexity, making them difficult to run in real time on embedded platforms.

[0029] To address the aforementioned technical problems, this invention provides a physical confidence assessment method based on the ionization quenching constant. This method utilizes the deterministic nature of the internal physical processes of the phototube to distinguish between real signals and random noise. Specifically, the falling edge decay constant of an avalanche pulse excited by real ultraviolet photons should be highly consistent with the inherent ionization quenching constant of the phototube, as this constant is determined by the quenching circuit and gas discharge physical characteristics within the tube and is an inherent property of the device. However, the decay constant of hot electron random avalanches induced by high-voltage electric fields often deviates from the inherent value due to the randomness of their excitation location, initial energy, and ionization path. Based on this physical law, this invention first performs a logarithmic transformation on the pulse decay sequence, converting it into a linear relationship, and obtains the measured decay constant through linear regression fitting. Then, it obtains the current DC bias voltage value and reads the hardware-specific ionization quenching constant corresponding to this bias voltage from the pre-stored parameter area; calculates the absolute deviation between the two; and finally, according to a preset deviation-confidence mapping function, converts the deviation value into a physical confidence factor between 0 and 1. The physical meaning of this confidence factor is: the degree of credibility of the pulse originating from a real ultraviolet signal; the smaller the deviation, the higher the confidence; when the deviation exceeds the threshold, the confidence approaches 0, and the pulse is judged as random noise. Compared with existing technologies, the following technical effects can be achieved: First, it achieves the distinction between signal and noise at the physical mechanism level, rather than relying on a simple comparison of time-domain morphology. This makes the method naturally adaptable to different discharge types and different device batches, without the need to establish a template library in advance. Second, the confidence factor is output in the form of continuous values ​​from 0 to 1, which allows for fine-grained processing of subsequent weighted energy calculations and avoids information loss caused by hard decision-making. Third, by introducing a DC bias voltage to correct the ionization quenching constant, the influence of device operating point changes on physical characteristics is considered, making the evaluation results more accurate. Fourth, it provides multiple deviation-confidence mapping functions, which can be flexibly selected according to different application scenarios and noise environments, taking into account both the determinism and smoothness of the decision.

[0030] It should be noted that linear regression is a conventional mathematical technique. The purpose of this invention is to apply it to a specific physical scenario, utilizing the physical characteristic of exponential decay of phototubes to transform the time-domain waveform into a logarithmic linear relationship that is easy to fit. In a preferred embodiment, to improve fitting accuracy, a weighted least squares method can be used, assigning higher weights to data points with high signal-to-noise ratios.

[0031] It should be noted that the ionization quenching constant is not a fixed value, but varies with the bias voltage. The higher the bias voltage, the stronger the internal electric field, and the more likely the avalanche quenching process will change. Therefore, this invention stores a bias voltage-quenching constant mapping table in the pre-stored parameter area. This mapping table is calibrated using a dedicated testing platform before the phototube leaves the factory.

[0032] Regarding the calculation of the physical confidence factor of the suspected pulse based on a preset deviation-confidence mapping function, this invention provides three different mapping functions as preferred embodiments to adapt to different application scenarios and noise environments.

[0033] In one specific implementation, a piecewise linear mapping function is used: Specifically, it is expressed as follows: ; In the formula, This represents the absolute deviation between the inherent ionization quenching constant of the hardware and the measured decay constant. ,in, This represents the inherent ionization quenching constant of the hardware. This represents the measured attenuation constant. ,in, (represents the slope of the regression line). This represents a preset deviation threshold. The advantage of this mapping method is that the physical meaning is intuitive and the implementation is simple. When the deviation value is within the threshold range, the confidence level decreases linearly as the deviation increases; when the deviation exceeds the threshold, the confidence level drops directly to 0, indicating that the pulse is judged as noise. This hard decision method is suitable for scenarios with relatively stable noise environments.

[0034] In another implementation, a continuous mapping in the form of a Gaussian function is used: ; In the formula, This represents the shape parameter of the Gaussian function, which is an adjustable time constant parameter. It is experimentally calibrated and used to control the decay rate of the physical confidence factor as the deviation increases. Its dimensions are... Consistent (i.e., time in seconds); this continuous mapping method is suitable for scenarios that require fine-grained weighted calculations.

[0035] In another implementation, an exponential mapping form with sensitivity adjustment is used: ; In the formula, This represents the environmental sensitivity adjustment coefficient, used to dynamically adjust the tolerance space for confidence level determination based on the background electromagnetic noise intensity, and is obtained through experimental calibration. This mapping method introduces the concept of relative deviation, which can be adjusted... Adaptable to different noise environments. For example, in high-noise environments, the noise level can be appropriately reduced. The value is increased to expand the fault tolerance margin and prevent too many valid pulses from being misjudged as noise; in low-noise environments, the value can be appropriately increased. This increases the rigor of the judgment.

[0036] By flexibly selecting the three mapping functions mentioned above, this invention can adapt to the needs of different application scenarios: piecewise linear mapping is suitable for embedded scenarios with strict requirements for computing resources; Gaussian function mapping is suitable for refined analysis scenarios that require smooth weighting; and exponential mapping with sensitivity adjustment is suitable for complex environments with dynamically changing noise. Regardless of the mapping method used, the core design idea is consistent: utilizing the inherent ionization quenching physical characteristics of phototubes, the morphological characteristics of pulses are quantified into calculable confidence factors, achieving effective differentiation between signals and noise at the physical mechanism level.

[0037] In one embodiment of the present invention, the steps of screening suspected pulses exceeding a preset confidence threshold based on the physical confidence factor to determine candidate valid pulses, performing time integration on each candidate valid pulse to obtain its apparent charge, and performing energy deconstruction and correction on candidate valid pulses exhibiting nonlinear stacking based on the rising edge slope distribution characteristics to obtain the effective discharge energy flow per unit time include: S41, filter suspected pulses whose physical confidence factor is greater than the preset confidence threshold as candidate valid pulses, and remove suspected pulses whose physical confidence factor is less than the preset confidence threshold as random thermal noise or electromagnetic interference. S42, for each candidate valid pulse, the voltage sequence is integrated over time from the pulse start point to the pulse end point to obtain the voltage time integral value, and the feedback resistance value of the pre-stored high-speed transimpedance sampling circuit and the internal ionization gain multiple of the phototube under the current DC bias are obtained, so as to perform dimensional conversion and gain correction on the voltage time integral value to obtain the apparent charge of the candidate valid pulse. S43, analyze the rising edge slope distribution of each candidate effective pulse, and detect whether there are continuous slope abrupt change points in the rising edge slope distribution; if at least two slope abrupt change points are detected, it is determined that the candidate effective pulse has nonlinear stacking of multiple avalanche processes, and the corresponding pulse waveform is divided into multiple sub-pulse units according to the slope abrupt change points. S44, perform time integration on each of the sub-pulse units, and correct it by combining the feedback resistance value and the internal ionization gain multiple to obtain the corresponding sub-unit charge. For example, the specific implementation process is as follows: Based on the sampling time of the abrupt change in the slope of the pulse rising edge, the starting and ending points of the voltage sequence sampling corresponding to each sub-pulse unit are determined, which are used as the voltage-time integration interval of the sub-pulse unit to ensure that the integration range only covers the effective waveform of the current sub-pulse unit and avoids introducing signal interference from adjacent sub-units; Discretized time integration calculation is performed on the original voltage sampling sequence of each sub-pulse unit within the above integration interval. In this invention, the trapezoidal numerical integration method is used to complete the integration operation. This method is a conventional numerical integration method in the field of signal processing, which can improve the efficiency of engineering implementation while ensuring calculation accuracy. The integration result represents the voltage-time cumulative value of the sub-pulse unit; After the electrical pulse signal output by the phototube is acquired by the high-speed transimpedance sampling circuit, the current to voltage conversion is realized. Moreover, the internal ionization gain of the phototube under high voltage bias will amplify the original discharge signal. Therefore, it is necessary to correct the integration result in combination with the inherent hardware parameters to restore the apparent charge of the sub-pulse unit. The specific conversion relationship is as follows: ; In the formula, This represents the charge of a sub-unit (i.e., the apparent charge of a sub-pulse unit). This indicates the start time of integration for the sub-pulse unit. Indicates the time when the integration of the sub-pulse unit ends. This represents the sampled voltage value within the integration interval. This represents the feedback resistor value of the high-speed transimpedance sampling circuit. It is a fixed hardware parameter, obtained directly from the circuit design parameters. This indicates the DC bias voltage of the phototube at the current moment. The internal ionization gain factor is obtained in real-time from the bias-gain mapping table calibrated before the phototube leaves the factory, and its value varies with... The dynamic adjustment and synchronous update (regarding the construction and calibration method of the bias-gain mapping table: This mapping table is based on the physical characteristics of the phototube and is obtained through systematic calibration experiments in a controlled laboratory environment. The specific implementation process is as follows: First, a calibration test platform including a standard pulse light source, a high-precision high-voltage power supply, and a picoampere-level current measurement unit is built. The back-illuminated solar-blind ultraviolet phototube under test is placed in a dark room environment to eliminate background noise interference; Second, within the safe operating voltage range of the phototube, the DC bias voltage applied between the cathode and anode of the phototube is adjusted in a stepwise manner with a preset step size (e.g., 0.5V or 1V). At each stable bias voltage point, a standard ultraviolet pulse signal with known photon flux and known energy distribution is injected using a standard light source; Then, the anode current response waveform output by the phototube is synchronously acquired, and the waveform is calculated by time integration. The actual output charge under the bias voltage is compared with the initial charge generated by the theoretical incident photon to calculate the internal ionization gain factor corresponding to the specific bias voltage point. Then, the above steps are repeated to traverse the entire operating voltage range, obtaining a series of discrete 'bias voltage value-gain factor' data pairs. Finally, cubic spline interpolation or least squares method is used to smooth and fit the discrete data points, generating a continuous bias voltage-gain function curve. This curve is then discretized into a high-resolution look-up table (LUT) and stored in the system's non-volatile memory. This allows the edge computing core to quickly and accurately index the corresponding internal ionization gain factor based on the real-time monitored DC bias voltage during real-time detection, thus achieving accurate inversion of the actual discharge energy under the pulse stacking effect. Through the above steps, the conversion of a single sub-pulse unit from voltage sampling sequence to actual charge quantity is completed, and the charge quantity calculation process of each sub-unit is independent of each other; S45, compare and verify the charge amount of the sub-unit with the apparent charge amount of the corresponding candidate effective pulse, and obtain the charge amount of the sub-unit that has passed the verification. S46, the verified sub-unit charge is multiplied by the physical confidence factor of the candidate effective pulse to obtain the corrected sub-unit effective charge, and the effective discharge energy flow per unit time is obtained by accumulating.

[0038] As described in steps S41-S46 above, this invention solves the following problem through this design: With the increase of partial discharge intensity, the time density of photons reaching the photosensitive surface increases, leading to a nonlinear pulse stacking effect in the internal ionization avalanche process. Traditional methods cannot accurately quantify the true discharge energy when pulse stacking is severe. Existing detection technologies mostly use pulse counting or overall integration of complete pulses. These methods have drawbacks when processing stacked pulses: if the stacked pulses are treated as a whole for integration, the result is the total charge of multiple avalanche processes, making it impossible to distinguish the contribution of each sub-pulse; if peak detection is used for pulse counting, multiple sub-pulses may be misjudged as a single pulse, resulting in a severely understated count. Both of these situations cause serious inaccuracies in energy quantization, creating an irreconcilable contradiction between detection sensitivity and quantization accuracy. Especially in the high-energy discharge stage, pulse stacking occurs frequently, and the energy quantization inaccuracy problem directly leads to a decrease in the reliability of hazard assessment.

[0039] To address the aforementioned technical problems, this invention provides a method for calculating effective discharge energy flow based on pulse stacking deconstruction. This method does not treat stacked pulses as an indivisible whole. Instead, it analyzes the slope distribution characteristics of the rising edges to identify the internal stacking structure of the pulses, reconstructs the composite waveform into multiple independent sub-pulse units, calculates the actual charge of each sub-unit, and then performs weighted correction using a physical confidence factor to finally synthesize an effective discharge energy flow with clear physical meaning. Specifically, firstly, candidate valid pulses are screened based on the physical confidence factor to eliminate noise interference; then, time integration is performed on each candidate valid pulse to obtain its apparent charge as a benchmark for subsequent verification; next, the rising edge slope distribution is analyzed to detect whether there are continuous slope abrupt changes—these abrupt changes are the physical traces left when multiple avalanche processes are triggered sequentially in time and the waveforms are superimposed; if at least two abrupt changes are detected, it is determined that there is nonlinear stacking, and the pulse waveform is divided into multiple sub-pulse units according to the abrupt changes; time integration is performed on each sub-pulse unit, and correction is made in combination with the feedback resistance value and the internal gain factor to obtain the true sub-unit charge; in particular, this invention designs a verification mechanism to compare the sum of the charges of each sub-unit with the apparent charge to verify the rationality of the segmentation; finally, the verified sub-unit charge is multiplied by the physical confidence factor of the pulse, and a DC bias voltage is introduced to convert the charge into energy, and the effective discharge energy flow per unit time is accumulated.

[0040] Based on the above-mentioned targeted design, this method achieves the following technical effects compared with existing technologies: First, it solves the problem of energy quantification under pulse stacking conditions by dividing the composite pulse into multiple independent sub-pulse units through waveform segmentation, so that the energy contribution of each avalanche process can be accurately measured; Second, it introduces apparent charge as a verification benchmark to verify the rationality of the segmentation results, avoiding energy calculation errors caused by false detection or missed detection of slope change points, thus improving the robustness of the method; Third, it combines physical confidence factor to weight the charge of sub-units, realizing refined processing of signal confidence weighting, making energy flow calculation more accurate.

[0041] It should be noted that, firstly, the basis for screening using the physical confidence factor—the confidence threshold—can be dynamically adjusted according to the ambient noise level, typically ranging from 0.3 to 0.6. For example, in a noisy environment, the threshold can be appropriately lowered to 0.3 to avoid excessive invalidation of valid pulses; in a quieter environment, the threshold can be increased to 0.6 to further improve the quality of the input signal.

[0042] It should be noted that for each candidate valid pulse, time integration is performed. Combined with the feedback resistor value of the high-speed transimpedance sampling circuit and the internal ionization gain multiple of the phototube under the current DC bias, the integral value is dimensionally transformed and the gain is corrected. Its function is to restore the sampled voltage value to the actual discharge charge, eliminate the influence of circuit gain and device gain. The apparent charge represents the total discharge charge of the pulse, which does not depend on any segmentation assumptions and is the benchmark value for subsequent verification of the rationality of the segmentation.

[0043] It should be noted that for rising edge slope distribution analysis and stacking detection, analyzing the rising edge slope distribution of each candidate valid pulse and detecting whether there are continuous slope abrupt changes is the main technical means for stacking identification in this scheme. Specifically, within the rising edge interval from the pulse start point to the peak time, the slope (i.e., the first-order difference value) of each sampling point is calculated to form a slope sequence. Under normal circumstances, the rising edge slope change of a single pulse is continuous; when multiple avalanche processes stack in time, the start of subsequent avalanches will superimpose a new rising trend on the rising edge of the waveform, causing abrupt changes in the slope. For example, in a specific implementation example, suppose the peak times of two sub-pulses differ by 0.8μs. When the first pulse rises to half amplitude, the second pulse begins to trigger. At this time, the rising edge slope changes from steep to gentle, forming a clear turning point. If at least two slope abrupt changes are detected (corresponding to at least three sub-pulses) or one abrupt change combined with other features (corresponding to two sub-pulses), it is determined that the candidate valid pulse has nonlinear stacking of multiple avalanche processes.

[0044] It should be noted that the segmentation of sub-pulse units is based on the following: each slope abrupt change point corresponds to the start time of a new avalanche process, and the waveforms before and after this moment belong to different sub-pulses. Based on the sampling time of the abrupt change point, the starting and ending points of the voltage sequence sampling for each sub-pulse unit are determined, serving as the voltage-time integration interval for that sub-pulse unit. Discrete time integration is performed on each sub-pulse unit within the aforementioned integration interval, which can be achieved using the trapezoidal numerical integration method. The integration result represents the voltage-time cumulative value of that sub-pulse unit. In particular, the design of this invention lies in the method of determining the integration interval, which is based on the slope abrupt change points identified by the physical mechanism for segmentation.

[0045] Regarding the verification mechanism for the rationality of the segmentation, after calculating the charge of each sub-unit, the sum of the charges of each sub-unit is compared with the apparent charge of the pulse for verification. If the comparison is successful, the segmentation result is deemed reasonable, and the sub-unit charge verification passes. If not, the stacking structure is deemed abnormal, possibly due to missed sub-units or falsely detected abrupt changes, requiring re-detection of slope abrupt changes or adjustment of the segmentation parameters. For example, if the relative deviation between the sum of the sub-unit charges and the apparent charge is less than 10%, the segmentation is considered reasonable; if the deviation exceeds 10%, a re-detection is triggered. This verification mechanism ensures the reliability of the stacking structure and avoids energy calculation errors caused by segmentation mistakes.

[0046] Regarding the synthesis of effective discharge energy flow, the formula for calculating effective discharge energy flow is as follows: ; ; In the formula, Indicates the effective discharge energy flow. This represents the total number of candidate valid pulses per unit time. Indicates the first The number of sub-pulse units after a candidate valid pulse is segmented. Indicates the first The physical confidence factor of each candidate valid pulse. Represents the charge of the sub-unit, i.e., the first... Among the candidate valid pulses, the first... The apparent charge of each sub-pulse unit The DC bias value provided for the high-voltage excitation module. Indicates the first The effective discharge energy of each candidate effective pulse.

[0047] In one embodiment of the present invention, the steps of obtaining the timestamps of adjacent candidate valid pulses, constructing a pulse interval sequence, performing information entropy analysis on the pulse interval sequence to obtain a time-related entropy value, and then calculating the discharge stability coefficient based on the coupling relationship between the time-related entropy value and the effective discharge energy flow include: S51, based on the selected candidate valid pulses, record the peak time of each candidate valid pulse. ,in The pulse number per unit time; S52, calculate the time interval between adjacent candidate valid pulses. Constructing pulse interval sequences ,in The total number of candidate valid pulses per unit time; S53, perform probability distribution statistics on the pulse interval sequence, divide the pulse interval value into a preset number of equally wide intervals, and calculate the probability of pulse interval occurrence in each interval. ,in, For range index; S54, calculate the time-related entropy value based on the probability of the pulse interval occurrence. The formula for the time-related entropy value is as follows: Among them, when the pulse intervals exhibit high regularity (such as periodic discharge), The value is relatively small; when the pulse interval exhibits a completely random distribution (such as white noise), The value is relatively large; S55, calculate the discharge stability coefficient based on the coupling relationship between the time-related entropy value and the effective discharge energy flow.

[0048] As described in steps S51-S55 above, through these steps, the present invention reveals the inherent characteristics of different discharge types from the perspective of the temporal distribution pattern of discharge pulses, providing a second key dimension besides energy intensity for subsequent hazard assessment. In practical applications, partial discharge and external electromagnetic interference may be highly similar in time-domain waveforms, making it difficult to effectively distinguish them based solely on pulse shape and energy intensity. However, there are fundamental differences in their temporal distribution patterns: real partial discharge is usually related to power frequency voltage, exhibiting obvious periodicity or phase clustering, with its pulse intervals showing a regular distribution; while external electromagnetic interference, random noise, etc., exhibit completely random temporal distributions. Existing detection methods mostly focus on the feature analysis of individual pulses, neglecting the important information dimension of the temporal correlation of pulse sequences. This makes it difficult to accurately distinguish between continuous partial discharge and intermittent random interference in complex electromagnetic environments, affecting the accuracy of discharge type identification and the reliability of hazard assessment.

[0049] To address the aforementioned technical problems, this invention provides a discharge stability analysis method based on time-related entropy. This method treats a pulse sequence as a point-in-time process, quantifies its regularity by analyzing the probability distribution of pulse intervals, and couples this regularity with energy flow to construct a composite index that comprehensively reflects the "intensity" and "stability" of the discharge. Specifically, firstly, based on the selected candidate effective pulses, the peak time of each pulse is recorded to construct a time interval sequence of adjacent pulses; then, the probability distribution of this interval sequence is statistically analyzed, discretizing the continuous interval values ​​into a probability distribution, and calculating the time-related entropy value based on information entropy theory—the smaller the entropy value, the more regular the pulse interval; the larger the entropy value, the more random the pulse interval; finally, the discharge stability coefficient is calculated based on the coupling relationship between the time-related entropy value and the effective discharge energy flow. In particular, this invention provides two different coupling calculation methods: the linear weighted form is suitable for scenarios where it is necessary to clearly distinguish between energy and regularity contributions; the entropy-weighted penalty form more intuitively reflects the physical idea of ​​"discounting equivalent energy due to randomness".

[0050] Based on the above-mentioned targeted design, this method achieves the following technical effects compared to existing technologies: First, it reveals the inherent regularity of discharge from the perspective of time distribution, providing an effective quantitative means to distinguish between periodic partial discharge and random electromagnetic interference; Second, by using information entropy, a mature uncertainty measurement tool, it compresses the complex pulse interval distribution characteristics into a single value, facilitating subsequent feature fusion and decision-making; Third, by coupling the time-related entropy with the effective discharge energy flow, the discharge stability coefficient simultaneously includes information from two orthogonal dimensions: "discharge intensity" and "discharge regularity," providing more comprehensive input features for hazard assessment; Fourth, it provides multiple coupling calculation methods, which can be flexibly selected according to different application scenarios, taking into account both the intuitiveness of physical meaning and computational efficiency.

[0051] It should be noted that information entropy is a classic indicator for measuring the uncertainty of random variables, and this invention applies it to the quantification of the regularity of pulse interval sequences. When the pulse intervals exhibit high regularity, such as periodic discharge under power frequency voltage, the pulse intervals are concentrated around the power frequency period (e.g., 20ms) and its multiples, and the probability distribution is concentrated in a few intervals, at which point the entropy value is small; when the pulse intervals exhibit a completely random distribution, such as white noise, the probability distribution is close to a uniform distribution, and the entropy value reaches its maximum.

[0052] In one implementation, the discharge stability coefficient is calculated using a linear weighted form: ; In the formula, Indicates the discharge stability coefficient. This indicates the preset maximum energy flow reference value. Indicates the effective discharge energy flow. This represents the preset maximum entropy reference value. Represents the time-dependent entropy value. This represents the energy flow weighting coefficient. Represents the entropy weighting coefficient ( The physical meaning of this formula is that discharge stability is determined by both energy intensity and regularity; the stronger the energy and the stronger the regularity (the smaller the entropy value), the higher the stability coefficient. The values ​​of the energy flow weighting coefficient and the entropy weighting coefficient can be adjusted according to the application scenario.

[0053] In one implementation, the discharge stability coefficient is calculated using an entropy-weighted penalty method: ; In the formula, This represents a preset entropy weight penalty factor, used to suppress the influence of background interference signals with high randomness on the calculation of the stability coefficient. This represents the normalized effective discharge energy flow. The physical meaning of this formula is: applying a discount to the discharge energy that is related to time randomness; the stronger the randomness (…). The larger the value of the entropy weight penalty factor, the greater the discount, the smaller the resulting stability coefficient, and the lower the sustained threat level of the discharge. The value of the entropy weight penalty factor determines the strength of the energy discount due to randomness, typically ranging from 0.1 to 0.5. The advantage of this approach is its intuitive physical meaning; the stability coefficient can be understood as "the equivalent energy after randomness discounting." Furthermore, it does not require a preset maximum reference value and is suitable for scenarios with a large energy dynamic range. The linear weighted form has the advantage that the contributions of energy and regularity can be adjusted independently, making it suitable for scenarios where there is clear prior knowledge of the importance of both. The entropy-weighted penalty form has the advantage of intuitive physical meaning and does not require a preset maximum energy value, making it suitable for scenarios where the dynamic range of energy is uncertain. In practical applications, the appropriate form can be selected based on the specific site conditions, or the stability coefficients of both forms can be calculated simultaneously as multi-dimensional inputs for subsequent hazard assessment.

[0054] In one embodiment of the present invention, the step of constructing a current state vector based on the effective discharge energy flow, the mean of the physical confidence factor, and the discharge stability coefficient, and calculating the dynamic hazard index based on the characteristic distance of the current state vector in the preset insulation degradation space and the slope of the historical state vector includes: S61, calculate the mean of the physical confidence factor of all candidate valid pulses per unit time; S62, normalize the effective discharge energy flow, the mean of the physical confidence factor, and the discharge stability coefficient to construct the current state vector. ,in, Indicates the index of the current time window; S63, Obtain the set of standard state vectors in the preset insulation degradation space. These represent three standard states: normal, warning, and alarm, respectively. S64, Calculate the Euclidean distance from the current state vector to each vector in the standard state vector set. And determine the instantaneous danger value of the current state based on the principle of minimum distance. ; S65, reads previous data from a pre-built historical database. A state vector sequence for each time window (Number of time windows) The value is set according to the system's real-time requirements (e.g., 5 to 20), and the corresponding instantaneous hazard value sequence is calculated. The slope of the change in the instantaneous danger value sequence was fitted by linear regression. S66, Calculate the dynamic hazard index based on the instantaneous hazard value and the change slope.

[0055] As described in steps S61-S66 above, this invention integrates the energy intensity, signal reliability, and discharge stability extracted in the preceding steps, and incorporates historical evolution trends to ultimately output a quantitative indicator that comprehensively reflects the degree of discharge hazard and its development trend. In practical applications, existing detection methods often use fixed thresholds to classify single features, for example, triggering an alarm when the energy exceeds a certain threshold. This static assessment method has two drawbacks. First, a single feature cannot fully describe the complex hazards of a discharge; high-energy but intermittent discharges and medium-energy but continuously deteriorating discharges may have different risk levels. Second, static thresholds cannot reflect the dynamic evolution trend of the discharge; a rapidly increasing discharge may already pose a greater potential risk than a stable but slightly higher-energy discharge, even if the current energy has not yet reached the threshold. This condition monitoring rather than trend prediction assessment mode makes it difficult for the system to achieve true early warning; often, by the time an alarm is triggered, insulation degradation has already reached a considerably severe level.

[0056] To address the aforementioned technical problems, this invention provides a dynamic risk index calculation method based on multi-dimensional feature fusion. This method describes the discharge process as a trajectory that evolves over time in a feature space. It assesses the instantaneous risk level by assessing the position of the current state in a preset degradation space, predicts the future evolution trend by assessing the slope of the historical trajectory, and integrates the two into a dynamic risk index that comprehensively considers both the "current state" and the "evolution trend". Specifically, firstly, the mean physical confidence factor of all candidate valid pulses per unit time is calculated, and together with the existing effective discharge energy flow and discharge stability coefficient, a three-dimensional state vector is formed. Then, these three characteristic components are normalized to eliminate dimensional differences. Next, a set of standard state vectors in a preset insulation degradation space is obtained. These standard vectors are calibrated through laboratory simulation experiments and represent typical states under different risk levels such as normal, warning, and alarm. The Euclidean distance from the current state vector to each standard state vector is calculated, and the instantaneous hazard value of the current state is determined according to the minimum distance principle. At the same time, the state vector sequences of the previous few time windows are read from the historical database, the corresponding instantaneous hazard value sequences are calculated, and the change slope of the sequence is fitted by linear regression to quantify the rate of hazard evolution. Finally, the instantaneous hazard value and the change slope are weighted and fused to obtain the dynamic hazard index, where the trend weight coefficient has the dimension of time, and its physical meaning is to consider the impact of the change trend over a future period on the current risk assessment. Compared with existing technologies, this method has the following advantages: First, it achieves the organic integration of multi-dimensional features, enabling hazard assessment to simultaneously consider the energy intensity of the discharge, signal reliability, and time stability, overcoming the one-sidedness of single-feature assessment. Second, it introduces historical evolution trend analysis, making the assessment results forward-looking and able to detect the trend of discharge deterioration in advance, thus achieving trend prediction. Third, by pre-setting a standard state vector in the insulation degradation space, the hazard assessment has a clear physical benchmark, making the assessment results of different equipment and different sites comparable. Fourth, the trend weighting system is designed with time as its dimension, allowing the system to flexibly adjust the prediction lead time according to the early warning requirements, such as considering the trend in the next 5 minutes or the trend in the next 30 minutes, adapting to the differentiated requirements of different equipment for timely early warning.

[0057] It should be noted that regarding the preset insulation degradation space: a set of standard state vectors is obtained within the preset insulation degradation space, representing three standard states: normal, warning, and alarm. The purpose of determining these standard vectors is to provide a physical benchmark for hazard assessment. In one specific implementation, the standard vectors are obtained through laboratory simulation experiments: under a controlled experimental environment, different discharge stages (normal weak discharge, moderate warning discharge, and severe alarm discharge) are simulated, a large number of samples are collected, and the state vectors of each stage are statistically analyzed. Typical values ​​for each stage are then taken as the standard vectors. For example, the calibration result might be: Normal state: ={0.2,0.9,0.8} indicates low energy, high confidence, and good stability; Warning status: ={0.5,0.7,0.5} indicates moderate energy, slightly decreased confidence, and moderate stability; Alarm status: ={0.8,0.5,0.2} indicates high energy, low confidence (possibly due to waveform distortion caused by stacking), and poor stability. It should be noted that the above values ​​are only examples and should be calibrated according to the specific equipment and field environment in actual applications.

[0058] In one embodiment, the formula for synthesizing the dynamic hazard index is: ; In the formula, This represents the dynamic risk index, a comprehensive risk value that takes into account the current discharge state and the trend of deterioration. Its value ranges from [0, +∞). This indicates the instantaneous danger value, reflecting the degree of discharge risk within the current time window. The slope represents the rate of change, reflecting the rate of evolution of discharge hazard. It is obtained through linear regression fitting of historical instantaneous hazard value sequences. This represents the trend weighting coefficient, used to adjust the degree of influence of the deterioration trend on the current risk assessment. It is obtained through experimental calibration and its unit is time (seconds). Preferably, The value should match the length of the time window; for example, when the window length is 10 seconds, The actual time corresponding to 30-300 seconds (i.e., 3-30 windows) can be taken. The mean of the physical confidence factor reflects the overall confidence level of candidate valid pulses per unit time and is obtained by arithmetic mean calculation.

[0059] In one embodiment of the present invention, after the step of calculating the dynamic hazard index, the method further includes: S71, determine whether the dynamic risk index exceeds a preset warning threshold or alarm threshold, wherein the alarm threshold is greater than the warning threshold; S72, if the dynamic hazard index is not less than the alarm threshold, an emergency alarm is triggered through the communication and alarm unit, and alarm information including the current state vector, dynamic hazard index and discharge type discrimination result is uploaded to the background monitoring system. S73, If the dynamic hazard index is between the warning threshold and the alarm threshold, a warning prompt is triggered, and a warning message containing the current state vector and the dynamic hazard index is uploaded to the background monitoring system. It is recommended to strengthen equipment monitoring. S74, if the dynamic hazard index is less than the warning threshold, it is determined that there is no dangerous discharge, no alarm is triggered, and only the current monitoring data is stored in the historical database. S75, simultaneously, executes an adaptive feedback adjustment process, specifically including: The ratio of the sum of the apparent charge of low-confidence pulses per unit time to the sum of the apparent charge of all pulses within the time window is used as the noise floor level of the current time window. The low-confidence pulses refer to suspected pulses whose physical confidence factor is not greater than a preset confidence threshold. If the mean of the physical confidence factor continues to decrease, the dynamic risk index continues to rise, and the noise floor level is lower than the preset noise threshold, it is determined that the discharge signal characteristics are distorted due to insulation degradation, and a feedback adjustment command is generated and sent to the high-voltage excitation module. If the mean of the physical confidence factor is detected to decrease and the noise floor level exceeds the preset noise threshold, it is determined that the external electromagnetic interference is enhanced, the high voltage regulation is suspended, and at least one of the following operations is performed: triggering an electromagnetic interference alarm, starting the digital filter enhancement mode, increasing the confidence threshold for pulse consistency discrimination, or extending the monitoring time window. The high-voltage excitation module, based on the feedback adjustment command, fine-tunes the DC bias applied to the back-illuminated solar-blind ultraviolet phototube within a preset safe voltage range to optimize the internal ionization gain and achieve adaptive tracking and monitoring of the entire discharge evolution process.

[0060] As described above, this invention, through this design, transforms the dynamic hazard index calculated in the preceding steps into specific operation and maintenance action instructions. This method uses the dynamic hazard index as a decision-making basis to achieve graded output of risk levels; simultaneously, by real-time monitoring of the noise floor level, it distinguishes between two different scenarios: "signal distortion caused by insulation degradation" and "enhanced external electromagnetic interference," and adopts differentiated response strategies for each scenario. Specifically, firstly, warning and alarm thresholds are set. Different levels of alarms are triggered based on the relationship between the dynamic hazard index and the thresholds, and the corresponding status information is uploaded to the background monitoring system. Simultaneously, the energy percentage of low-confidence pulses per unit time is statistically analyzed in real time as the noise floor level for the current time window. Then, based on the changing trends of the physical confidence factor mean, the changing trends of the dynamic hazard index, and the noise floor level, scenario judgment is made. If the confidence decreases, the hazard increases, and the noise floor level is low, it is determined to be signal characteristic distortion caused by insulation degradation, and a feedback adjustment command is generated to adjust the phototube bias voltage. If the confidence decreases but the noise floor level is high, it is determined to be enhanced external electromagnetic interference, and high-voltage regulation is suspended while other anti-interference measures are initiated. Finally, the high-voltage excitation module fine-tunes the bias voltage value within the safe voltage range according to the feedback adjustment command, optimizes the internal ionization gain ratio, and achieves adaptive tracking and monitoring of the entire discharge evolution process.

[0061] The above scheme enables graded output of hazard levels, allowing maintenance personnel to take differentiated countermeasures based on alarm levels. By introducing a noise floor level, it effectively distinguishes between signal distortion and noise interference scenarios, avoiding performance degradation caused by blindly adjusting the bias voltage when external interference increases. Furthermore, it allows the system to adaptively adjust its operating point according to the discharge evolution process, maintaining better detection sensitivity. In addition, it provides multiple anti-interference measures as alternatives, enhancing the system's adaptability in complex electromagnetic environments.

[0062] It should be noted that the definition of noise floor level is significant because it uses the proportion of energy rather than the proportion of pulse quantity to assess the level of interference, since a small amount of high-amplitude noise is more harmful than a large amount of low-amplitude noise.

[0063] It should be noted that, in order to distinguish between the two completely different physical scenarios of "signal distortion caused by insulation degradation" and "enhanced external electromagnetic interference", this invention adopts differentiated countermeasures: Scenario 1: Signal distortion caused by insulation degradation. If the mean of the physical confidence factor continues to decrease, the dynamic hazard index continues to increase, and the noise floor level is lower than the preset noise threshold, it is determined that the discharge signal characteristics are distorted due to insulation degradation. The logic here is: a decrease in the mean of the physical confidence factor means that the pulse waveform deviates from the inherent characteristics of the device, which may be due to waveform changes caused by the evolution of the discharge type; an increase in the dynamic hazard index means an increase in risk; a low noise floor level indicates that the noise proportion is not large, so the signal distortion mainly comes from changes in the actual discharge rather than external interference. At this time, a feedback adjustment command is generated and sent to the high-voltage excitation module to optimize the detection sensitivity by adjusting the bias voltage to adapt to the changed signal characteristics.

[0064] Scenario 2: Increased External Electromagnetic Interference. If a decrease in the mean of the physical confidence factor and an increase in the noise floor level are detected, it is determined that external electromagnetic interference has increased. The logic here is: a decrease in the mean of the physical confidence factor and an increase in the noise floor level indicate a surge of low-confidence pulses, which are likely external interference rather than actual discharges. In this case, high-voltage regulation is paused (because adjusting the bias voltage cannot eliminate external interference), and at least one of the following operations is performed instead: triggering an electromagnetic interference alarm, activating a digital filter enhancement mode, increasing the confidence threshold for pulse consistency discrimination, or extending the monitoring time window. The purpose of these measures is to enhance the system's ability to suppress interference, rather than changing the device operating point.

[0065] It should be noted that the safe voltage range is set to avoid damage to the phototube due to excessive bias or insufficient sensitivity due to excessively low bias. A typical range is 800V~1500V. The fine-tuning step size can be set as needed, for example, adjusting 10V~50V each time. The direction and magnitude of the adjustment can be determined based on the changing trends of the physical confidence factor mean and the dynamic hazard index: if the physical confidence factor mean decreases and the dynamic hazard index increases, a small increase in bias voltage can be tried to improve the gain and make the signal characteristics more obvious; if the physical confidence factor rebounds after adjustment, it indicates that the adjustment is effective and adjustment in the same direction can continue; if the physical confidence factor continues to decrease, adjustment in the opposite direction or pausing is required.

[0066] like Figure 2 As shown, the present invention also provides a phototube high-voltage excitation solar-blind ultraviolet discharge detection system, comprising: The ultraviolet detection unit adopts a back-illuminated solar-blind ultraviolet phototube with a built-in high-voltage ionization structure. Under the DC bias voltage of 800V-1500V provided by the high-voltage excitation module, the phototube converts the received solar-blind ultraviolet light signal into an avalanche pulse with an inherent ionization quenching constant. The high-voltage excitation module, electrically connected to the ultraviolet detection unit, is used to provide an adjustable DC bias voltage to the built-in high-voltage ionization structure in order to control the internal ionization gain ratio. A high-speed transimpedance sampling circuit is directly connected to the output of the ultraviolet detection unit to capture the avalanche electrical pulse and convert it into raw voltage sequence data, without the need for an external multi-stage amplification unit. The edge computing core, connected to the high-speed transimpedance sampling circuit, has a built-in non-volatile memory. The memory pre-stores the hardware physical characteristic parameters of the phototube, including at least the inherent ionization quenching constant of the phototube. The edge computing core is used to run the following logic modules in real time: The signal acquisition module is used to acquire the raw voltage sequence data collected in real time by the high-speed transimpedance sampling circuit; The data extraction module is used to locate the peak point of each suspected pulse based on the original voltage sequence data, extract the rising edge data before the peak, obtain the rising edge slope distribution characteristics, and extract the falling edge data points within a preset time window after the peak based on the peak point to construct a pulse decay sequence. The pulse evaluation module is used to calculate the measured attenuation constant of each suspected pulse based on the pulse attenuation sequence, and compare the measured attenuation constant with the pre-stored hardware inherent ionization quenching constant. Based on the comparison deviation value, the physical confidence factor of each suspected pulse is obtained. The screening and reconstruction module is used to screen suspected pulses that exceed the preset confidence threshold according to the physical confidence factor, determine candidate valid pulses, perform time integration on each candidate valid pulse to obtain its apparent charge, and perform energy deconstruction and correction on the candidate valid pulses with nonlinear stacking according to the rising edge slope distribution characteristics to obtain the effective discharge energy flow per unit time. The stability analysis module is used to obtain the timestamps of adjacent candidate effective pulses, construct a pulse interval sequence, perform information entropy analysis on the pulse interval sequence to obtain a time-related entropy value, and then calculate the discharge stability coefficient based on the coupling relationship between the time-related entropy value and the effective discharge energy flow. The risk assessment module is used to construct a current state vector based on the effective discharge energy flow, the mean of the physical confidence factor, and the discharge stability coefficient, and to calculate a dynamic risk index based on the characteristic distance of the current state vector in the preset insulation degradation space and the slope of the historical state vector. The alarm communication module is used to trigger alarms of the corresponding level and upload corresponding status information based on the relationship between the dynamic risk index and multiple preset risk thresholds.

[0067] The feedback adjustment module is used to generate a feedback adjustment command based on the changing trends of the physical confidence factor and the dynamic risk index, and send it to the high-voltage excitation module to fine-tune the DC bias value of the phototube.

[0068] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a phototube high-voltage excitation solar-blind ultraviolet discharge detection method.

[0069] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a phototube high-voltage excitation solar-blind ultraviolet discharge detection method.

[0070] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0071] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for detecting solar-blind ultraviolet discharge by high-voltage excitation of a phototube, characterized in that, include: The raw voltage sequence data is acquired in real time by a high-speed transimpedance sampling circuit. The raw voltage sequence data is generated by a back-illuminated solar-blind ultraviolet phototube with a built-in high-voltage ionization structure converting the received solar-blind ultraviolet light signal into an electrical pulse signal under the DC bias provided by the high-voltage excitation module and then sampling it through transimpedance. Based on the original voltage sequence data, the feature points of each suspected pulse are located, and the pulse waveform features are extracted. The pulse waveform features include peak points, rising edge slope distribution, and falling edge attenuation characteristics. The measured attenuation constant of each suspected pulse is calculated based on the falling edge attenuation characteristics, and the measured attenuation constant is compared with the pre-stored hardware inherent ionization quenching constant to obtain the physical confidence factor of each suspected pulse. Suspected pulses are screened based on the physical confidence factor to determine candidate valid pulses. Energy quantization is performed on each candidate valid pulse. Based on the rising edge slope distribution, energy deconstruction and correction are performed on candidate valid pulses with nonlinear stacking to obtain the effective discharge energy flow per unit time. The time distribution characteristics of adjacent candidate effective pulses are obtained, and the time distribution characteristics are subjected to regular quantitative analysis to obtain the time-related entropy value. Then, the discharge stability coefficient is calculated based on the coupling relationship between the time-related entropy value and the effective discharge energy flow. The current state vector is constructed based on the effective discharge energy flow, the physical confidence factor, and the discharge stability coefficient. The dynamic hazard index is calculated based on the characteristic distance of the current state vector in the preset insulation degradation space and the slope of the historical state vector.

2. The phototube high-voltage excitation method for detecting solar-blind ultraviolet discharge according to claim 1, characterized in that, The step of locating feature points of each suspected pulse based on the original voltage sequence data and extracting pulse waveform features, including peak points, rising edge slope distribution, and falling edge attenuation characteristics, includes: The original voltage sequence data is subjected to waveform rate of change analysis. Based on the zero-crossing characteristics of the waveform rate of change, the start and end points of each suspected pulse are located, and the waveform range of each suspected pulse is determined according to the start and end points. Within each waveform interval, the peak point of the suspected pulse is located with subsampling point precision based on local extremum analysis, and the peak amplitude and peak time corresponding to the peak point are recorded. Based on the peak point and the waveform interval, the rising edge data before the peak point is extracted to obtain the rising edge slope distribution characteristics, and the falling edge data points within a preset time window after the peak point are extracted to construct a pulse decay sequence for characterizing pulse decay characteristics.

3. The phototube high-voltage excitation method for detecting solar-blind ultraviolet discharge according to claim 1, characterized in that, The step of calculating the measured attenuation constant of each suspected pulse based on the falling edge attenuation characteristics, and comparing the measured attenuation constant with the pre-stored hardware-inherent ionization quenching constant to obtain the physical confidence factor of each suspected pulse includes: The pulse decay sequence was fitted with an exponential decay model to obtain the measured decay constant of the suspected pulse. Obtain the DC bias voltage applied to the phototube by the high-voltage excitation module at the current moment, and read the inherent ionization quenching constant of the phototube under the current DC bias voltage from the pre-stored parameters; Calculate the deviation between the measured attenuation constant and the inherent ionization quenching constant of the hardware; According to the preset deviation-confidence mapping relationship, the deviation value is mapped to a physical confidence factor between 0 and 1, wherein the mapping relationship satisfies the following: the smaller the deviation value, the larger the physical confidence factor; when the deviation value exceeds the preset range, the physical confidence factor tends to 0.

4. The phototube high-voltage excitation method for detecting solar-blind ultraviolet discharge according to claim 1, characterized in that, The steps of screening suspected pulses based on the physical confidence factor to determine candidate valid pulses, performing energy quantization on each candidate valid pulse, and deconstructing and correcting the energy of candidate valid pulses with nonlinear stacking based on the rising edge slope distribution to obtain the effective discharge energy flow per unit time include: Possible pulses whose physical confidence factor is greater than a preset confidence threshold are selected as candidate valid pulses; For each candidate valid pulse, energy quantization is performed based on its waveform range, and the feedback resistance value of the pre-stored high-speed transimpedance sampling circuit and the internal ionization gain multiple of the phototube under the current DC bias are obtained. The voltage time integral value is then subjected to dimensional conversion and gain correction to obtain the apparent charge of the candidate valid pulse. Analyze the rising edge slope distribution of each candidate valid pulse, and detect whether there are waveform distortion features in the rising edge slope distribution that characterize the superposition of multiple avalanche processes; if so, determine that the candidate valid pulse has nonlinear stacking, and divide the corresponding pulse waveform into multiple sub-pulse units according to the waveform distortion features. Each of the sub-pulse units is subjected to energy quantization processing and corrected in combination with inherent hardware parameters to obtain the corresponding sub-unit charge. The charge quantity of the sub-unit is compared and verified with the apparent charge quantity of the corresponding candidate valid pulse to obtain the charge quantity of the sub-unit that has passed the verification. The corrected effective charge of the sub-unit is obtained by comparing the verified charge of the sub-unit with the physical confidence factor of the corresponding candidate effective pulse, and the effective discharge energy flow per unit time is obtained by accumulating them.

5. The phototube high-voltage excitation method for detecting solar-blind ultraviolet discharge according to claim 1, characterized in that, The steps of obtaining the time distribution characteristics of adjacent candidate effective pulses, performing regularity quantification analysis on the time distribution characteristics to obtain time-related entropy values, and then calculating the discharge stability coefficient based on the coupling relationship between the time-related entropy values ​​and the effective discharge energy flow include: Based on the timestamps of the candidate valid pulses, construct a time interval sequence of adjacent pulses; The probability distribution statistics of the time interval sequence are performed, and the time correlation entropy value of the time interval sequence is calculated based on the information entropy theory. The time correlation entropy value is used to quantify the regularity of the pulse time distribution. The discharge stability coefficient is calculated based on the coupling relationship between the time-related entropy value and the effective discharge energy flow. The discharge stability coefficient is used to comprehensively characterize the energy intensity and temporal regularity of the discharge process.

6. The phototube high-voltage excitation method for detecting solar-blind ultraviolet discharge according to claim 1, characterized in that, The steps of constructing a current state vector based on the effective discharge energy flow, the physical confidence factor, and the discharge stability coefficient, and calculating the dynamic hazard index based on the characteristic distance of the current state vector in the preset insulation degradation space and the slope of the historical state vector include: Calculate the mean of the physical confidence factor for all candidate valid pulses per unit time. The effective discharge energy flow, the mean of the physical confidence factor, and the discharge stability coefficient are normalized to construct the current state vector. Obtain a set of standard state vectors in a preset insulation degradation space, wherein the standard state vectors represent different risk levels; The instantaneous danger value of the current state is determined based on the similarity between the current state vector and each of the standard state vectors. The state vector sequences of the previous several time windows are read from the historical database, and the evolution trend of the state vector sequences is analyzed to obtain the rate of change of the danger level. The dynamic hazard index is calculated based on the instantaneous hazard value and the rate of change. The dynamic hazard index is used to reflect the current discharge risk and its development trend.

7. The phototube high-voltage excitation method for detecting solar-blind ultraviolet discharge according to claim 1, characterized in that, After the step of calculating the dynamic hazard index, the method further includes: Based on the relationship between the dynamic risk index and multiple preset risk thresholds, alarms of the corresponding level are triggered, and corresponding status information is uploaded.

8. The phototube high-voltage excitation method for detecting solar-blind ultraviolet discharge according to claim 7, characterized in that, After the step of calculating the dynamic hazard index, the process further includes executing an adaptive feedback adjustment procedure: The energy contribution ratio of low-confidence pulses is statistically analyzed in real time per unit time, which is used as the noise base level of the current time window. The low-confidence pulses refer to suspected pulses whose physical confidence factor is not greater than a preset confidence threshold. Based on the changing trends of the statistical characteristics of the physical confidence factor, the changing trends of the dynamic risk index, and the noise floor level, the type of the current abnormal state is comprehensively determined. Based on the type of the current abnormal state, a feedback adjustment command is generated and sent to the high-voltage excitation module to fine-tune the DC bias value of the phototube.

9. A phototube high-voltage excitation solar-blind ultraviolet discharge detection system, characterized in that, include: The ultraviolet detection unit adopts a back-illuminated solar-blind ultraviolet phototube with a built-in high-voltage ionization structure. Under the preset DC bias voltage provided by the high-voltage excitation module, the phototube converts the received solar-blind ultraviolet light signal into an avalanche pulse with inherent ionization quenching characteristics. The high-voltage excitation module, electrically connected to the ultraviolet detection unit, is used to provide an adjustable DC bias voltage to the built-in high-voltage ionization structure in order to control the internal ionization gain ratio. A high-speed transimpedance sampling circuit is directly connected to the output terminal of the ultraviolet detection unit to capture the avalanche electrical pulse and convert it into raw voltage sequence data; The edge computing core is connected to the high-speed transimpedance sampling circuit and has a built-in non-volatile memory. The memory pre-stores the hardware physical characteristic parameters of the phototube, including at least the inherent ionization quenching constant of the phototube.

10. The phototube high-voltage excitation solar-blind ultraviolet discharge detection system according to claim 9, characterized in that, The edge computing core is used to run the following logic modules in real time: The signal acquisition module is used to acquire the raw voltage sequence data collected in real time by the high-speed transimpedance sampling circuit; The data extraction module is used to locate the feature points of each suspected pulse based on the original voltage sequence data and extract the pulse waveform features; The pulse evaluation module is used to calculate the measured attenuation constant of each suspected pulse based on the falling edge attenuation characteristics, and compare the measured attenuation constant with the pre-stored hardware inherent ionization quenching constant to obtain the physical confidence factor of each suspected pulse. The screening and reconstruction module is used to screen suspected pulses according to the physical confidence factor, determine candidate valid pulses, perform energy quantization processing on each candidate valid pulse, and perform energy deconstruction and correction on the candidate valid pulses with nonlinear stacking according to the rising edge slope distribution to obtain the effective discharge energy flow per unit time. The stability analysis module is used to obtain the time distribution characteristics of adjacent candidate effective pulses, perform regularity quantification analysis on the time distribution characteristics to obtain the time-related entropy value, and then calculate the discharge stability coefficient based on the coupling relationship between the time-related entropy value and the effective discharge energy flow. The risk assessment module is used to construct a current state vector based on the effective discharge energy flow, the physical confidence factor, and the discharge stability coefficient, and to calculate a dynamic risk index based on the characteristic distance of the current state vector in the preset insulation degradation space and the slope of the historical state vector. The alarm communication module is used to trigger alarms of the corresponding level based on the relationship between the dynamic risk index and multiple preset risk thresholds, and upload the corresponding status information. The feedback adjustment module is used to generate a feedback adjustment command based on the changing trends of the physical confidence factor and the dynamic risk index, and send it to the high-voltage excitation module to fine-tune the DC bias value of the phototube.

Images