A real-time monitoring fault early warning method for an insulated bus

Abstract

The application relates to the technical field of insulation bus fault early warning, and discloses a real-time monitoring fault early warning method for an insulation bus, which comprises the following steps: S1, temperature sensors and partial discharge sensors are arranged at key points of the insulation bus; an anti-interference acquisition circuit and a temperature-compensated crystal oscillator differential synchronous acquisition mechanism are adopted to synchronously acquire temperature and partial discharge original signals; S2, the original signals of step S1 are input into an error self-correction model combined with Kalman filtering and LSTM (Long Short Term Memory) to perform calibration and dynamic compensation, and high-fidelity temperature and partial discharge characteristic data are output. The real-time monitoring fault early warning method for the insulation bus improves data fidelity from the source through the anti-interference acquisition circuit and the differential synchronous acquisition mechanism, realizes high-precision calibration of signals in combination with the error self-correction model, and applies the principle of magnetohydrodynamics to partial discharge monitoring to solve the problems that signals are easily submerged and the sensitivity is insufficient in traditional electric measurement methods.

Description

Technical Field

[0001] This invention relates to the field of insulated busbar fault early warning technology, and in particular to a real-time monitoring and fault early warning method for insulated busbars. Background Technology

[0002] Solid-insulated tubular busbars are widely used in power transmission and distribution systems of power plants, substations, and large industrial and mining enterprises due to their advantages such as compact structure, good insulation performance, and large current carrying capacity.

[0003] Traditional partial discharge monitoring methods have low signal fidelity in strong electromagnetic environments, making it difficult to capture weak physical field disturbances caused by discharge. Furthermore, assessments of a single physical quantity cannot fully reflect the complex evolution of insulation aging, leading to misjudgments of aging status and inaccurate predictions of remaining life, thus failing to provide reliable decision-making basis for maintenance personnel. Summary of the Invention

[0004] The technical problem to be solved by this invention is that the existing technology has the disadvantage of inaccurate prediction. To address this, we propose a real-time monitoring and fault early warning method for insulated busbars.

[0005] To achieve the above objectives, this application adopts the following technical solution: a real-time monitoring and fault early warning method for insulated busbars, comprising the following steps:

[0006] S1. Temperature sensors and partial discharge sensors are deployed at key points on the insulated busbar. An anti-interference acquisition circuit and a temperature-compensated crystal oscillator differential synchronous acquisition mechanism are used to synchronously acquire the original temperature and partial discharge signals.

[0007] S2, input the original signal from step S1 into the error self-correction model combining Kalman filtering and LSTM, perform calibration and dynamic compensation, and output high-fidelity temperature and partial discharge characteristic data.

[0008] S3 embeds a miniature MHD sensor in the busbar shell. It applies a constant magnetic field to the magnetic field generator through its electrode pair, collects the induced current generated by the partial discharge plasma to obtain electromagnetic characteristics, and simultaneously captures the flow field pattern through a high-speed camera to extract flow characteristics.

[0009] S4. The temperature data from step S2 is spatiotemporally synchronized with the dual-modal features from step S3. After the fluid characteristics are compensated by the actual temperature measured by thermocouples, a discharge risk assessment model is constructed through a Bayesian fusion algorithm to realize discharge location and intensity quantification.

[0010] S5, based on discharge risk indicators and temperature characteristic data, constructs an aging assessment model that integrates multiple physical quantities to quantitatively assess the aging status and estimate the remaining service life.

[0011] S6 compares the aging status and remaining lifespan with preset thresholds, and triggers a graded warning when the limit is exceeded.

[0012] Preferably, the temperature sensor is a PT1000 platinum resistance sensor with a sampling range of -40℃ to 150℃, a sampling accuracy of ±0.1℃, and an adjustable sampling frequency of 1Hz to 10Hz; the partial discharge sensor is a high-frequency current sensor with a sampling frequency band of 10kHz to 1MHz and a maximum sampling rate of ≥10MS / s.

[0013] Preferably, the error self-correction model has a built-in standardized error database, which contains three types of prior data: sensor factory calibration error, interference error under different electromagnetic environments, and drift error under different temperatures. The model performs online incremental updates to the database based on real-time operating data.

[0014] Preferably, the miniature MHD sensors are arranged in an array with a deployment spacing of 1m-3m and are encapsulated with IP67-level waterproof, dustproof and insulating properties.

[0015] Preferably, the flow features are extracted using an image processing algorithm, which includes four steps in sequence: image denoising, image enhancement, edge detection, and contour extraction.

[0016] Preferably, the temperature compensation is based on the ideal gas law, and the correlation between gas density and temperature is corrected by measuring the ambient temperature, thereby calibrating the fluid dynamics model; the Bayesian fusion algorithm treats electromagnetic features and fluid features as two independent sources of evidence, combines prior probability and conditional probability to calculate posterior probability, and selects the hypothesis with the highest posterior probability as the fusion evaluation result.

[0017] Preferably, the aging assessment model adopts any one of random forest, support vector machine or convolutional neural network, and is a joint classification and regression model. The input features of the model include partial discharge amplitude, discharge frequency, discharge phase distribution characteristics, fluid disturbance intensity, real-time temperature, temperature change rate and cumulative running time.

[0018] Preferably, the error self-correction model is jointly implemented using the Kalman filter algorithm and the LSTM long short-term memory neural network algorithm. The Kalman filter algorithm is used to dynamically filter out Gaussian white noise and random electromagnetic interference in the original signal, and the LSTM neural network algorithm is used to compensate for the time drift and temperature drift errors of the sensor online.

[0019] Preferably, the partial discharge location accuracy of the discharge risk assessment model is ≤5mm, and the discharge intensity quantification error is ≤8%.

[0020] Preferably, the aging assessment model divides the insulation aging state into four levels: normal state, slight degradation, moderate degradation, and severe degradation, and the graded warning corresponds to the four warning thresholds.

[0021] The technical effects and advantages of this invention are as follows:

[0022] This invention improves data fidelity from the source by using an anti-interference acquisition circuit and a differential synchronous acquisition mechanism, and achieves high-precision signal calibration by combining an error self-correction model. It applies magnetohydrodynamic principles to partial discharge monitoring, capturing plasma disturbances caused by discharge through an electromagnetic fluid dual-mode scheme, thus solving the problems of signal overload and insufficient sensitivity in traditional electrical measurement methods. Furthermore, it achieves probabilistic fusion of multimodal data through a Bayesian fusion algorithm, significantly improving discharge location accuracy and quantification accuracy. Finally, it integrates temperature and discharge multi-physical characteristics to construct an aging assessment and lifetime prediction model, enabling accurate assessment of insulation status and early fault warning. This effectively solves the technical challenges of existing technologies in data fidelity, discharge monitoring sensitivity, and aging assessment accuracy, significantly improving the reliability and safety of insulated busbar operation. Attached Figure Description

[0023] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts:

[0024] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0025] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0026] Reference Figure 1 As shown, the present invention provides a technical solution: a real-time monitoring and fault early warning method for insulated busbars, characterized by comprising the following steps:

[0027] S1, temperature sensors and partial discharge sensors are distributed at key monitoring points of the insulated busbar, including intermediate joints, weak sections of the main body insulation, and terminal connection parts. In response to the strong electromagnetic operating environment of the insulated busbar, the anti-interference acquisition circuit and differential synchronous acquisition mechanism synchronously acquire the original temperature and partial discharge signals of the insulated busbar during operation.

[0028] The dedicated anti-interference acquisition circuit includes, in sequence, a low-pass combined filter and a band-pass combined filter, a shielded drive unit and a low-noise instrument amplification stage; wherein the filter is used to filter out power frequency harmonics and high-frequency electromagnetic noise, the shielded drive unit eliminates the influence of parasitic capacitance of sensor cables through equipotential drive technology, and the amplification factor of the instrument amplification stage can be linearly adjusted in the range of 100 times to 1000 times to adapt to the acquisition requirements of weak signals.

[0029] The differential synchronous acquisition mechanism synchronizes all acquisition channels through a temperature-compensated crystal oscillator high-precision clock, reduces clock synchronization error, ensures strict alignment of temperature signal and partial discharge signal in the time dimension, and eliminates subsequent fusion errors caused by timing misalignment.

[0030] The temperature sensor uses a PT1000 platinum resistance sensor, with a sampling range of -40℃ to 150℃, a sampling accuracy of ±0.1℃, and an adjustable sampling frequency of 1Hz to 10Hz; the partial discharge sensor uses a high-frequency current sensor (HFCT), with a sampling frequency band of 10kHz to 1MHz, a maximum sampling rate of ≥10MS / s, and is suitable for the acquisition requirements of nanosecond-level pulse signals of partial discharge.

[0031] S2. Construct an error self-correction system model. The error self-correction system model has a built-in error compensation algorithm based on historical calibration data and insulation degradation theory analysis. Input the original temperature and partial discharge signal collected in step S1 into the error self-correction system model to perform high-precision calibration and dynamic compensation on the signal. Eliminate systematic and random errors introduced by strong electromagnetic interference, sensor zero-point drift, gain error, and changes in ambient temperature and humidity. Output high-fidelity temperature characteristic data and partial discharge characteristic data.

[0032] The error self-correction system model is jointly implemented using the Kalman filter algorithm and the LSTM long short-term memory neural network algorithm. The Kalman filter algorithm is used to dynamically filter out Gaussian white noise and random electromagnetic interference in the original signal, while the LSTM neural network algorithm is used to compensate for the time drift and temperature drift errors of the sensor online, so as to achieve high-fidelity signal restoration.

[0033] The self-correcting error system model has a built-in standardized error database, which contains three types of prior data: sensor factory calibration error, interference error under different electromagnetic environments, and drift error under different temperatures. The database can be updated online incrementally based on real-time operating data to improve the adaptability of error compensation under different operating scenarios.

[0034] S3: An array of MHD sensors is embedded in the insulated busbar shell to achieve blind-spot-free distributed monitoring of the entire busbar section; the core of the MHD sensor consists of symmetrically arranged electrode pairs and a uniform magnetic field generator. A constant uniform magnetic field is applied in the monitoring area through the magnetic field generator, and the magnetic field strength can be flexibly adjusted in the range of 0.1T-0.5T according to the busbar voltage level.

[0035] When partial discharge occurs inside the insulated busbar, the high-energy particles generated by the discharge ionize the surrounding air, forming a conductive plasma cloud. This alters the conductivity and flow field distribution of the local gas within the monitoring area. As the plasma cloud moves with the airflow in a constant uniform magnetic field, it cuts magnetic field lines, generating an induced current. The MHD sensor collects the dynamic changes of this induced current in real time through electrode pairs, extracting the amplitude, frequency, phase distribution, and duration of the induced current. Simultaneously, a miniature high-speed camera is deployed next to the MHD sensor. Through a pre-set highly insulated transparent observation window on the busbar shell, it captures in real time the changes in plasma streamlines, eddy current structure, and disturbance range caused by partial discharge. Image processing algorithms are used to extract the eddy current intensity, airflow velocity, and disturbance diffusion rate.

[0036] The MHD sensor is embedded inside the busbar housing without direct contact with the busbar body, which ensures electrical insulation performance and enables close-range, high-sensitivity sensing of discharge plasma; the array deployment spacing is 1m-3m, which can be flexibly adjusted according to the busbar voltage level and installation scenario;

[0037] The electrode pair is made of oxygen-free copper or gold-plated copper alloy, which has high conductivity and corrosion resistance. The uniform magnetic field generator uses a small neodymium iron boron permanent magnet, which does not require an additional power supply and can output a constant uniform magnetic field stably for a long time.

[0038] The miniature high-speed camera has a shooting frame rate of ≥1000fps and a resolution of ≥1080P. It is equipped with a macro lens and a supplementary lighting unit to meet the shooting requirements of the observation window. The observation window is made of high-transparency tempered glass or epoxy resin, and its insulation withstand voltage level is fully matched with the busbar shell, without affecting the overall insulation performance of the busbar.

[0039] The image processing algorithm consists of four steps: image noise reduction, image enhancement, edge detection, and contour extraction. It can effectively filter out interference from ambient light, dust, and other factors, and accurately extract fluid feature parameters.

[0040] S4. First, the high-fidelity temperature data output in step S2 is spatiotemporally synchronized with the electromagnetic features and fluid visualization features obtained in step S3. Based on the high-precision clock in step S1, the time synchronization error of the three types of data is ensured to be ≤1μs, and the spatial positions correspond one-to-one. At the same time, the actual ambient temperature of the monitoring area is collected in real time by the thermocouple integrated with the MHD sensor. Based on the ideal gas law, the changes in the fluid density and viscosity of the plasma are compensated in real time to correct the measurement error of the fluid characteristic parameters.

[0041] A Bayesian fusion algorithm is adopted, which treats the temperature-compensated electromagnetic features and fluid features as two independent sources of evidence. Combined with the prior probability and conditional probability obtained from the partial discharge calibration experiment, the posterior probability under different discharge intensities and different discharge location assumptions is calculated. The hypothesis with the highest posterior probability is selected as the final fusion evaluation result, and a probabilistic evaluation model for partial discharge risk is constructed to achieve accurate location of discharge and quantitative assessment of discharge intensity and hazard level.

[0042] S5. Based on the partial discharge risk quantification index fused from step S4, and combined with the long-term temperature time series characteristic data obtained in step S2, an insulation bus aging assessment model integrating multiple physical quantities and multi-modal characteristics is constructed. The model is used to quantitatively classify and assess the current insulation aging state of the insulation bus, and the expected remaining service life of the insulation bus is estimated through the remaining service life prediction algorithm built into the model.

[0043] The aging assessment model adopts any one of random forest, support vector machine or convolutional neural network, and is a joint classification and regression model. The input features of the model include at least: partial discharge amplitude, discharge frequency, discharge phase distribution characteristics, fluid disturbance intensity, real-time temperature, temperature change rate, and cumulative operating time. The model divides the insulation aging state into four levels: normal state, slight degradation, moderate degradation, and severe degradation.

[0044] The remaining service life prediction algorithm adopts the Arrhenius variable-temperature thermal life loss model or the data-driven degradation trajectory model based on LSTM network. The Arrhenius model calculates the thermal aging loss of the insulation material based on the temperature change rate and corrects the degradation rate by combining the discharge risk index. The data-driven model fits the insulation degradation trajectory based on long-term time-series monitoring data to achieve quantitative prediction of the remaining service life.

[0045] S6. Pre-set four-level early warning thresholds corresponding to insulation aging level and remaining service life. Compare the insulation aging status and remaining service life calculated in step S5 with the preset thresholds in real time. When the evaluation result exceeds the corresponding threshold, the system automatically triggers the corresponding level of graded early warning signal. At the same time, it outputs maintenance prompts including fault location, degree of deterioration, remaining service life, and maintenance suggestions, and pushes them to the local monitoring platform and mobile terminal of maintenance personnel to guide maintenance personnel to intervene in a timely manner and realize early warning of insulation faults.

[0046] The four-level early warning thresholds are specifically as follows:

[0047] Normal level, no abnormal discharge;

[0048] Attention level, slight degradation;

[0049] Warning level, moderate degradation;

[0050] Alarm level, severely degraded;

[0051] The early warning methods include local audible and visual warnings, pop-up warnings from cloud monitoring platforms, and SMS and APP push warnings from operation and maintenance terminals.

[0052] Example 1:

[0053] This embodiment addresses the intermediate joints and weak sections of the insulation of the insulated busbar (distance from the terminal).

[0054] Deployed in the location and terminal connection part Each monitoring unit integrates the following sensors and data acquisition modules:

[0055] The temperature sensor is a PT1000 platinum resistance temperature sensor, surface-mounted on the busbar casing. Its sensing range covers... In this embodiment, the sampling frequency is set to be The acquisition accuracy was calibrated to be .

[0056] The partial discharge sensor uses a high-frequency current sensor (HFCT) and is connected to the grounding wire of the busbar shielding layer. Its acquisition frequency band covers... The maximum sampling rate is set to It is adapted to the acquisition requirements of partial discharge pulse signals.

[0057] The MHD dual-modal sensor is an array of miniaturized MHD sensors embedded in the busbar housing, with a deployment spacing of [missing information]. The sensor core consists of symmetrically arranged oxygen-free copper electrode pairs and a small neodymium iron boron permanent magnet, with a constant uniform magnetic field strength applied. .

[0058] Miniature high-speed cameras are deployed next to each MHD sensor and take pictures through a pre-set high-transparency tempered glass viewing window on the busbar housing.

[0059] The data acquisition unit integrates a dedicated anti-interference acquisition circuit, including a low-pass combined filter and a band-pass combined filter, a shielded drive unit, and a low-noise instrumentation amplifier. All acquisition channels are synchronized with a temperature-compensated crystal oscillator for high-precision clocking.

[0060] After the system starts up, it synchronously collects raw signals of temperature and partial discharge. Taking a monitoring process that lasts for 30 minutes as an example:

[0061] PT1000 sensor output per minute Temperature values ​​were collected in total. Temperature data points. Initial stage bus temperature: As the load current increases, the temperature gradually rises to HFCT sensor with Continuous sampling at the sampling rate;

[0062] The collected raw data is input into the error self-correction system model, which is implemented by combining the Kalman filter algorithm and the LSTM neural network algorithm. The built-in standardized error database contains three types of prior data: sensor factory calibration error, interference error under different electromagnetic environments, and drift error under different temperatures.

[0063] Taking temperature signal processing as an example, in the original temperature sequence, electromagnetic interference causes... An abnormal jump occurred within seconds, and the temperature reading changed from... Change to This effectively filters out random interference. Simultaneously, the LSTM neural network, based on historical temperature drift data, performs online compensation for the sensor's nonlinear drift across the entire temperature range, outputting a corrected temperature value. .

[0064] For partial discharge signals, the smoothness of the waveform is improved by filtering out high-frequency noise through a Kalman filter on the original pulse waveform.

[0065] When a partial discharge occurs inside the busbar, the air within the discharge channel is ionized at the instant of discharge, forming a plasma cloud. In a constant magnetic field, the plasma cloud moves and cuts magnetic field lines, according to the Lorentz force formula... An induced electromotive force is generated on the electrodes perpendicular to the magnetic field and the flow direction. The electrode pairs of the MHD sensor collect the dynamic changes of the induced current in real time.

[0066] When the plasma cluster moves at speed magnetic field strength Electrode spacing The theoretical value of the induced electromotive force is approximately:

[0067] ;

[0068] The actual measured induced current passes through After magnification by 10 times, it becomes Converted to the sensor input terminal Meanwhile, a miniature high-speed camera captures the fluid visualization pattern within the observation window. Image processing algorithms sequentially perform image denoising, image enhancement, edge detection, and contour extraction, extracting the following flow feature parameters:

[0069] The maximum vorticity value in the image is ; Particle displacement calculation based on continuous frames, average airflow velocity Plasma clusters in The diffusion from the point source to the region;

[0070] First, spatiotemporal synchronization matching is performed. Based on a high-precision clock, the time synchronization error between the aforementioned MHD electromagnetic characteristics and the high-speed camera image sequence is <0.01ms, and the spatial location is the intermediate connector area where the monitoring unit is located.

[0071] Thermocouples collect real-time measured ambient temperature data for this area. Based on the ideal gas law Temperature compensation is applied to the fluid density. The air density variation factor compared to the initial calibration temperature is:

[0072] Therefore, fluid characteristic parameters such as eddy intensity and airflow velocity are modified accordingly. The modified eddy intensity is: ;

[0073] A Bayesian fusion algorithm was used for discharge risk assessment. Based on historical calibration experiments, a weak discharge occurred in the intermediate joint region, and the discharge amount was... Prior probability A moderate discharge occurs, with a discharge quantity of Prior probability Prior probability of no discharge

[0074] When a discharge hypothesis occurs At that time, electromagnetic characteristics were observed. Field probability Observe fluid characteristics conditional probability .

[0075] The calculation formula is as follows:

[0076] ;

[0077] After normalization, the hypothesis with the highest posterior probability is taken as the fusion evaluation result. Simultaneously, the calibration relationship between discharge quantity and characteristic parameters is considered. Estimate the discharge amount The discharge location is determined based on the time difference or amplitude ratio of the multi-sensor array. .

[0078] Based on the quantified partial discharge risk index from the fusion output, combined with medium- and long-term temperature time-series characteristic data, the data is input into the insulation aging assessment model. A joint random forest classification and regression model is adopted, with a 12-dimensional input feature vector. The model outputs the current insulation aging state, and the remaining lifetime prediction uses a thermal lifetime loss model based on the Arrhenius equation.

[0079] ,in Reference temperature The baseline lifespan, The activation energy of insulating materials, Let Boltzmann's constant be substituted into the actual operating temperature. The calculated thermal aging acceleration factor is:

[0080] ;

[0081] Taking into account discharge degradation correction factors The overall acceleration factor is:

[0082] ; Elapsed running time The equivalent aging time is calculated as follows: ;

[0083] The remaining life prediction value is: Prediction relative error .

[0084] The system is pre-set with four warning thresholds corresponding to the insulation aging level and remaining service life:

[0085] Normal grade, remaining service life aging index ;

[0086] Attention level, remaining service life aging index ;

[0087] Warning level, remaining service life aging index 。;

[0088] Alarm level, remaining service life aging index .

[0089] The insulation aging status and remaining life calculated in step S5 are compared with the aforementioned preset thresholds in real time. When the evaluation result exceeds the corresponding threshold, the system automatically triggers the corresponding level of graded early warning signal and pushes it to the maintenance personnel, realizing early warning of insulation faults.

[0090] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A real-time monitoring and fault early warning method for insulated busbars, characterized in that, Includes the following steps: S1. Temperature sensors and partial discharge sensors are deployed at key points on the insulated busbar. An anti-interference acquisition circuit and a temperature-compensated crystal oscillator differential synchronous acquisition mechanism are used to synchronously acquire the original temperature and partial discharge signals. S2, input the original signal from step S1 into the error self-correction model combining Kalman filtering and LSTM, perform calibration and dynamic compensation, and output high-fidelity temperature and partial discharge characteristic data. S3 embeds a miniature MHD sensor in the busbar shell. It applies a constant magnetic field to the magnetic field generator through its electrode pair, collects the induced current generated by the partial discharge plasma to obtain electromagnetic characteristics, and simultaneously captures the flow field pattern through a high-speed camera to extract flow characteristics. S4. The temperature data from step S2 is spatiotemporally synchronized with the dual-modal features from step S3. After the fluid characteristics are compensated by the actual temperature measured by thermocouples, a discharge risk assessment model is constructed through a Bayesian fusion algorithm to realize discharge location and intensity quantification. S5, based on discharge risk indicators and temperature characteristic data, constructs an aging assessment model that integrates multiple physical quantities to quantitatively assess the aging status and estimate the remaining service life. S6 compares the aging status and remaining lifespan with preset thresholds, and triggers a graded warning when the limit is exceeded.

2. The real-time monitoring and fault early warning method for insulated busbars according to claim 1, characterized in that: The temperature sensor uses a PT1000 platinum resistance sensor, with a sampling range of -40℃ to 150℃, a sampling accuracy of ±0.1℃, and an adjustable sampling frequency of 1Hz to 10Hz; the partial discharge sensor uses a high-frequency current sensor, with a sampling frequency band of 10kHz to 1MHz and a maximum sampling rate of ≥10MS / s.

3. The real-time monitoring and fault early warning method for insulated busbars according to claim 1, characterized in that: The self-correcting error model has a built-in standardized error database, which contains three types of prior data: sensor factory calibration error, interference error under different electromagnetic environments, and drift error under different temperatures. The model performs online incremental updates to the database based on real-time operating data.

4. The real-time monitoring and fault early warning method for insulated busbars according to claim 1, characterized in that: The miniature MHD sensors are arranged in an array with a deployment spacing of 1m-3m and are encapsulated with IP67-level waterproof, dustproof and insulating properties.

5. The real-time monitoring and fault early warning method for insulated busbars according to claim 1, characterized in that: The flow features are extracted using an image processing algorithm, which includes four steps in sequence: image denoising, image enhancement, edge detection, and contour extraction.

6. The real-time monitoring and fault early warning method for insulated busbars according to claim 1, characterized in that: The temperature compensation is based on the ideal gas law, and the correlation between gas density and temperature is corrected by measuring the ambient temperature, thereby calibrating the fluid dynamics model. The Bayesian fusion algorithm treats electromagnetic features and fluid features as two independent sources of evidence, combines prior probability and conditional probability to calculate posterior probability, and selects the hypothesis with the highest posterior probability as the fusion evaluation result.

7. The real-time monitoring and fault early warning method for insulated busbars according to claim 1, characterized in that: The aging assessment model adopts any one of random forest, support vector machine or convolutional neural network, and is a joint classification and regression model. The input features of the model include partial discharge amplitude, discharge frequency, discharge phase distribution characteristics, fluid disturbance intensity, real-time temperature, temperature change rate and cumulative running time.

8. The real-time monitoring and fault early warning method for insulated busbars according to claim 3, characterized in that: The error self-correction model is implemented by combining the Kalman filter algorithm and the LSTM long short-term memory neural network algorithm. The Kalman filter algorithm is used to dynamically filter out Gaussian white noise and random electromagnetic interference in the original signal, while the LSTM neural network algorithm is used to compensate for the time drift and temperature drift errors of the sensor online.

9. The real-time monitoring and fault early warning method for insulated busbars according to claim 1, characterized in that: The partial discharge location accuracy of the discharge risk assessment model is ≤5mm, and the discharge intensity quantification error is ≤8%.

10. The real-time monitoring and fault early warning method for insulated busbars according to claim 1, characterized in that: The aging assessment model classifies the insulation aging state into four levels: normal state, slight degradation, moderate degradation, and severe degradation. The graded warning corresponds to the four warning thresholds.

Images