Evaluation method and test device for combined stress aging state of hot chamber high-voltage electrical penetration assembly

Abstract

The invention discloses a hot chamber high-voltage electrical penetration assembly composite stress aging state evaluation method and a test device, and relates to the technical field of electrical equipment aging tests, and the method specifically comprises the steps: synchronously obtaining the leakage current, the partial discharge amplitude spectrum, the temperature distribution matrix and the high-frequency current spectrum of a to-be-tested electrical penetration assembly; extracting a time domain feature from the leakage current, extracting a frequency domain feature from the partial discharge spectrum, and extracting a space gradient feature from the temperature field; carrying out weighted fusion on the time domain features, the frequency domain features and the spatial gradient features by adopting an attention mechanism; reducing the 128-dimensional features after the attention mechanism fusion to three-dimensional spatial features; processing the fused three-dimensional space characteristics by adopting a bidirectional long-term memory network to obtain a residual life prediction value of the electrical penetration assembly; judging the aging stage of the electrical penetration assembly according to the residual life prediction value; presetting an aging threshold value, and when the predicted aging stage exceeds the threshold value, automatically adjusting the temperature and humidity parameters of the environmental test box to a safe working condition; the method is used for solving the problem that a traditional evaluation result is inaccurate.

Description

Technical Field

[0001] The present invention relates to the technical field of aging tests for electric power equipment, and in particular to a composite stress aging state assessment method and a testing device for a high-voltage electrical penetration component in a hot chamber. Background Art

[0002] The high-voltage electrical penetrations in a hot cell are part of the hot cell wall. Relying on excellent shielding and sealing capabilities, they form a safety barrier to prevent the leakage of radioactive materials within the hot cell and safeguard the health of operating personnel. High-voltage conductor assemblies are core components that ensure the reliability of these high-voltage electrical penetrations. They are typically made of composite insulating materials. During operation, they are affected by factors such as electricity, heat, and radiation, causing performance degradation, impacting the reliability of the entire electrical penetration and even the hot cell system. In particular, the recent improvement in design standards for newly built hot cells, such as increasing voltage levels and harsher working environments, has placed higher demands on the reliability of these high-voltage electrical penetrations. Therefore, it is extremely important to assess the aging status of these high-voltage electrical penetrations in hot cells.

[0003] However, due to the generally long lifespan of composite insulation materials, real-time aging status assessment is not feasible. Therefore, it is necessary to build an accelerated aging device in the laboratory to conduct accelerated aging tests, and then use the corresponding mathematical model to evaluate the aging status. However, existing aging devices are not flexible and versatile enough. Although they can simulate several aging factors simultaneously, they cannot achieve multi-stress coordinated loading and multi-parameter correlation analysis, making it difficult to accurately assess the aging status and difficult to fully apply to actual engineering.

[0004] The Chinese patent publication number CN111650540A discloses an intelligent state monitoring device and method for main transformers based on multi-source data. The device includes multiple transformer state monitoring, and each monitoring network monitors multiple data. When one of the monitoring data is abnormal, the edge IoT agent device obtains the monitoring data from the monitoring network with a linkage relationship, thereby integrating the multi-source monitoring data to make a comprehensive judgment on the state of the main transformer, which can achieve a more accurate state judgment. At the same time, the invention sets a conventional monitoring mode and an intensive care mode for each type of monitoring device. When it is initially determined that an abnormal signal has occurred, the monitoring device with a linkage relationship is switched from the conventional monitoring mode to the intensive care mode, increasing the frequency of data collection and data transmission, thereby utilizing more data to further judge the state of the main transformer and further improving the accuracy of the state judgment. However, when the multi-source detection data is used to judge the state of the main transformer in this method, multiple state judgments are still performed in parallel, resulting in inaccurate evaluation results.

[0005] Therefore, we propose a state assessment method and test device with accurate assessment results. Summary of the Invention

[0006] The purpose of the present invention is to provide a method and a test device for evaluating the composite stress aging state of high-voltage electrical penetrations in a hot chamber, which are used to solve the problem of inaccurate traditional evaluation results.

[0007] The present invention is achieved through the following technical solutions:

[0008] A method for evaluating the composite stress aging state of a hot chamber high-voltage electrical penetration component, specifically comprising:

[0009] Synchronously acquire the leakage current I(t), partial discharge amplitude spectrum S(f,t), temperature distribution matrix T(x,y,t) and high-frequency current spectrum of the electrical penetration component to be tested, where t is the timestamp, f is the frequency, and x,y are the spatial coordinates;

[0010] Extract time domain features from leakage current I(t), frequency domain features from partial discharge spectrum S(f,t), and spatial gradient features from temperature field T(x,y,t);

[0011] The attention mechanism is used to weightedly fuse time domain features, frequency domain features and spatial gradient features;

[0012] The 128-dimensional features fused by the attention mechanism are embedded into 3-dimensional spatial features through t-distributed random neighborhood embedding;

[0013] A bidirectional long-term memory network is used to process the fused 3D spatial features to obtain the remaining life prediction value of electrical penetration parts.

[0014] Determine the aging stage of electrical penetrations based on the predicted remaining life of the electrical penetrations;

[0015] A preset aging threshold value is set. When the predicted aging stage exceeds the threshold value, the temperature and humidity parameters of the environmental test chamber are automatically adjusted to safe working conditions.

[0016] Furthermore, the leakage current I(t) is decomposed into five IMF components according to the intrinsic mode function, and the sample entropy of the 2nd to 4th IMF components is extracted as the time domain feature input to the attention mechanism.

[0017] Furthermore, when the temperature gradient of the temperature distribution matrix is greater than 5°C / cm and a characteristic peak appears in the high-frequency current spectrum within the range of 3-10 MHz, it is defined as corona discharge, and the intensity of the corona discharge is linearly correlated with the energy value of the third IMF component.

[0018] Furthermore, the frequency domain characteristics include a quality factor Q value of each 100 kHz sub-band in the 0.5-2 MHz frequency band.

[0019] Furthermore, the calculation formula for weighted fusion of time domain features, frequency domain features and spatial gradient features using the attention mechanism is:

[0020] α i =soft max(W T tanh(V·h i ));

[0021] F fused =Σα i h i ;

[0022] where h i is a single modal feature, W and V are training parameters.

[0023] Furthermore, the attention mechanism adopts multi-head attention, and the number of heads k=4.

[0024] Furthermore, a bidirectional long-term memory network is used to process the fused 3D spatial features to obtain the remaining life prediction value of the electrical penetration parts. The calculation formula is:

[0025] RUL=f LSTM (F fused ;θ)+ε CNN (T(x,y,t));

[0026] Among them, RUL is the predicted value of the remaining life of the electrical penetration, θ is the network parameter, and ε is the temperature field correction term extracted by the convolutional neural network.

[0027] Furthermore, the hidden layer dimension of the bidirectional LSTM network is 64, and the dropout rate is 0.3.

[0028] Furthermore, the perplexity in the t-distributed random neighborhood embedding is set to 6% to 10% of the total number of samples.

[0029] A hot chamber high-voltage electrical penetration composite stress aging test device includes an operating workstation, a control cabinet, a test and detection system, a voltage regulating and transforming device, and an environmental test chamber. The operating workstation is connected to the control cabinet via a communication control cable, and the control cabinet is connected to the test and detection system and the voltage regulating and transforming device via communication control cables. The control cabinet is also electrically connected to the voltage regulating and transforming device via a power cable.

[0030] The test detection system includes an insulation resistance detection device, a high-voltage isolation cabinet, a high-frequency current monitoring device, a partial discharge detection device, a temperature measurement and acquisition device, and a leakage current detection device; the sensor system is installed in an environmental test chamber, and the sensor system includes a high-frequency current sensor, an ultra-high frequency and ultrasonic partial discharge sensor, a thermocouple sensor, and a leakage current sensor;

[0031] The power input end of the insulation resistance detection device is electrically connected to the voltage regulating and transforming equipment through a power cable, and the power output end of the insulation resistance detection device is electrically connected to the high-frequency current sensor and the leakage current sensor through a high-voltage isolation cabinet, respectively. The high-frequency current sensor is electrically connected to the positive electrode of the electrical penetration component to be tested, and the leakage current sensor is electrically connected to the negative electrode of the electrical penetration component to be tested;

[0032] The detection input end of the high-frequency current monitoring device is signal-connected to the detection output end of the high-frequency current sensor; the detection input end of the partial discharge detection device is signal-connected to the detection output end of the ultra-high frequency and ultrasonic partial discharge sensor; the detection input end of the temperature measurement and acquisition device is signal-connected to the detection output end of the thermocouple sensor; and the detection input end of the leakage current detection device is signal-connected to the detection output end of the leakage current sensor.

[0033] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0034] The present invention discloses a composite stress aging state assessment method and test device for high-voltage electrical penetrations in a hot chamber. By synchronously collecting leakage current, partial discharge spectrum, temperature field, and high-frequency current spectrum, this method implements multi-stress coupling loading of electric field, temperature and humidity, and time series, more realistically simulating actual working conditions and overcoming the limitations of traditional single stress tests (such as voltage or temperature and humidity only).

[0035] By using a weighted fusion method to analyze the correlation between high-frequency current spectrum characteristics and temperature field mutations, potential defects that cannot be discovered by traditional single detection are identified. This solves the problem of misjudgment caused by traditional methods relying on single parameter detection and improves the accuracy of evaluation results.

[0036] In addition, the 128-dimensional fusion features are reduced to 3 dimensions, which significantly simplifies the data complexity while retaining key discriminant information, thereby optimizing the input structure of the bidirectional LSTM, reducing redundant calculations, and improving the model convergence speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A schematic flow chart of a method of the present invention is shown;

[0038] Figure 2 A schematic diagram of the structure of a device of the present invention;

[0039] Figure 3 It is a structural schematic diagram of the electronic device in the present invention. DETAILED DESCRIPTION

[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0041] Example 1

[0042] like Figure 1 A method for evaluating the composite stress aging state of a high-voltage electrical penetration component in a hot chamber is shown, specifically comprising:

[0043] Synchronously acquire the leakage current I(t), partial discharge amplitude spectrum S(f,t), temperature distribution matrix T(x,y,t) and high-frequency current spectrum of the electrical penetration component to be tested, where t is the timestamp, f is the frequency, and x,y are the spatial coordinates;

[0044] The leakage current I(t), partial discharge amplitude spectrum S(f,t), temperature distribution matrix T(x,y,t) and high-frequency current spectrum are obtained by collecting data from the corresponding leakage current sensor, ultra-high frequency and ultrasonic partial discharge sensor, thermocouple sensor and high-frequency current sensor in the environmental test chamber respectively by the leakage current detection device, partial discharge detection device, temperature measurement and acquisition device and high-frequency current monitoring device.

[0045] In addition, the leakage current I(t), partial discharge amplitude spectrum S(f,t), temperature distribution matrix T(x,y,t), and high-frequency current spectrum are all test parameters for stress aging testing of the electrical penetrations under test. They are used to reflect the leakage current, partial discharge level, temperature distribution during the test, and high-frequency input current level of the electrical penetrations, respectively. In addition, the environmental test chamber can simulate the actual temperature and humidity environment, which is achieved by using a humidifier and heater.

[0046] Extract time domain features from leakage current I(t), frequency domain features from partial discharge spectrum S(f,t), and spatial gradient features from temperature field T(x,y,t);

[0047] The time domain features of leakage current I(t) are extracted by using mean, kurtosis and intrinsic mode function (IMF) energy of Hilbert-Huang transform.

[0048] The process of extracting frequency domain features of the partial discharge spectrum S(f,t) is to extract wavelet packet energy entropy in the 1-10 MHz frequency band to identify the partial discharge spectrum S(f,t) and extract frequency domain features; and the frequency domain features include the quality factor Q value of each 100 kHz sub-band in the 0.5-2 MHz frequency band;

[0049] The process of extracting spatial gradient features of the temperature field T(x,y,t) is to use the spatial gradient (maximum temperature difference ΔT max , hotspot moving rate v hotspot ) Capture local overheating defects of the electrical penetration component under test and quantify local abnormalities with significantly improved sensitivity;

[0050] The attention mechanism is used to weightedly fuse time domain features, frequency domain features and spatial gradient features;

[0051] The specific calculation formula is:

[0052] α i =soft max(W T tanh(V·h i ));

[0053] F fused =Σα i h i ;

[0054] where h i is a single modal feature, i.e., leakage current I(t), partial discharge amplitude spectrum S(f,t) or temperature distribution matrix T(x,y,t), W and V are training parameters;

[0055] In addition, the attention mechanism adopts multi-head attention (Multi-Head Attention), and the number of heads k = 4;

[0056] The 128-dimensional features fused with the attention mechanism are reduced to three-dimensional spatial features through t-distributed Stochastic Neighbor Embedding (t-SNE). Healthy and aged samples are linearly separable in the low-dimensional space (cluster margin > 0.87), significantly simplifying data complexity while retaining key discriminant information. This feature optimizes the data structure required for subsequent input into the bidirectional LSTM, reduces redundant computation, and improves model convergence speed. In particular, the perplexity in the t-SNE is set to 6% to 10% of the total number of samples and can be dynamically adjusted based on the data size, avoiding bias caused by manual parameter adjustment. It ensures the preservation of local details, capturing subtle differences between aged and healthy samples (such as transient fluctuations in leakage current), and ensures global distribution stability, maintaining the overall clustering structure of different aging stages (such as the separation of healthy, initially degraded, and severely aged samples).

[0057] A bidirectional long-term memory (LSTM) network is used to process the fused 3D spatial features to obtain the remaining life prediction value of the electrical penetration parts.

[0058] The calculation formula is:

[0059] RUL=f LSTM (F fused ;θ)+ε CNN (T(x,y,t));

[0060] Where RUL is the remaining useful life prediction value (RUL) of the electrical penetration fitting, θ is the network parameter, and ε is the temperature field correction term extracted by the convolutional neural network (CNN).

[0061] In addition, the bidirectional LSTM network has a hidden layer dimension of 64 and a dropout rate of 0.3, which prevents overfitting and can effectively capture the temporal dependencies of the aging process (such as the correlation between the sudden increase in leakage current and the temperature hysteresis).

[0062] The aging stage of the electrical penetrations is determined based on the predicted remaining life of the electrical penetrations. Certain thresholds are preset for healthy, initial degradation, and severe aging. The aging stage of the electrical penetrations is determined based on which of the two thresholds the predicted remaining life of the electrical penetrations falls between.

[0063] Preset aging thresholds. When the predicted aging stage exceeds the threshold, the temperature and humidity parameters of the environmental test chamber are automatically adjusted to safe working conditions.

[0064] When the leakage current exceeds the threshold, the humidity is automatically reduced to avoid flashover;

[0065] And when the leakage current exceeds the preset threshold, it automatically reduces the humidity and triggers partial discharge detection, while avoiding flashover.

[0066] Example 2

[0067] like Figure 2 In one embodiment shown, the leakage current I(t) is decomposed into five IMF components according to the intrinsic mode function, and the sample entropy of the 2nd to 4th IMF components is extracted as time-domain features and input into the attention mechanism. In addition, the first IMF component is a high-frequency noise component and is not related to aging. The second to fourth IMF components are medium-high frequency components, corresponding to the transient current characteristics of key aging processes such as partial discharge and corona discharge of insulating materials. The fifth IMF component is a low-frequency, slowly varying component and has a weak correlation with aging. Therefore, selecting the second to fourth IMF components can effectively filter out non-critical noise and low-frequency interference, retaining the characteristic frequency band directly related to insulation degradation. Sample entropy is used to quantify the complexity of the signal and is highly sensitive to small changes in non-stationary signals.

[0068] In addition, if all five IMF components are used, high-frequency noise (IMF1) and low-frequency redundancy (IMF5) may be introduced, leading to model overfitting. Therefore, extracting only the sample entropy of the 2nd to 4th IMF components can significantly reduce the input feature dimension (from 5 dimensions to 3 dimensions), thereby reducing model complexity. In addition, the attention mechanism fuses different modal features (time domain, frequency domain, and space) through weighted fusion. The sample entropy of the 2nd to 4th IMF components is the core feature in the time domain, and its contribution to the aging state can be highlighted through the attention weight.

[0069] In addition, when the temperature gradient of the temperature distribution matrix is greater than 5°C / cm and the high-frequency current spectrum has a characteristic peak in the range of 3-10 MHz, it is defined as corona discharge, and the intensity of the corona discharge is linearly correlated with the energy value of the third IMF component. This is because the sample entropy of the third IMF component is strongly correlated with the corona discharge intensity (its determination coefficient R 2 =0.92), which can directly reflect the degree of insulation damage caused by discharge events. The sample entropy of the second and fourth IMF components characterizes the suddenness and randomness of partial discharge, supplementing the detection of potential defects. In addition, corona discharge can identify potential defects that cannot be discovered by traditional single detection through the correlation analysis of high-frequency current spectrum characteristics and temperature field mutations.

[0070] Example 3

[0071] A hot chamber high-voltage electrical penetration composite stress aging test device includes an operating workstation, a control cabinet, a test and detection system, a voltage regulating and transforming device, and an environmental test chamber. The operating workstation is connected to the control cabinet via a communication control cable, and the control cabinet is respectively connected to the test and detection system and the voltage regulating and transforming device via communication control cables. The control cabinet is also electrically connected to the voltage regulating and transforming device via a power cable. This ensures the rapid issuance and execution of control instructions (such as voltage regulation and temperature and humidity setting). At the same time, test data (leakage current, partial discharge spectrum, temperature field, etc.) is centrally uploaded to the operating workstation via the control cabinet, realizing full-process digital management. The control cabinet serves as an intermediate hub, independently managing the communication and power supply of the test and detection system, the voltage regulating and transforming device, and the environmental test chamber, avoiding signal interference and improving system stability.

[0072] In addition, the voltage regulating and transforming equipment has the function of multi-mode voltage loading. Its power frequency mode is a 50Hz / 35kV sine wave voltage, which can simulate the long-term operating conditions of electrical penetrations.

[0073] The high-frequency mode is a 12kHz / ±15kV square wave voltage, which is used to accelerate the aging process of the insulation material in the electrical penetrations.

[0074] This enables the voltage-regulating and transformer equipment to support a flexible combination of electric field stress and time series to meet the needs of different aging tests (such as accelerated aging and fatigue testing). It can also quickly adjust the output voltage through the control cabinet based on real-time detection parameters (such as leakage current threshold triggering) to achieve closed-loop control of stress loading.

[0075] The test detection system includes an insulation resistance detection device, a high-voltage isolation cabinet, a high-frequency current monitoring device, a partial discharge detection device, a temperature measurement and acquisition device, and a leakage current detection device; the sensor system is installed in an environmental test chamber, and the sensor system includes a high-frequency current sensor, an ultra-high frequency and ultrasonic partial discharge sensor, a thermocouple sensor, and a leakage current sensor;

[0076] The insulation resistance test device can prevent damage to the insulation resistance test device caused by the high-voltage test system, and the relevant insulation resistance data can also be used to verify the insulation condition of the electrical penetration fittings under test. In addition, the environmental test chamber has the function of multi-stress coupling loading. That is, the environmental test chamber simulates the actual working conditions of the hot chamber by adjusting the temperature and humidity (such as high temperature and high humidity, low temperature and dryness). Combined with the electric field stress (output of the voltage regulating device) and the time series, it realizes the coordinated loading of electric, thermal and humidity multi-stress, which is closer to the actual aging environment.

[0077] The system also has a real-time feedback control function. When it detects that the leakage current exceeds the limit or the temperature gradient suddenly increases, it automatically adjusts the temperature and humidity to safe operating conditions (such as reducing the humidity to below 60%), suppressing the risk of flashover and extending the life of the electrical penetration component under test.

[0078] The power input end of the insulation resistance detection device is electrically connected to the voltage regulating and transforming equipment through a power cable, and the power output end of the insulation resistance detection device is electrically connected to the high-frequency current sensor and the leakage current sensor through a high-voltage isolation cabinet, respectively. The high-frequency current sensor is electrically connected to the positive electrode of the electrical penetration component to be tested, and the leakage current sensor is electrically connected to the negative electrode of the electrical penetration component to be tested;

[0079] The detection input end of the high-frequency current monitoring device is signal-connected to the detection output end of the high-frequency current sensor; the detection input end of the partial discharge detection device is signal-connected to the detection output end of the ultra-high frequency and ultrasonic partial discharge sensor; the detection input end of the temperature measurement and acquisition device is signal-connected to the detection output end of the thermocouple sensor; the detection input end of the leakage current detection device is signal-connected to the detection output end of the leakage current sensor;

[0080] Among them, the high-frequency current sensor is installed at the positive pole of the electrical penetration part to be tested, directly capturing the high-frequency discharge signal, with strong anti-power frequency interference capability and an improvement of the signal-to-noise ratio by more than 30%;

[0081] Ultra-high frequency and ultrasonic partial discharge sensors are installed at multiple angles on the inner wall of the environmental test chamber to support azimuth identification of discharge signals, such as the spatial distribution differences between surface discharge and internal discharge.

[0082] Thermocouple sensor array covers the surface of the electrical penetration to be tested and the interior of the environmental chamber, and ΔT is calculated by spatial gradient max , achieving millimeter-level positioning of local overheating defects;

[0083] The leakage current sensor is installed at the negative pole of the electrical penetration piece to be tested, and works in conjunction with the insulation resistance detection device. Combined with the high-voltage isolation design, it ensures accurate measurement of microampere-level leakage current.

[0084] Example 4

[0085] As attached Figure 3 An electronic device as shown is characterized by comprising:

[0086] Processor, memory, communication interface;

[0087] The memory is used to store executable instructions of the processor;

[0088] The processor is configured to execute the above-mentioned method for evaluating the composite stress aging state of high-voltage electrical penetration parts in a hot chamber by executing the executable instructions.

[0089] A readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the above-mentioned method for evaluating the composite stress aging state of high-voltage electrical penetration parts in a hot chamber is implemented.

[0090] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A method for evaluating the composite stress aging state of high-voltage electrical penetrations in hot chambers, characterized in that: Specifically include: Synchronously obtain the leakage current, partial discharge amplitude spectrum, temperature distribution matrix and high-frequency current spectrum of the electrical penetration component to be tested; Extract time domain features for leakage current, frequency domain features for partial discharge spectrum, and spatial gradient features for temperature field; The attention mechanism is used to weightedly fuse time domain features, frequency domain features and spatial gradient features; The 128-dimensional features fused by the attention mechanism are embedded into 3-dimensional spatial features through t-distributed random neighborhood embedding; A bidirectional long-term memory network is used to process the fused 3D spatial features to obtain the remaining life prediction value of electrical penetration parts. Determine the aging stage of electrical penetrations based on the predicted remaining life of the electrical penetrations; A preset aging threshold value is set. When the predicted aging stage exceeds the threshold value, the temperature and humidity parameters of the environmental test chamber are automatically adjusted to safe working conditions.

2. The method for evaluating the composite stress aging state of a hot chamber high-voltage electrical penetration according to claim 1, characterized in that: The leakage current I(t) is decomposed into five IMF components according to the intrinsic mode function, and the sample entropy of the 2nd to 4th IMF components is extracted as the time domain feature and input into the attention mechanism.

3. The method for evaluating the composite stress aging state of a hot chamber high-voltage electrical penetration according to claim 2, characterized in that: When the temperature gradient of the temperature distribution matrix T(x, y, t) is greater than 5°C / cm and a characteristic peak appears in the high-frequency current spectrum within the range of 3-10 MHz, it is defined as corona discharge, and the intensity of the corona discharge is linearly correlated with the energy value of the third IMF component.

4. The method for evaluating the composite stress aging state of a hot chamber high-voltage electrical penetration according to claim 1, characterized in that: The frequency domain characteristics of the partial discharge amplitude spectrum S(f,t) include a quality factor Q value of each 100 kHz sub-band in the 0.5-2 MHz frequency band.

5. The method for evaluating the composite stress aging state of a hot chamber high-voltage electrical penetration according to claim 1, characterized in that: The calculation formula for weighted fusion of time domain features, frequency domain features and spatial gradient features using the attention mechanism is: α i =soft max(W T tanh(V·h i )); F fused =∑α i h i ; where h i is a single modal feature, W and V are training parameters.

6. The method for evaluating the composite stress aging state of a hot chamber high-voltage electrical penetration according to claim 1, characterized in that: The attention mechanism adopts multi-head attention, and the number of heads k=4.

7. The method for evaluating the composite stress aging state of a hot chamber high-voltage electrical penetration according to claim 5, characterized in that: The bidirectional long-term memory network is used to process the fused 3D spatial features to obtain the remaining life prediction value of the electrical penetration parts. The calculation formula is: RUL=f LSTM (F fused ;θ)+ε CNN (T(x,y,t)); Among them, RUL is the predicted value of the remaining life of the electrical penetration parts, θ is the network parameter, and ε is the temperature field correction term extracted by the convolutional neural network.

8. The method for evaluating the composite stress aging state of a hot chamber high-voltage electrical penetration according to claim 7, characterized in that: The hidden layer dimension of the bidirectional LSTM network is 64, and the dropout rate is 0.

3.

9. The method for evaluating the composite stress aging state of a hot chamber high-voltage electrical penetration according to claim 1, characterized in that: The perplexity in the t-distributed random neighborhood embedding is set to 6% to 10% of the total number of samples.

10. A hot chamber high voltage electrical penetration composite stress aging test device, which implements the hot chamber high voltage electrical penetration composite stress aging state assessment method according to any one of claims 1 to 9, characterized in that: It includes an operating workstation, a control cabinet, a test and detection system, a voltage regulating and transforming device, and an environmental test chamber. The operating workstation is connected to the control cabinet signal via a communication control cable, the control cabinet is respectively connected to the test and detection system and the voltage regulating and transforming device signal via a communication control cable, and the control cabinet is electrically connected to the voltage regulating and transforming device via a power cable; The test detection system includes an insulation resistance detection device, a high-voltage isolation cabinet, a high-frequency current monitoring device, a partial discharge detection device, a temperature measurement and acquisition device, and a leakage current detection device; the sensor system is installed in an environmental test chamber, and the sensor system includes a high-frequency current sensor, an ultra-high frequency and ultrasonic partial discharge sensor, a thermocouple sensor, and a leakage current sensor; The power input end of the insulation resistance detection device is electrically connected to the voltage regulating and transforming equipment through a power cable, and the power output end of the insulation resistance detection device is electrically connected to the high-frequency current sensor and the leakage current sensor through a high-voltage isolation cabinet, respectively. The high-frequency current sensor is electrically connected to the positive electrode of the electrical penetration component to be tested, and the leakage current sensor is electrically connected to the negative electrode of the electrical penetration component to be tested; The detection input end of the high-frequency current monitoring device is signal-connected to the detection output end of the high-frequency current sensor; the detection input end of the partial discharge detection device is signal-connected to the detection output end of the ultra-high frequency and ultrasonic partial discharge sensor; the detection input end of the temperature measurement and acquisition device is signal-connected to the detection output end of the thermocouple sensor; and the detection input end of the leakage current detection device is signal-connected to the detection output end of the leakage current sensor.

Images