Intelligent switching method for GIS power frequency and partial discharge test

Abstract

This invention relates to an intelligent switching method for GIS power frequency and partial discharge testing, comprising the following steps: Step S1, test task analysis and switching path planning: analyzing the test task instructions, combining the power frequency withstand voltage test and partial discharge test of the test chamber into a continuous test sequence, and planning the switching path between terminals; Step S2, mechanical switching execution and multi-source signal synchronous acquisition: responding to the switching path, performing a mechanical switching operation to disconnect the test lead from the current primary terminal and connect it to the target primary terminal, while simultaneously acquiring the insertion / extraction force timing signal and the contact circuit impedance timing signal during the switching process; Step S3, feature vector construction: extracting the force energy feature component from the insertion / extraction force timing signal; This invention enables intelligent switching in GIS testing, improving the accuracy of wiring quality assessment and test safety.

Description

Technical Field

[0001] This invention relates to the field of GIS equipment, and more specifically to a smart switching cable method for GIS power frequency and partial discharge testing. Background Technology

[0002] During factory testing, field acceptance testing, and periodic preventative testing of GIS equipment, power frequency withstand voltage testing and partial discharge testing must be carried out sequentially on each gas chamber. The terminals for these two types of tests are usually located in different positions, and the applied voltage levels and test circuit parameters are different. Therefore, after completing one test item, the high-voltage test lead must be disconnected from the current terminal and switched to another terminal. Currently, this switching operation mainly relies on manual methods. Operators need to enter the test area to manually disconnect and install leads and switch the corresponding grounding and measurement circuits. The entire process is time-consuming and labor-intensive. Furthermore, given the compact structure and dense terminal arrangement of GIS equipment, manual switching poses risks to personal safety and test quality, such as accidental contact with nearby live parts, inadequate wiring tightening, and failure to remove oxide film from terminal surfaces.

[0003] In attempts to improve automation of connection switching, existing technologies mostly use robotic arms with preset fixed paths or simple switching switches to perform the switching, focusing only on the execution of the action itself and lacking the ability to effectively perceive and diagnose the wiring quality. A few solutions judge the wiring qualification by measuring a single contact resistance after the switching, but this method can only reflect the static contact state and is difficult to identify transient poor contact caused by terminal wear, contact finger bounce or foreign object intervention. It also cannot identify measurement distortion caused by abnormal wiring shape or sensor failure. Once a misjudgment occurs, it may lead to overheating and burning of contact points or even insulation breakdown accidents during high-voltage testing.

[0004] In addition, repeated insertion and removal of GIS terminals inevitably cause wear on the contact surface and relaxation of elastic elements, resulting in a gradual increase in contact resistance. Existing technologies lack online monitoring and adaptive compensation methods for this degradation process, and usually rely on periodic maintenance or replacement after failure, which is lagging maintenance and costly. To address the aforementioned issues, a smart switching cable method for GIS power frequency and partial discharge testing is proposed. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for intelligent switching cables used in GIS power frequency and partial discharge testing, comprising the following steps: Step S1, Test Task Analysis and Switching Path Planning: Analyze the test task instructions, combine the power frequency withstand voltage test and partial discharge test of the gas chamber under test into a continuous test sequence, and plan the switching path between the terminals. Step S2: Mechanical switching is performed and multi-source signals are acquired synchronously. In response to the switching path, a mechanical switching operation is performed to disconnect the test lead from the current primary terminal and connect it to the target primary terminal. At the same time, the insertion and extraction force timing signal and the contact circuit impedance timing signal during the switching process are acquired synchronously. Step S3, feature vector construction: extract force energy feature components from the insertion and extraction force timing signal and extract impedance transient feature components from the contact circuit impedance timing signal, and concatenate the force energy feature components and impedance transient feature components into a joint feature vector; Step S4: Physical and data-driven hybrid diagnostic reasoning. The joint feature vector is input into the hybrid diagnostic model, which is composed of a physical constraint model and a data-driven reconstruction model in parallel. The physical constraint model outputs the contact resistance estimate based on the terminal electromechanical contact equivalent parameters, and the data-driven reconstruction model outputs the reconstruction error of the joint feature vector based on the deep autoencoder network. The contact resistance estimate and the reconstruction error are used together as a multidimensional health indicator. Step S5, Multidimensional Threshold Compliance Judgment and Loop Control: Multidimensional health indicators are judged for multidimensional threshold compliance. If either the estimated contact resistance or the reconstruction error exceeds the preset safety threshold, the current test loop is locked and a warning message containing the specific fault dimension is generated. If both the estimated contact resistance and the reconstruction error are within the preset safety threshold, the high-voltage test power supply is unlocked, and the test voltage is allowed to be applied.

[0006] Furthermore, the process of acquiring the insertion / extraction force timing signal and the contact circuit impedance timing signal in step S2 is specifically as follows: During the entire stroke of the end effector when it is inserted into the target terminal block, the insertion and extraction force timing signal F(t) is acquired at a preset sampling rate. After the metal contact finger of the end effector makes initial contact with the conductive part of the target primary terminal, and before the locking mechanism is activated, a multi-harmonic excitation current with an amplitude not exceeding the safety extra-low voltage limit is injected into the contact circuit. The response voltage is then synchronously acquired and calculated into a complex impedance vector sequence containing real and imaginary parts, forming the contact circuit impedance timing signal Z(t).

[0007] Furthermore, in step S3, the extraction process of the force energy characteristic components is specifically as follows: Define the effective time period of the switching process Short-time energy analysis was performed on the acquired insertion and extraction force time-series signal F(t). The square of the signal amplitude was integrated over the effective time period and divided by the time period length to obtain the average energy. The calculation process is as follows: ; Simultaneously, by traversing the effective time period using a sliding window w, the integral value of the square of the signal amplitude within each window is calculated as the window energy. Then, calculate the standard deviation of the energy for all windows to obtain the energy variability. The energy variability calculation process involves first calculating the energy of each window. mean Then, the square root of the mean of the squared deviations of the energy from the mean of each window is calculated. The specific calculation process is as follows: ; Where n is the total number of sliding windows; Average energy With energy variability The combination is a force-energy characteristic component.

[0008] Furthermore, in step S3, the extraction process of the transient characteristic components of impedance is specifically as follows: The impedance vector sequence is modulo-operated to obtain the impedance amplitude time-series signal |Z(t)|. The impedance amplitude time-series signal is then subjected to first-order forward differential processing to solve for the impedance rate of change sequence. The calculation process is as follows: the difference in impedance amplitude between two adjacent sampling times is divided by the sampling interval between the two sampling times to obtain the impedance abrupt change characteristic. The calculation process is as follows: ; in, The sampling interval is set according to the preset sampling rate; the average impedance amplitude during the stable contact phase between the end effector and the target primary terminal is extracted as the steady-state impedance characteristic. ; impedance abrupt change characteristic With steady-state impedance characteristic quantity The combination forms the transient characteristic component of impedance.

[0009] Furthermore, in step S4, the construction and operation of the hybrid diagnostic model are specifically as follows: The physical constraint model is the electromechanical contact equivalent model of the target primary terminal. This model uses finite element simulation to obtain the nonlinear mapping relationship between force energy characteristic components and contact resistance under different contact degradation states, generating a priori knowledge base. This knowledge base stores corresponding contact resistance values ​​indexed by force energy characteristic components. Using real-time extracted force energy characteristic components as input, the system retrieves matching force energy characteristic component intervals from the priori knowledge base, performs interpolation calculations, and outputs estimated contact resistance values. ; The data-driven reconstruction model is a pre-trained deep autoencoder network (DEC). The DEC uses a set of joint feature vector samples collected under qualified wiring conditions as training data. The encoder of the DEC compresses the input joint feature vector into a low-dimensional latent space. The decoder reconstructs the joint feature vector from the low-dimensional latent space and calculates the mean square error between the reconstructed vector and the input vector. The calculation process is as follows: square the difference between corresponding elements of the reconstructed vector and the input vector, sum all the squared values, and divide by the vector dimension to obtain the reconstruction error. The specific calculation process is as follows: ; Where m is the dimension of the joint feature vector. Let j be the j-th element of the input joint feature vector. To reconstruct the j-th element of the joint eigenvector; Multidimensional health indicators are estimated from contact resistance values and reconstruction error constitute.

[0010] Furthermore, during the training process of the deep autoencoder network, the training data consists only of joint feature vectors of qualified wiring states collected during the experimental system debugging phase using standard verification terminals and under preset standard operating conditions. This training data enables the deep autoencoder network to learn the low-dimensional manifold distribution of the joint feature vectors of qualified wiring states. When the real-time input joint feature vector represents an abnormal wiring or acquisition failure state, the real-time input joint feature vector deviates from the low-dimensional manifold distribution. The encoder of the deep autoencoder network maps this real-time input joint feature vector to a region deviating from the normal manifold, and the decoder of the deep autoencoder network reconstructs the reconstruction error from this region. Exceeding the normal range.

[0011] Furthermore, the multidimensional threshold compliance judgment in step S5 specifically includes: Set a safe threshold for contact resistance and reconstruction error safety threshold Among them, the contact resistance safety threshold The determination process is as follows: Based on the rated current capacity and allowable temperature rise limit of the target primary terminal, the maximum contact resistance value corresponding to the allowable temperature rise of the terminal under rated operating current is obtained by reverse calculation through the conversion relationship between contact resistance and Joule heat power. This maximum contact resistance value is taken as the contact resistance safety threshold. The conversion process is as follows: ; in, To allow for an upper limit of temperature rise, The rated operating current is k, which is the heat dissipation coefficient related to the structure, material and environmental heat dissipation conditions of the terminal block. k is obtained through experimental calibration or by consulting an engineering manual. When the real-time calculated estimated value of contact resistance Less than the contact resistance safety threshold And reconstruction error Less than the reconstruction error safety threshold When the wiring is deemed acceptable, a power supply permission command is issued. When the estimated contact resistance value Greater than or equal to the contact resistance safety threshold When this occurs, it is determined to be a poor contact fault, and an alarm for excessive contact resistance is generated; When reconstruction error Greater than or equal to the reconstruction error safety threshold If the wiring configuration is abnormal or the data acquisition process fails, a comprehensive early warning system for both wiring abnormalities and sensor failures will be generated.

[0012] Furthermore, after completing the power frequency withstand voltage and partial discharge tests of the current gas chamber, the control parameters of the next switching path are adaptively corrected based on historical data of multidimensional health indicators. The specific process is as follows: Set a threshold for slight degradation of contact resistance And the contact resistance slightly degrades to the threshold. <Contact resistance safety threshold> The slight degradation range is defined as the estimated contact resistance value. satisfy The interval; If the estimated contact resistance value calculated during the current switching process is... If the device falls into the slight degradation range, in the next switching path, the target insertion and extraction force of the end effector will be increased by a preset step size through the proportional control model. The proportional control model estimates the value based on the contact resistance. With the threshold of slight degradation of contact resistance The deviation between the two is calculated to compensate for the incremental insertion and extraction force. The calculation process is as follows: the current estimated contact resistance value is... With the threshold of slight degradation of contact resistance The difference is taken as the control deviation e, and the control deviation e is multiplied by the preset force compensation ratio coefficient. Receive the compensation increment The calculation process is as follows: . The present invention has the following advantages over the prior art: By combining power frequency withstand voltage and partial discharge tests into a continuous test sequence and automatically performing the switching, the insertion / extraction force timing signal and the contact circuit impedance timing signal are simultaneously acquired. Force energy characteristic components and impedance transient characteristic components are extracted to construct a joint feature vector, which is then input into a hybrid diagnostic model composed of a physical constraint model and a data-driven reconstruction model. The physical constraint model outputs an estimated contact resistance value based on electromechanical contact equivalent parameters, while the data-driven reconstruction model uses a deep autoencoder network trained only with the joint feature vector of qualified wiring conditions to output the reconstruction error. Both serve as multi-dimensional health indicators for threshold compliance judgment, enabling simultaneous identification of poor contact faults caused by excessive contact resistance, as well as abnormal wiring configurations or acquisition failures. This achieves multi-dimensional, high-precision diagnostics. Sensitive switching quality diagnosis avoids missed and false judgments in single-dimensional detection. Furthermore, the deep autoencoder network can generate high reconstruction errors for unknown abnormal states by learning only normal wiring characteristics, reducing dependence on fault samples. It further utilizes the Joule heating temperature rise relationship to back-calculate the contact resistance safety threshold, giving the safety judgment a clear physical basis and ensuring the safety of high-voltage testing. At the same time, it sets a slight degradation range based on historical contact resistance estimates, and adaptively adjusts the target insertion and extraction force of subsequent switching through a proportional control model. This can actively compensate for contact pressure when the terminal experiences slight wear and degradation, delaying performance deterioration and maintaining long-term switching reliability. As a result, it improves the automation level, wiring quality assessment accuracy, and test safety of GIS power frequency and partial discharge testing. Attached Figure Description

[0013] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0014] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0015] like Figure 1 As shown, a method for intelligent switching cable for GIS power frequency and partial discharge testing includes the following steps: Step S1, Test Task Analysis and Switching Path Planning: Analyze the test task instructions, combine the power frequency withstand voltage test and partial discharge test of the gas chamber under test into a continuous test sequence, and plan the switching path between the terminals. Step S2: Mechanical switching is performed and multi-source signals are acquired synchronously. In response to the switching path, a mechanical switching operation is performed to disconnect the test lead from the current primary terminal and connect it to the target primary terminal. At the same time, the insertion and extraction force timing signal and the contact circuit impedance timing signal during the switching process are acquired synchronously. Step S3, feature vector construction: extract force energy feature components from the insertion and extraction force timing signal and extract impedance transient feature components from the contact circuit impedance timing signal, and concatenate the force energy feature components and impedance transient feature components into a joint feature vector; Step S4: Physical and data-driven hybrid diagnostic reasoning. The joint feature vector is input into the hybrid diagnostic model, which is composed of a physical constraint model and a data-driven reconstruction model in parallel. The physical constraint model outputs the contact resistance estimate based on the terminal electromechanical contact equivalent parameters, and the data-driven reconstruction model outputs the reconstruction error of the joint feature vector based on the deep autoencoder network. The contact resistance estimate and the reconstruction error are used together as a multidimensional health indicator. Step S5, Multidimensional Threshold Compliance Judgment and Loop Control: Multidimensional health indicators are judged for multidimensional threshold compliance. If either the estimated contact resistance or the reconstruction error exceeds the preset safety threshold, the current test loop is locked and a warning message containing the specific fault dimension is generated. If both the estimated contact resistance and the reconstruction error are within the preset safety threshold, the high-voltage test power supply is unlocked, and the test voltage is allowed to be applied.

[0016] The specific process of acquiring the insertion / extraction force timing signal and the contact circuit impedance timing signal in step S2 is as follows: During the entire stroke of the end effector when it is inserted into the target terminal block, the insertion and extraction force timing signal F(t) is acquired at a preset sampling rate. After the metal contact finger of the end effector makes initial contact with the conductive part of the target primary terminal, and before the locking mechanism is activated, a multi-harmonic excitation current with an amplitude not exceeding the safety extra-low voltage limit is injected into the contact circuit. The response voltage is synchronously acquired and calculated into a complex impedance vector sequence containing real and imaginary parts, forming the contact circuit impedance timing signal Z(t). By continuously acquiring force signals throughout the entire insertion stroke, the mechanical evolution of the contact finger from contact, sliding in to limit position can be fully recorded, avoiding omission of transient force characteristics reflecting terminal deformation or foreign object jamming. After the metal contact finger makes initial contact with the conductive part and before the locking mechanism operates, a multi-harmonic excitation current is injected and the response voltage is acquired to calculate the complex impedance. This not only eliminates the interference of locking pressure on the contact interface state, but also utilizes multiple harmonics to excite multiple frequencies simultaneously, obtaining a wideband impedance spectrum within a single short time window. This allows for rapid capture of interface characteristics such as contact resistance and surface film capacitive reactance. Furthermore, the injected amplitude does not exceed the safety extra-low voltage limit, ensuring operational safety.

[0017] For example, after a 220kV GIS bay power frequency withstand voltage test, the test lead needs to be switched from the withstand voltage terminal to the partial discharge test terminal. During the process of the end effector being inserted into the target partial discharge terminal, the strain gauge force sensor installed between the end effector and the drive joint collects the insertion and extraction force timing signal F(t) at a sampling rate of 200Hz. It records that the contact finger gradually rises from 0N at the initial contact to a peak value of 15.3N when it passes the anti-retraction structure, and then falls back to a stable holding force of 10.1N, which fully reflects the mechanical state of the insertion process. The impedance detection circuit is triggered at the instant the contact finger makes initial contact with the conductive component, injecting a multi-harmonic current with an amplitude of 10mA into the contact circuit. A. Simultaneously acquire the response voltage v(t) at a sampling rate of 10 kS / s, extract a short time window of data of 20 ms, and calculate the complex voltage at 1 kHz and 10 kHz frequencies using Fast Fourier Transform. For example, obtain the 1 kHz complex voltage at the middle of the window. Given that the phasor of the injected current at this frequency is Then complex impedance Amplitude phase ; Similarly, calculate the complex impedance at 10kHz. Amplitude phase This yields the complex impedance vector sequence Z(t) containing dual-frequency information at the sampling moment, providing wideband, high-resolution contact interface data for subsequent extraction of impedance abrupt change characteristics and steady-state characteristics, thereby effectively improving the sensitivity and accuracy of wiring condition diagnosis.

[0018] In step S3, the extraction process of force energy characteristic components is as follows: Define the effective time period of the switching process Short-time energy analysis was performed on the acquired insertion and extraction force time-series signal F(t). The square of the signal amplitude was integrated over the effective time period and divided by the time period length to obtain the average energy. The calculation process is as follows: ; Simultaneously, by traversing the effective time period using a sliding window w, the integral value of the square of the signal amplitude within each window is calculated as the window energy. Then, calculate the standard deviation of the energy for all windows to obtain the energy variability. The energy variability calculation process involves first calculating the energy of each window. mean Then, the square root of the mean of the squared deviations of the energy from the mean of each window is calculated. The specific calculation process is as follows: ; Where n is the total number of sliding windows; Average energy With energy variability The combination is a force-energy characteristic component; By defining the effective time period of the switching process and calculating the average energy and energy variability, the overall level and dynamic stability of the insertion and extraction force can be quantified simultaneously. The average energy reflects whether the terminal contact pressure is sufficient, while the energy variability captures the force fluctuations caused by terminal deformation, foreign object jamming, or poor operation of the drive mechanism, thereby avoiding omissions and misjudgments when judging wiring quality solely based on peak force or single-point force values.

[0019] In a scenario where a 220kV GIS bay is switched to a partial discharge test terminal after power frequency withstand voltage testing, the force sensor collects the insertion / extraction force timing signal F(t) at a sampling rate of 200Hz, defining the effective time period. The moment when the metal contact finger just touches the conductive component =0s until it is inserted and remains stable for 0.5s. =2s.

[0020] Short-time energy calculations were performed on F(t) from 401 sampling points within this time period. First, the integral of the square of the signal amplitude was calculated, and then trapezoidal numerical integration was used to obtain the result. The average energy is: ; Subsequently, a sliding window of w=0.2s (40 sampling points) with a step size of 0.1s was used to traverse the effective time period, obtaining a total of 19 window energies. Each For the window The integral values ​​are as follows: (Near the stress peak),... (Stable phase).

[0021] Calculate the mean energy of all windows. Then, the energy variability is calculated. ; Average energy With energy variability The combined force energy characteristic components (160.25, 7.9) indicate that the overall contact pressure level is normal (compared to historical acceptable data). about This also reflects that there are slight fluctuations in the insertion process, but the degree of variation does not exceed the limit, providing a clear mechanical state basis for the subsequent hybrid diagnostic model.

[0022] In step S3, the extraction process of the transient characteristic components of impedance is as follows: The impedance vector sequence is modulo-operated to obtain the impedance amplitude time-series signal |Z(t)|. The impedance amplitude time-series signal is then subjected to first-order forward differential processing to solve for the impedance rate of change sequence. The calculation process is as follows: the difference in impedance amplitude between two adjacent sampling times is divided by the sampling interval between the two sampling times to obtain the impedance abrupt change characteristic. The calculation process is as follows: ; in, The sampling interval is set according to the preset sampling rate; the average impedance amplitude during the stable contact phase between the end effector and the target primary terminal is extracted as the steady-state impedance characteristic. ; impedance abrupt change characteristic With steady-state impedance characteristic quantity The combination forms the transient characteristic components of impedance; By taking the modulus of the complex impedance vector sequence to obtain the amplitude timing signal, and performing first-order forward differential extraction to extract impedance abrupt change characteristics, it is possible to accurately capture the moment and rate of impedance abrupt change caused by sudden changes in contact area, surface oxide film rupture, or impurity compression during the process of the finger sliding into the conductive part. This compensates for the defect of steady-state impedance value's lag in responding to short-term contact failure events. At the same time, the average impedance amplitude during the contact stabilization stage is extracted as a steady-state impedance characteristic, providing a reference benchmark for evaluating long-term contact quality. The impedance transient characteristic component formed by the combination of the two can detect instantaneous contact faults and reflect the overall contact status, effectively improving the sensitivity and reliability of wiring diagnosis.

[0023] In the scenario of switching from the 220kV GIS bay power frequency withstand voltage to the partial discharge test terminal, the impedance spectrum detection circuit obtains the complex impedance vector sequence Z(t) at a sampling rate of 10kS / s, and collects a total of 200 sampling points within a 20ms window after the initial contact between the contact finger and the conductive part.

[0024] Two adjacent sampling points, t=5ms and t=5.1ms, were selected, and their complex impedances were calculated using Fast Fourier Transform. At a frequency of 1kHz, Amplitude ; Amplitude | .

[0025] Sampling interval The rate of change at that moment is obtained using the formula for calculating the characteristic quantity of impedance abrupt change: ; Continuing to calculate the impedance abrupt change characteristic at subsequent time points, a dynamic process of the reactive contact can be formed. A sequence in which at a certain moment occurs A sharp drop in the above values ​​indicates a momentary contact failure.

[0026] During the contact stabilization phase from t=15ms to t=20ms, the average impedance amplitude of 50 sampling points within this period is taken as the steady-state impedance characteristic quantity. It is assumed that the amplitudes at each point are as follows: The calculation yields: ; impedance abrupt change characteristic sequence With steady-state impedance characteristic quantity The combination of impedance transient characteristic components provides dual information on the dynamic process and steady-state characteristics of the contact interface for the hybrid diagnostic model, thereby accurately determining whether there is a risk of instantaneous disconnection or long-term contact degradation in the wiring condition.

[0027] In step S4, the construction and operation of the hybrid diagnostic model are specifically as follows: The physical constraint model is the electromechanical contact equivalent model of the target primary terminal. This model uses finite element simulation to obtain the nonlinear mapping relationship between force energy characteristic components and contact resistance under different contact degradation states, generating a priori knowledge base. This knowledge base stores corresponding contact resistance values ​​indexed by force energy characteristic components. Using real-time extracted force energy characteristic components as input, the system retrieves matching force energy characteristic component intervals from the priori knowledge base, performs interpolation calculations, and outputs estimated contact resistance values. ; The data-driven reconstruction model is a pre-trained deep autoencoder network (DEC). The DEC uses a set of joint feature vector samples collected under qualified wiring conditions as training data. The encoder of the DEC compresses the input joint feature vector into a low-dimensional latent space. The decoder reconstructs the joint feature vector from the low-dimensional latent space and calculates the mean square error between the reconstructed vector and the input vector. The calculation process is as follows: square the difference between corresponding elements of the reconstructed vector and the input vector, sum all the squared values, and divide by the vector dimension to obtain the reconstruction error. The specific calculation process is as follows: ; Where m is the dimension of the joint feature vector. Let j be the j-th element of the input joint feature vector. To reconstruct the j-th element of the joint eigenvector; Multidimensional health indicators are estimated from contact resistance values and reconstruction error constitute; The physical constraint model, based on the equivalent parameters of terminal electromechanical contact, establishes a nonlinear mapping between force energy characteristic components and contact resistance through finite element simulation. It can directly calculate the contact resistance value with clear physical meaning from the insertion and extraction force signal, providing a quantifiable and interpretable evaluation basis for wiring qualification judgment. The data-driven reconstruction model adopts a deep autoencoder network, which is trained only using the joint feature vector of qualified wiring state. This allows the network to learn the low-dimensional manifold distribution of normal state. It can automatically identify abnormal patterns that deviate from the normal state through reconstruction error without acquiring various fault samples, solving the problem of scarce fault data and unpredictable abnormal types in GIS switching scenarios. Both models output the contact resistance estimate and reconstruction error in parallel. The multi-dimensional health indicators simultaneously characterize the wiring state from both physical mechanism and statistical pattern dimensions. The complementary advantages effectively avoid misjudgment and omission caused by the lack of dynamic modeling or data noise in a single model, significantly improving the accuracy and robustness of diagnosis.

[0028] Continuing the scenario described above where the 220kV GIS bay power frequency withstand voltage is switched to the partial discharge test terminal, the force energy characteristic component extracted in step S3 is: The transient characteristic component of impedance is The four-dimensional joint feature vector is formed by splicing them together. .

[0029] In the physical constraint model, a model is pre-generated using finite element simulation. and Given an indexed contact resistance mapping table, for the input (160.25, 7.9), retrieve the contact resistance corresponding to the nearest neighbor grid point (160, 8). Corresponding to (161,8) The estimated contact resistance value was obtained by linear interpolation. .

[0030] The data-driven reconstruction model employs a pre-trained deep autoencoder network, where the encoder will... The encoded vector is obtained by mapping to a two-dimensional latent space, and the decoder reconstructs the output. The reconstruction error is calculated using the mean square error formula as follows: ; This yields multidimensional health indicators. .

[0031] If a preset contact resistance safety threshold is used Reconstruct the error safety threshold =5, at this time This indicates that the static contact resistance is acceptable, but This indicates that the overall pattern of the joint feature vector input in real time deviates significantly from the normal distribution when the wiring is qualified. Further investigation revealed that... The component anomaly (-700Ω / s) deviates significantly from the reconstruction value of the component by the deep autoencoder network (approximately -295Ω / s). This can be identified as a momentary bounce or foreign object jamming during the finger insertion process, resulting in poor dynamic contact and triggering an early warning, thus avoiding transient hazards that cannot be detected by relying solely on the physical model.

[0032] During the training of the deep autoencoder network, the training data consists only of joint feature vectors of qualified wiring states acquired during the experimental system debugging phase, using standard verification terminals and under preset standard operating conditions. This training data enables the deep autoencoder network to learn the low-dimensional manifold distribution of the joint feature vectors of qualified wiring states. When the real-time input joint feature vector represents an abnormal wiring or acquisition failure state, the real-time input joint feature vector deviates from the low-dimensional manifold distribution. The encoder of the deep autoencoder network maps this real-time input joint feature vector to a region deviating from the normal manifold, and the decoder of the deep autoencoder network reconstructs the reconstruction error from this region. Exceeding the normal range; The training data consists only of the joint feature vectors of qualified wiring states collected using standard calibration terminals under preset standard operating conditions during the debugging phase of the experimental system. Model training can be completed without acquiring various fault samples, solving the problems of diverse and exhaustive fault modes, scarce fault data, and high acquisition costs in GIS switching scenarios. The network learns the low-dimensional manifold distribution of the joint feature vectors under qualified wiring states, and automatically maps the real-time input vectors when abnormal wiring or failure of the acquisition link occurs to the region deviating from the normal manifold, thereby generating reconstruction errors beyond the normal range. Fault detection is achieved by utilizing the difference in statistical distribution between normal and abnormal conditions, enabling the model to have the ability to generalize and identify unknown fault modes. This avoids the dependence of supervised classification methods on fault labels and the risk of missed judgments due to incomplete coverage of fault types. At the same time, the training data is acquired under standard calibration terminal conditions, ensuring the consistency and reproducibility of the normal state benchmark.

[0033] In a scenario involving the replacement of the partial discharge test terminal after the power frequency withstand voltage test of a 220kV GIS bay, during the test system commissioning phase, brand-new standard calibration terminals were used, and the ambient temperature was controlled at 25℃ and humidity at 50%RH. The replacement actuator performed 50 qualified wiring operations at a preset standard insertion and extraction speed. During each replacement process, insertion / extraction force timing signals and contact circuit impedance timing signals were collected, and the force energy characteristic components were extracted. quantity The 50 four-dimensional joint feature vectors are concatenated to form a training dataset.

[0034] Taking three typical training samples as examples: Sample 1 is ; Sample 2 is ; Sample 3 is .

[0035] Deep autoencoder networks use an encoder to compress a four-dimensional input into a two-dimensional latent space, and a decoder to reconstruct the four-dimensional output. After training, the average reconstruction error on 50 training samples is reduced. Converging to within 0.15, for example, for a sample The network outputs a reconstructed vector. The reconstruction error is: ; This error value is extremely small, indicating that the network has learned the low-dimensional manifold distribution of the joint feature vector of qualified wiring states. In actual experiments, if a slight deformation of the contact finger causes an abnormally large increase in the force energy variability during a connection change, the extracted real-time joint feature vector will be... Input the trained deep autoencoder network and reconstruct the output. The reconstruction error is calculated as follows: ; If the preset reconstruction error safety threshold =5, at this time This indicates that the joint feature vector deviates significantly from the normal distribution of qualified wiring conditions, triggering an abnormal wiring pattern warning, thus achieving accurate identification of unknown abnormal conditions without the need to collect finger deformation fault samples in advance.

[0036] Step S5, the multidimensional threshold compliance judgment, specifically includes: Set a safe threshold for contact resistance and reconstruction error safety threshold Among them, the contact resistance safety threshold The determination process is as follows: Based on the rated current capacity and allowable temperature rise limit of the target primary terminal, the maximum contact resistance value corresponding to the allowable temperature rise of the terminal under rated operating current is obtained by reverse calculation through the conversion relationship between contact resistance and Joule heat power. This maximum contact resistance value is taken as the contact resistance safety threshold. The conversion process is as follows: ; in, To allow for an upper limit of temperature rise, The rated operating current is k, which is the heat dissipation coefficient related to the structure, material and environmental heat dissipation conditions of the terminal block. k is obtained through experimental calibration or by consulting an engineering manual. When the real-time calculated estimated value of contact resistance Less than the contact resistance safety threshold And reconstruction error Less than the reconstruction error safety threshold When the wiring is deemed acceptable, a power supply permission command is issued. When the estimated contact resistance value Greater than or equal to the contact resistance safety threshold When this occurs, it is determined to be a poor contact fault, and an alarm for excessive contact resistance is generated; When reconstruction error Greater than or equal to the reconstruction error safety threshold When the wiring configuration is abnormal or the data acquisition process fails, a comprehensive early warning for both wiring abnormality and sensor failure is generated. The contact resistance safety threshold is obtained by inversely solving the physical relationship between Joule heat power and temperature rise based on the rated current capacity and allowable temperature rise limit of the target primary terminal. This provides a clear physical basis for safety judgment rather than relying on experience, ensuring that the contact resistance will not cause over-temperature risk under test current. At the same time, a reconstruction error safety threshold is set, forming a two-dimensional criterion with the contact resistance threshold. This can identify static contact failures caused by excessive contact resistance and dynamic anomalies caused by abnormal wiring or failure of the acquisition link. This enables accurate differentiation of fault types, avoids misjudgment or omission due to a single indicator, and provides clear guidance for subsequent targeted maintenance measures, thereby significantly improving the safety of high-voltage testing and the intelligent diagnostic capability of the switching system.

[0037] In the scenario of switching the connection from the 220kV GIS bay power frequency withstand voltage test terminal to the partial discharge test terminal, the rated operating current of the target primary terminal is... Allowable temperature rise limit The heat dissipation coefficient k was experimentally calibrated to be The contact resistance safety threshold is determined by the following formula: The calculation yielded: ; Simultaneously set a reconstruction error safety threshold. .

[0038] The multidimensional health indicators for a certain handover have been calculated in the previous example. , ,at this time This indicates that the static contact resistance meets the thermal stability requirements, but The system was determined to be either due to abnormal wiring or failure in the data acquisition process, generating a comprehensive early warning for both wiring abnormalities and sensor failures. Actual inspection revealed that the touch finger rebound caused a momentary disconnection.

[0039] Another example is the result calculated after a certain switchover. ,at this time and The fault was identified as poor contact, and an alarm for excessive contact resistance was generated. Upon inspection, it was confirmed that there was an oxide film on the terminal surface that had not been removed.

[0040] For another example, after a normal switching connection, , it is satisfied that and , the wiring is determined to be qualified, the central control unit issues a voltage application permission instruction, and the test voltage is successfully applied subsequently. Through the above double-threshold conformity judgment, different types of abnormalities are effectively distinguished, and reliable unlocking of normal wiring is ensured.

[0041] After completing the power frequency withstand voltage and partial discharge tests of the current gas chamber, the control parameters of the next switching path are adaptively corrected based on the historical data of multi-dimensional health indicators, and the specific process is as follows: A slight degradation threshold of contact resistance is set , and the slight degradation threshold of contact resistance is less than the safety threshold of contact resistance , and the slight degradation interval is defined as the interval where the estimated value of contact resistance satisfies ; If the estimated value of contact resistance solved in the current switching process falls into the slight degradation interval, in the next switching path, the target insertion and extraction force of the end effector is increased by a preset step size through a proportional control model; The proportional control model calculates the compensation increment of the target insertion and extraction force according to the deviation between the estimated value of contact resistance and the slight degradation threshold of contact resistance , and the calculation process is as follows: the difference between the current estimated value of contact resistance and the slight degradation threshold of contact resistance is taken as the control deviation e, and the control deviation e is multiplied by a preset force compensation proportional coefficient to obtain the compensation increment , the calculation process of which is: ; By setting a slight degradation threshold lower than the safety threshold, the performance degradation trend of the terminal can be actively identified when the contact resistance increases slightly due to repeated insertion and extraction but no safety risk is yet constituted; the proportional control model is used to calculate the compensation increment of the target insertion and extraction force according to the current contact resistance deviation, so as to automatically increase the insertion force of the end effector, and increase the contact pressure to compensate for the reduction of contact area or the relaxation of elastic elements caused by terminal wear, thereby delaying the further deterioration of the contact resistance, preventing it from eventually developing into an over-limit fault, avoiding the blindness of scheduled maintenance and the cost of emergency shutdown maintenance, realizing adaptive preventive maintenance based on the actual degradation state, and significantly improving the switching reliability and system availability in long-term operation.

[0042] ​In the scenario of switching the connection from the power frequency withstand voltage test terminal to the partial discharge test terminal of a 220kV GIS bay, after completing the power frequency withstand voltage and partial discharge tests of the gas chamber, the system records the estimated contact resistance value calculated during this switching. .

[0043] Known contact resistance safety threshold Preset contact resistance slight degradation threshold ,satisfy The current terminal is determined to be in a slightly degraded range. Based on this, adaptive correction is initiated when planning the switching path for the next test chamber. The force compensation proportional coefficient in the proportional control model... Values Control deviation The compensation increment for the target insertion and extraction force is calculated as follows: ; In this replacement, the original target insertion and extraction force of the end effector was a stable holding force of 10.1 N. Therefore, the corrected target insertion and extraction force is adjusted to... The next switching operation will be performed by inserting the device with the adjusted target insertion and extraction force to compensate for the contact pressure attenuation caused by terminal wear, thereby maintaining the contact resistance within a safe range.

[0044] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for intelligent switching cables used in GIS power frequency and partial discharge testing, characterized in that, Includes the following steps: Step S1, Test Task Analysis and Switching Path Planning: Analyze the test task instructions, combine the power frequency withstand voltage test and partial discharge test of the gas chamber under test into a continuous test sequence, and plan the switching path between the terminals. Step S2: Mechanical switching is performed and multi-source signals are acquired synchronously. In response to the switching path, a mechanical switching operation is performed to disconnect the test lead from the current primary terminal and connect it to the target primary terminal. At the same time, the insertion and extraction force timing signal and the contact circuit impedance timing signal during the switching process are acquired synchronously. Step S3, feature vector construction: extract force energy feature components from the insertion and extraction force timing signal and extract impedance transient feature components from the contact circuit impedance timing signal, and concatenate the force energy feature components and impedance transient feature components into a joint feature vector; Step S4: Physical and data-driven hybrid diagnostic reasoning. The joint feature vector is input into the hybrid diagnostic model, which is composed of a physical constraint model and a data-driven reconstruction model in parallel. The physical constraint model outputs the contact resistance estimate based on the terminal electromechanical contact equivalent parameters, and the data-driven reconstruction model outputs the reconstruction error of the joint feature vector based on the deep autoencoder network. The contact resistance estimate and the reconstruction error are used together as a multidimensional health indicator. Step S5, Multidimensional threshold compliance judgment and loop control: Multidimensional health indicators are judged for multidimensional threshold compliance. If either the estimated contact resistance or the reconstruction error exceeds the preset safety threshold, the current test loop is locked and a warning message containing specific fault dimensions is generated. If both the estimated contact resistance and the reconstruction error are within the preset safety threshold, the high-voltage test power supply will be unlocked, allowing the test voltage to be applied.

2. The intelligent switching cable method for GIS power frequency and partial discharge testing according to claim 1, characterized in that: The specific process of acquiring the insertion / extraction force timing signal and the contact circuit impedance timing signal in step S2 is as follows: During the entire stroke of the end effector when it is inserted into the target terminal block, the insertion and extraction force timing signal F(t) is acquired at a preset sampling rate. After the metal contact finger of the end effector makes initial contact with the conductive part of the target primary terminal, and before the locking mechanism is activated, a multi-harmonic excitation current with an amplitude not exceeding the safety extra-low voltage limit is injected into the contact circuit. The response voltage is then synchronously acquired and calculated into a complex impedance vector sequence containing real and imaginary parts, forming the contact circuit impedance timing signal Z(t).

3. The intelligent switching cable method for GIS power frequency and partial discharge testing according to claim 2, characterized in that: In step S3, the extraction process of force energy characteristic components is as follows: Define the effective time period of the switching process Short-time energy analysis was performed on the acquired insertion and extraction force time-series signal F(t). The square of the signal amplitude was integrated over the effective time period and divided by the time period length to obtain the average energy. ; Simultaneously, by traversing the effective time period using a sliding window w, the integral value of the square of the signal amplitude within each window is calculated as the window energy. Then, calculate the standard deviation of the energy for all windows to obtain the energy variability. ; Average energy With energy variability The combination is a force-energy characteristic component.

4. The intelligent switching cable method for GIS power frequency and partial discharge testing according to claim 3, characterized in that: In step S3, the extraction process of the transient characteristic components of impedance is as follows: The impedance vector sequence is modulo-operated to obtain the impedance amplitude time-series signal |Z(t)|. The impedance amplitude time-series signal is then subjected to first-order forward differential processing to solve for the impedance rate of change sequence. The calculation process is as follows: the difference in impedance amplitude between two adjacent sampling times is divided by the sampling interval between the two sampling times to obtain the impedance abrupt change characteristic. ; impedance abrupt change characteristic With steady-state impedance characteristic quantity The combination forms the transient characteristic component of impedance.

5. The intelligent switching cable method for GIS power frequency and partial discharge testing according to claim 4, characterized in that: In step S4, the construction and operation of the hybrid diagnostic model are specifically as follows: The physical constraint model is the electromechanical contact equivalent model of the target primary terminal. This model uses finite element simulation to obtain the nonlinear mapping relationship between force energy characteristic components and contact resistance under different contact degradation states, generating a priori knowledge base. This knowledge base stores corresponding contact resistance values ​​indexed by force energy characteristic components. Using real-time extracted force energy characteristic components as input, the system retrieves matching force energy characteristic component intervals from the priori knowledge base, performs interpolation calculations, and outputs estimated contact resistance values. ; The data-driven reconstruction model is a pre-trained deep autoencoder network (DEC). The DEC uses a set of joint feature vector samples collected under qualified wiring conditions as training data. The DEC encoder compresses the input joint feature vectors into a low-dimensional latent space, and the DEC decoder reconstructs the joint feature vectors from this low-dimensional latent space. The mean squared error between the reconstructed vector and the input vector is calculated by squaring the differences between corresponding elements of the reconstructed and input vectors, summing all the squared values, and then dividing by the vector dimension to obtain the reconstruction error. ; Multidimensional health indicators are estimated from contact resistance values and reconstruction error constitute.

6. The intelligent switching cable method for GIS power frequency and partial discharge testing according to claim 5, characterized in that: During the training of a deep autoencoder network, the training data consists only of joint feature vectors of qualified wiring states acquired during the experimental system debugging phase, using standard calibration terminals and under preset standard operating conditions. This training data enables the deep autoencoder network to learn the low-dimensional manifold distribution of the joint feature vectors of qualified wiring states. When the real-time input joint feature vectors represent abnormal wiring or acquisition failure states, the real-time input joint feature vectors deviate from the low-dimensional manifold distribution. The encoder of the deep autoencoder network maps the real-time input joint feature vectors to a region deviating from the normal manifold, and the decoder reconstructs the reconstruction error from this deviated region. Exceeding the normal range.

7. The intelligent switching cable method for GIS power frequency and partial discharge testing according to claim 6, characterized in that: Step S5, the multidimensional threshold compliance judgment, specifically includes: Set a safe threshold for contact resistance and reconstruction error safety threshold Among them, the contact resistance safety threshold The determination process is as follows: Based on the rated current capacity and allowable temperature rise limit of the target primary terminal block, the maximum contact resistance value corresponding to the terminal block not exceeding the allowable temperature rise under rated operating current is obtained by reverse calculation through the conversion relationship between contact resistance and Joule heat power. The maximum contact resistance value is used as the contact resistance safety threshold. ; When the real-time calculated estimated value of contact resistance Less than the contact resistance safety threshold And reconstruction error Less than the reconstruction error safety threshold When the wiring is deemed acceptable, a power supply permission command is issued. When the estimated contact resistance value Greater than or equal to the contact resistance safety threshold When this occurs, it is determined to be a poor contact fault, and an alarm for excessive contact resistance is generated; When reconstruction error Greater than or equal to the reconstruction error safety threshold If the wiring configuration is abnormal or the data acquisition process fails, a comprehensive early warning system for both wiring abnormalities and sensor failures will be generated.

8. The intelligent switching cable method for GIS power frequency and partial discharge testing according to claim 7, characterized in that: After completing the power frequency withstand voltage and partial discharge tests of the current gas chamber, the control parameters of the next switching path are adaptively corrected based on historical data of multidimensional health indicators. The specific process is as follows: Set a threshold for slight degradation of contact resistance And the contact resistance slightly degrades to the threshold. <Contact resistance safety threshold> The slight degradation range is defined as the estimated contact resistance value. satisfy The interval; If the estimated contact resistance value calculated during the current switching process is... If the device falls into the slight degradation range, in the next switching path, the target insertion and extraction force of the end effector will be increased by a preset step size through the proportional control model. The proportional control model estimates the value based on the contact resistance. With the threshold of slight degradation of contact resistance The deviation between the two is calculated to compensate for the incremental insertion and extraction force. The calculation process is as follows: the current estimated contact resistance value is... With the threshold of slight degradation of contact resistance The difference is taken as the control deviation e, and the control deviation e is multiplied by the preset force compensation ratio coefficient. Receive the compensation increment .

Images