A method and system for evaluating vacuum condition of a vacuum interrupter

Abstract

The present invention discloses a method and system for evaluating the vacuum condition of a vacuum interrupter, belonging to the technical field of vacuum circuit breakers. The evaluation method combines RF with XGBoost algorithms to achieve heterogeneous integration, while adding a GAN discriminator to enhance physical constraints and construct a universal relationship model between X-ray intensity and vacuum degree. The feature importance of RF can guide the feature selection or weighting of XGBoost, enhancing feature interaction. The adversarial mechanism of GAN can coordinate data fitting with physical constraints, using adversarial training to enhance the physical rationality and data authenticity of model predictions. The discriminator is used to double-verify the generated results, ensuring that the model conforms to both the experimental data distribution and the constraints of theoretical physical laws, significantly improving the credibility of the results. The evaluation method of the present invention can achieve accurate fitting of the stable attenuation trend in the high vacuum stage and robust prediction of fluctuation data in the critical vacuum stage, proposing new ideas for vacuum condition assessment.

Description

Technical Field

[0001] The present invention belongs to the technical field of vacuum circuit breakers and relates to a method and system for evaluating the vacuum condition of a vacuum interrupter. Background Art

[0002] Vacuum circuit breakers are highly favored by power system users for their excellent dielectric strength, arc-extinguishing capability, and pollution-free properties, and they hold a dominant position in the medium-voltage field. As a core component, the vacuum interrupter plays a crucial role, and research on it has primarily focused on optimizing the electric field to increase breakdown voltage and improving the magnetic field to enhance high-current interrupting capability. The vacuum level is also one of the primary factors determining the interrupting capacity of the vacuum interrupter; a decrease in the vacuum level will also weaken the interrupting capacity. Long-term use of a vacuum circuit breaker will reduce the vacuum level within the interrupter, reducing the circuit breaker's arc-extinguishing capability and insulation performance, impacting its protective function and shortening the life of the equipment. Therefore, it is of great significance to have a convenient and accurate method for evaluating its vacuum level.

[0003] The traditional commonly used power frequency withstand voltage method and magnetron discharge method, although these two methods are simple in principle and easy to operate, the power frequency withstand voltage method can only detect arc chambers with serious air leakage, and is powerless for vacuum arc chambers in a critical state; the magnetron discharge method has disadvantages such as poor measurement repeatability. Although the field emission current method, post-arc emission current method, high-frequency pre-breakdown current method, arc voltage method, etc. have been proposed, they have not been studied in depth due to experimental conditions. Taking advantage of the fact that the magnitude of the field emission current is proportional to the intensity of the X-rays generated, some scholars proposed the X-ray quantity tracking method in the early 21st century. This method integrates the X-ray release intensity measured by the X-ray dose rate meter and the vacuum degree inside the vacuum arc chamber, and obtains the rule that the lower the vacuum degree, the smaller the X-ray release amount. However, it is only applicable to discrete data, and the data sample is relatively simple. It has not formed a universal system for judging the vacuum condition, and the accuracy is low. Although the X-ray intensity method does not require a magnetic field, it only fits through physical formulas (such as exponential decay). In the critical range of vacuum (10 -2 ~10 -1 The data of Pa (Pa) fluctuate violently due to the electron avalanche effect, which is difficult to be accurately described by traditional mathematical models.

[0004] The development of machine learning technology has provided new methods and insights for exploring the underlying patterns of complex nonlinear relationships. The application of machine learning can deeply explore the relationship between X-ray intensity and vacuum level, improving the accuracy and reliability of vacuum level detection compared to traditional methods. Bagging, a method based on bootstrapping, is typified by Random Forest (RF). It further introduces random feature selection based on bagging to reduce model correlation. RF can be used to predict curves, reducing variance by constructing multiple decision trees in parallel. It is suitable for high-dimensional data and noisy scenarios, but may not adequately fit complex nonlinear relationships. Boosting is an iterative ensemble learning method that trains multiple weak learners sequentially, adjusting sample weights or optimizing the objective function each time to gradually correct the errors of the previous models and ultimately combine them into a strong learner. Extreme Gradient Boosting (XGBoost) is a typical example. XGBoost uses gradient boosting to serially optimize residuals. It excels at capturing subtle local patterns but is more susceptible to overfitting. Summary of the Invention

[0005] In response to the problems existing in the prior art, the present invention provides a method and system for evaluating the vacuum condition of a vacuum interrupter, thereby solving the technical problems in the prior art of insufficient RF fitting of complex nonlinear relationships and overfitting of XGBoost in the critical vacuum range.

[0006] The present invention is achieved through the following technical solutions:

[0007] A method for evaluating the vacuum condition of a vacuum interrupter comprises the following steps:

[0008] S1: Obtain the X-ray intensity when there is field emission current inside the vacuum interrupter;

[0009] S2: pre-processing the acquired X-ray intensity;

[0010] S3: Obtain the vacuum degree in the vacuum interrupter using the pre-processed X-ray intensity and the hybrid segmented model;

[0011] The construction process of the hybrid segmented model is specifically as follows:

[0012] Establish an X-ray-vacuum relationship including the electron mean free path and field emission current, perform logarithmic transformation on the vacuum degree and X-ray intensity, and extract the physical characteristic matrix ;

[0013] The physical feature matrix is ​​analyzed by random forest algorithm. Perform pre-training and transform the physical feature matrix The data in the image are divided into a stable region and a critical region, and asymmetric noise that is adaptively enhanced with vacuum degradation is injected into the critical region.

[0014] XGBoost is used to perform segmented training on the stable region and the critical region. During the training process, a generative adversarial network discriminator is introduced to perform physical constraint enhancement to complete the construction of the hybrid segmented model.

[0015] Preferably, the establishing of an X-ray-vacuum degree relationship including electron mean free path and field emission current comprises:

[0016] Based on the bremsstrahlung theory, an X-ray-vacuum relationship including the electron mean free path and the field emission current is established. The X-ray-vacuum relationship is specifically:

[0017]

[0018]

[0019]

[0020] in, is the X-ray intensity, is the material constant of the X-ray tube, is the external applied voltage, is the field emission current, is the electron mean free path, is the gas absorption coefficient, is the actual gas pressure in the system, is the reference vacuum degree, is the cathode material constant, is the electric field strength, is the Boltzmann constant, is the gas temperature, is the diameter of a gas molecule.

[0021] Preferably, the physical characteristic matrix Including electron energy characteristics and collision frequency characteristics:

[0022] The electron energy characteristics are:

[0023]

[0024] in, is the electron energy, represents the basic charge of the elementary charge, is the external applied voltage;

[0025] The collision frequency characteristics are:

[0026]

[0027] in, is the electron collision frequency, is the contact gap, is the electron mean free path.

[0028] Preferably, the asymmetric noise injected into the critical region that is adaptively enhanced with vacuum degradation has a noise model as follows:

[0029]

[0030] in, is the X-ray intensity measurement value after adding noise, is the logarithmically transformed X-ray intensity measurement value, The mean is 0 and the standard deviation is Asymmetric Gaussian noise, is the skewness coefficient, is the system vacuum degree, express for function.

[0031] Preferably, the physical feature matrix is ​​subjected to random forest algorithm. Perform pre-training and transform the physical feature matrix The data in is divided into stable area and critical area, including:

[0032] Through the Gini coefficient and feature importance, the physical feature matrix The data in is divided into stable area and critical area;

[0033] The objective function of the stable region is:

[0034]

[0035] The robust loss function used in the critical region is:

[0036]

[0037] in, is the objective function in the stable region, For the i The true X-ray intensity of the sample, For the model i The predicted value of the sample, n is the number of samples in the stable region, is the weight coefficient of the physical constraint term, is the number of samples in the critical region, For the jThe weight of the feature, To predict the results, is the gas absorption coefficient, is the X-ray intensity, is the material constant of the X-ray tube, is the accelerating voltage, is the field emission current, is the electron mean free path, Expressed as a natural constant The natural logarithm of base ;

[0038] is the critical region robust loss function, For the i The standard deviation of the noise of the samples, is the L1 regularization coefficient.

[0039] Preferably, the total loss function of the hybrid segmentation model is:

[0040]

[0041]

[0042]

[0043] in, is the total loss function value, is the number of samples in the stable region, Indicates the The true value of the sample, For the model The predicted value of the sample, is the regularization coefficient, is the physical constraint term, is the boundary constraint; Expressed as a natural constant The natural logarithm of base ; is the gas absorption coefficient, For the i The measured value of X-ray intensity of the sample, is the material constant of the X-ray tube, For the i The applied voltage of the sample, is the field emission current, is the electron mean free path, Represents the maximum value function.

[0044] Preferably, the stable region and the critical region are segmented and trained using XGBoost. During the training process, a generative adversarial network discriminator is introduced to perform physical constraint enhancement to complete the construction of a hybrid segmented model, including:

[0045] Setting up the generator and the discriminator ;

[0046] The discriminator The loss function includes real data loss, generated data loss and physical compliance supervision;

[0047] The generator The loss function includes the original MSE loss, the original physical constraint, the adversarial loss, and the physical compliance incentive.

[0048] Preferably, after obtaining the vacuum degree in the vacuum interrupter using the pre-processed X-ray intensity and the mixed segment model, the method further includes:

[0049] Uncertainty quantification is used to output prediction confidence intervals for risk warning. The warning results are output with probability through Monte Carlo sampling based on Bayesian probability, and a three-level warning mechanism is established to complete compound triggering and failure protection, and to perform adaptive updates.

[0050] A vacuum condition assessment system for a vacuum interrupter, comprising:

[0051] A data acquisition module, used to obtain the X-ray intensity when a field emission current exists inside the vacuum interrupter;

[0052] A first data processing module, configured to pre-process the acquired X-ray intensity;

[0053] The second data processing module uses the pre-processed X-ray intensity and the hybrid segmentation model to obtain the vacuum degree in the vacuum interrupter;

[0054] The construction process of the hybrid segmented model is specifically as follows:

[0055] Establish an X-ray-vacuum relationship including the electron mean free path and field emission current, perform logarithmic transformation on the vacuum degree and X-ray intensity, and extract the physical characteristic matrix ;

[0056] The physical feature matrix is ​​analyzed by random forest algorithm. Perform pre-training and transform the physical feature matrix The data in the image are divided into a stable region and a critical region, and asymmetric noise that is adaptively enhanced with vacuum degradation is injected into the critical region.

[0057] XGBoost is used to perform segmented training on the stable region and the critical region. During the training process, a generative adversarial network discriminator is introduced to perform physical constraint enhancement to complete the construction of the hybrid segmented model.

[0058] A computer system includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the above-mentioned method for evaluating the vacuum condition of a vacuum interrupter.

[0059] Compared with the prior art, the present invention has the following beneficial technical effects:

[0060] The present invention provides a method for evaluating the vacuum condition of a vacuum interrupter. The method combines RF with the XGBoost algorithm. The differentiated learning mechanism can realize heterogeneous integration. At the same time, a generative adversarial network (GAN) discriminator is added to enhance physical constraints. Through heterogeneous integrated learning and physical constraint feature engineering, a universal relationship model between X-ray intensity and vacuum degree is constructed. The feature importance of RF can guide the feature selection or weighting of XGBoost to enhance feature interaction. The adversarial mechanism of GAN can coordinate data fitting with physical constraints. With the help of adversarial training, the physical rationality and data authenticity of the model prediction are enhanced. The discriminator is used to double-verify the generated results to ensure that the model conforms to the experimental data distribution and the constraints of theoretical physical laws at the same time, which significantly improves the credibility of the results. The method for evaluating the vacuum condition of a vacuum interrupter proposed by the present invention can achieve high vacuum stage (10 -5 ~10 -2 Pa) stable decay trend and the critical vacuum stage (10 -2 ~10 -1 The robust prediction of the fluctuation data of Pa) is proposed, thus proposing a new idea for evaluating the vacuum condition. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0062] Figure 1 Schematic diagram of a flow chart of a method for evaluating the vacuum condition of a vacuum interrupter in the present invention;

[0063] Figure 2 This is a diagram of the vacuum comparison system;

[0064] Figure 3This is a flow chart of the method for determining the vacuum degree of a vacuum interrupter in Example 2 of the present invention;

[0065] Figure 4 This is a schematic diagram of X-ray measurement;

[0066] Figure 5 A flow chart for establishing a hybrid segmented model in Example 2 of the present invention;

[0067] Figure 6 This is a flow chart of model training and result output in Example 2 of the present invention;

[0068] Figure 7 The figure is a structural diagram of a vacuum condition evaluation system for a vacuum interrupter in the present invention.

[0069] Among them: 1. Nitrogen cylinder, 2. Inlet valve, 3. Ionization silicon tube, 4. Molecular pump, 5. Mechanical pump, 6. Vacuum pool, 7. Needle valve, 8. Vacuum interrupter, 9. X-ray dosimeter probe, 10. Power frequency voltage source, 11. Computer. DETAILED DESCRIPTION

[0070] 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.

[0071] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort shall fall within the scope of protection of the present invention.

[0072] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not require further definition or explanation in subsequent drawings.

[0073] In the description of the embodiments of the present invention, it should be noted that if the terms "upper," "lower," "horizontal," "inner," etc. appear, the orientation or positional relationship indicated is based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the inventive product is typically placed when in use. These terms are merely for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or component referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limitations on the present invention. In addition, the terms "first," "second," etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0074] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0075] In the description of the embodiments of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," "connected," and "connected" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to internal connections between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0076] The present invention is described in further detail below with reference to the accompanying drawings:

[0077] Example 1

[0078] like Figure 1 As shown, the present invention discloses a method for evaluating the vacuum condition of a vacuum interrupter, comprising the following steps:

[0079] S1: Obtain the X-ray intensity when there is field emission current inside the vacuum interrupter;

[0080] S2: Preprocessing the acquired X-ray intensity, wherein the preprocessing includes logarithmic transformation, physical feature construction, and asymmetric noise injection;

[0081] S3: Obtain the vacuum degree in the vacuum interrupter using the pre-processed X-ray intensity and the hybrid segmented model;

[0082] The construction process of the hybrid segmented model is specifically as follows:

[0083] Establish an X-ray-vacuum relationship including the electron mean free path and field emission current, perform logarithmic transformation on the vacuum degree and X-ray intensity, and extract the physical characteristic matrix ;

[0084] The physical feature matrix is ​​analyzed by random forest algorithm. Perform pre-training and transform the physical feature matrix The data in the image are divided into a stable region and a critical region, and asymmetric noise that is adaptively enhanced with vacuum degradation is injected into the critical region.

[0085] XGBoost is used to perform segmented training on the stable region and the critical region. During the training process, a generative adversarial network discriminator is introduced to perform physical constraint enhancement to complete the construction of the hybrid segmented model.

[0086] Specifically, the present invention explores the X-ray intensity based on the bremsstrahlung theory of electron-anode collision. and vacuum degree The physical formulas referred to are as follows:

[0087]

[0088] in, is the X-ray intensity, is the material constant of the X-ray tube, and is related to the atomic number of the anode metal related; is the external applied voltage, is the field emission current, where , is the cathode material constant, is the electric field strength, is the electron mean free path, , is the Boltzmann constant, is the gas temperature, is the diameter of gas molecules, is the gas absorption coefficient, is the actual gas pressure in the system, is the reference vacuum degree.

[0089] Since the analytical solution of the inverse function of vacuum degree with respect to X-ray intensity is complex, machine learning is used to approximate the mapping.

[0090] First, the vacuum degree and X-ray intensity are logarithmically transformed to construct the electron energy characteristics and collision frequency characteristics. Specifically, the electron energy characteristics are:

[0091]

[0092] in, is the electron energy, represents the basic charge of the elementary charge, is the external applied voltage.

[0093] The collision frequency characteristics are:

[0094]

[0095] in, is the electron collision frequency, is the contact gap, is the electron mean free path, is the external applied voltage;

[0096] Should Proportional to ,Right now ,

[0097] is the actual gas pressure in the system, Indicates proportionality.

[0098] That is, by performing logarithmic transformation on the X-ray-vacuum relationship including the electron mean free path and field emission current, the physical characteristic matrix is ​​extracted. , the physical characteristic matrix Contains electron energy characteristics and collision frequency characteristics, where the electron energy characteristics are , the collision frequency characteristic is .

[0099] A critical region noise model is set for data enhancement. To strengthen physical constraints, the traditional Gaussian noise is upgraded to an asymmetric noise model. Asymmetric noise that is adaptively enhanced as the vacuum deteriorates is injected into the critical region to reflect the skewed distribution of the signal caused by gas ionization when the vacuum degree decreases. The noise model is:

[0100]

[0101] in, is the X-ray intensity measurement value after adding noise, is the logarithmically transformed X-ray intensity measurement value, The mean is 0 and the standard deviation is Asymmetric Gaussian distributed noise ensures that the noise intensity changes adaptively with the signal strength; is the skewness coefficient, is the system vacuum degree, express for function.

[0102] Establish a hybrid segmented model and determine the critical region based on the standard deviation of the training data. Set the total loss function and introduce physical constraints based on the Bremsstrahlung theory. and boundary constraints , the implementation formula is as follows:

[0103]

[0104]

[0105]

[0106] Right now

[0107] in, is the number of samples in the stable region, Indicates the The true value of the sample, For the model The predicted value of the sample, is the regularization coefficient, which controls the weight of physical constraints and boundary constraints; is the total loss function value, Expressed as a natural constant The natural logarithm of base ; is the gas absorption coefficient, For the i The measured value of X-ray intensity of the sample, is the material constant of the X-ray tube, For the i The applied voltage of the sample, is the field emission current, is the electron mean free path, represents the maximum value function, is the physical constraint term, is the boundary constraint; Expressed as a natural constant The natural logarithm of base ; is the gas absorption coefficient, For the i The measured value of X-ray intensity of the sample, is the material constant of the X-ray tube, For the i The applied voltage of the sample, is the field emission current, is the electron mean free path, represents the maximum value function, i is the total number of samples.

[0108] Add a Generative Adversarial Network (GAN) discriminator to the original hybrid model to enhance physical constraints and ensure that the prediction results are doubly credible in terms of physical laws and real data distribution. Set the generator and the discriminator , output the predicted value of X-ray intensity And the binary discrimination results :

[0109]

[0110]

[0111] in, For the model The predicted value of the sample, is the generator model, For the i The input feature vector of samples, For the i The measured value of X-ray intensity of the sample, For the i The applied voltage of the sample, For the i The contact gap of each sample;

[0112] is the discriminator output, is the probability of judging the authenticity of the data, is the probability of physical constraint satisfaction; , used to determine whether the data comes from the real distribution, , used to determine whether the physical constraints are met, is the real sample data.

[0113] The discriminator loss function is introduced, which mainly consists of the following three parts: real data loss, generated data loss, and physical compliance supervision:

[0114]

[0115] in, is the total loss function of the discriminator, For the real data distribution expectations, For the discriminator to the real data The logarithm of the discriminant probability, Input feature distribution to the generator expectations, is the predicted value of X-ray intensity output by the generator, is the input feature vector, is the weight coefficient of physical compliance supervision, For the joint distribution expectations, For the discriminator to The probability of physical compliance, is the physical compliance indicator function; Expressed as a natural constant The natural logarithm of base .

[0116] , is the physical compliance indicator function;

[0117] is the joint distribution of real sample data and input features The physical constraints of For real sample data Boundary constraints of

[0118] is the tolerance threshold. Represents the real sample The distribution of Represents the generator input features X distribution.

[0119] The generator loss function is introduced, which mainly consists of the following four parts: original mean squared error (MSE) loss, original physical constraint, adversarial loss, and physical compliance incentive:

[0120]

[0121] in, is the total loss function of the generator, Input feature distribution to the generator expectations, is the predicted value of X-ray intensity output by the generator, is the true X-ray intensity value, is the physical constraint term, is the boundary constraint, is the probability of the discriminator judging the authenticity of the generated data, is the input feature vector, is the probability of physical compliance of the discriminator to the generated data; Expressed as a natural constant The natural logarithm of base ;

[0122] , is the weight coefficient of the physical constraint term, usually set to [0.1, 0.5];

[0123] The weight coefficient of the adversarial loss term balances the motivation of the generator to deceive the discriminator, usually set to [0.01, 0.1]; is the weight coefficient of the physical compliance incentive item.

[0124] The generator scores the authenticity of the generated data for the discriminator, and hopes that this value is close to 1 (deceiving the discriminator).

[0125] For the implementation of physical compliance judgment, the degree of constraint violation is directly calculated based on the hard-coded physical rules. :

[0126]

[0127] in, is the Sigmoid function, is the scaling factor, is the degree of violation of the physical constraint, is the degree of boundary constraint violation; is the X-ray intensity value predicted by the model after logarithmic transformation, is the input feature vector.

[0128] After adding the generative adversarial network (GAN) discriminator, a dual discriminant mechanism is formed to simultaneously constrain data distribution and physical laws, and adversarial training reduces the prediction variance in the critical region.

[0129] In order to generate feature weights and divide the data area, random forest pre-training is performed and the feature matrix is ​​input. ,

[0130] in, For the i The measured value of X-ray intensity of the sample, For the i The applied voltage of the sample, For the i The contact gap of each sample, i is the total number of samples. Z Each decision tree selects features through the following splitting criteria, calculates feature importance, and determines the critical area, that is, through the Gini coefficient and feature importance, the physical feature matrix The data in is divided into stable area and critical area;

[0131]

[0132]

[0133] in, For the dataset Gini impurity, also known as the Gini coefficient, is the total number of categories, is the input sample set, For the r A subset of samples of categories, is the total number of samples;

[0134] For the j The global importance weight of each feature, that is, the feature importance when dividing the stable area and the critical area, is the total number of decision trees, is the current decision tree number, For the z The first tree j The number of samples of the left child node when the feature is split, for u The sum of the split gains of all features of all trees under the total number of features, Indicates the samples, For the The first tree j The importance of nodes when splitting features;

[0135] NI is the node importance and VI is the number of splits.

[0136] Then perform XGBoost training in stages, add a lightweight physical network layer before XGBoost to directly solve the electron mean free path As a supplement to feature engineering, the approximate analytical solution of is implemented as follows:

[0137]

[0138] in, is the mean free path of electrons in the gas, is the system vacuum degree, is the diameter of gas molecules, is a proportional constant that is related to gas type and temperature.

[0139] The objective function set in the stable region and the robust loss function used in the critical region are as follows:

[0140]

[0141]

[0142] in, is the objective function in the stable region, For the i The true X-ray intensity of the sample, For the model i The predicted value of the sample, n is the number of samples in the stable region, is the weight coefficient of the physical constraint term, determined by cross-validation, is the number of samples in the critical region, For the j The weight of the feature, To predict the results, is the gas absorption coefficient, is the X-ray intensity, is the material constant of the X-ray tube, is the accelerating voltage, is the field emission current, is the electron mean free path, Expressed as a natural constant The natural logarithm of base ;

[0143] is the critical region robust loss function, For the i The standard deviation of the noise of the samples, It is the L1 regularization coefficient, which controls feature sparsity, reduces the impact of non-critical features, and improves the model's noise resistance.

[0144] In order to adaptively distribute feature importance and be compatible with physical priors, dynamic feature weighting is performed:

[0145]

[0146] in, is the weighted feature matrix, is the input feature vector, is the weight vector, Represents the Hadamard product, which realizes the multiplication of corresponding matrix elements. For the j The exponential transformation of the importance score of the features, where I j represents the feature importance, It is the exponentially transformed sum of all feature importance scores, used to normalize the weights. is the total number of feature importance scores, For the The importance score of a feature.

[0147] Finally, feature construction and vacuum degree prediction are carried out, and uncertainty quantification is introduced. The implementation process is as follows:

[0148]

[0149]

[0150]

[0151] Where, is the vacuum degree prediction feature matrix, is the measured value of X-ray intensity, Externally applied voltage, Contact gap, is the basic charge of the elementary charge, is the Boltzmann constant, is the gas temperature, is the actual gas pressure in the system;

[0152] is the reference vacuum degree, is the predicted vacuum ratio, is the predicted vacuum degree, is the number of sampling times, is the sum of the prediction values ​​of the integrated model;

[0153] is the standard deviation of the current forecast value, For the t subsampled prediction output, is the predicted vacuum degree, is the sensitivity of the predicted value to the input measured value of X-ray intensity, is the error of the measured value of X-ray intensity.

[0154] The above prediction results are subjected to Monte Carlo sampling to achieve Bayesian probability output:

[0155]

[0156] in, Indicates that the prediction result belongs to the level q The probability of is the total number of tests, For the s The predicted value of the test, is the indicator function.

[0157] when Falling on grade q When the vacuum degree is within the corresponding range, the function value is 1; when When it is not within the range, the function value is 0. It is used to determine whether the predicted value of each sample meets the level q range conditions.

[0158] Based on the probability output of Monte Carlo sampling, a three-level early warning mechanism is established:

[0159]

[0160] in, is the prediction variance, which reflects the uncertainty of the model’s prediction of the current state. is the uncertainty threshold; , is the mean historical forecast error, is the standard deviation of historical forecast values;

[0161] is the rate of change of failure probability, is the actual pressure change in the system, is the time interval.

[0162] Theoretically, this method can reduce the relative error of vacuum degree prediction, improve accuracy, enhance the robustness of the critical region, meet the needs of real-time monitoring, and thus realize the evaluation of the vacuum degree condition of the vacuum interrupter.

[0163] This paper proposes an X-ray dose tracking and assessment method that combines random forest (RF), generative adversarial network (GAN), and XGBoost, enabling assessment of the vacuum condition of vacuum interrupters. First, based on bremsstrahlung theory, an explicit physical correlation between X-ray intensity and vacuum is established. Through microscopic modeling of the electron free path, key parameters such as the anode atomic number and field emission current are incorporated into an interpretable framework, resulting in a clearer physical meaning than traditional black-box models. In the feature extraction stage, a logarithmic transformation is performed to compress the data scale, highlighting the physical characteristics of electron energy and collision frequency. Furthermore, an asymmetric noise model is employed to overcome the limitations of Gaussian noise. The skewness coefficient is used to simulate the ionization skew distribution during vacuum degradation, making data enhancement more physically accurate. In the hybrid segmented modeling stage, random forest pre-training dynamically demarcates the critical region using the Gini coefficient and feature importance (NI, VI), addressing the rigidity of traditional thresholding methods. XGBoost staged training employs a physically constrained robust loss to suppress outlier interference in the critical region while maintaining high accuracy in the stable region. The present invention also carries out a deep fusion of physics and machine learning: the lightweight physical network layer directly calculates the analytical solution of the electron free path as the feature input, realizing the cross-scale coupling of mechanism and data; dynamic feature weighting modifies the physical parameter weights through the importance of random forest features, balancing data-driven and prior knowledge. In order to perform double verification, the generative adversarial network is introduced: the GAN framework implements a dual discrimination mechanism, Used to verify data distribution, Hard-coded constraints enforce compliance The generator loss dynamically balances MSE accuracy, physical compliance, and countermeasures against deception by setting loss term weight coefficients, aiming to reduce the prediction variance in the critical region. Finally, to adapt to engineering applications, this invention uses uncertainty quantification to output prediction confidence intervals and support risk warnings. Based on Bayesian probability, Monte Carlo sampling is used to output the probability of prediction results, establishing a three-level warning mechanism that implements composite triggering, failure protection, and adaptive updates.

[0164] Example 2

[0165] In order to verify the effectiveness of the evaluation method in the present invention, the following vacuum comparison system is established. The system consists of a vacuum interrupter 8, whose sample model can be BD390, BD392 or BD395, a vacuum pool 6, a mechanical pump 5, a molecular pump 4, an air inlet valve 2, a needle valve 7, a nitrogen bottle 1 and an ionized silicon tube 3. Figure 2 It is a vacuum comparison system, including a nitrogen bottle 1, an air inlet valve 2, an ionized silicon tube 3, a molecular pump 4, a mechanical pump 5, a vacuum pool 6, a needle valve 7, and a vacuum interrupter 8. One side of the vacuum interrupter 8 is connected to the vacuum pool 6 via the needle valve 7, and the other side of the vacuum pool 6 is connected to the nitrogen bottle 1 via the air inlet valve 2. The vacuum pool 6 is equipped with an ionized silicon tube 3, and the mechanical pump 5 and the molecular pump 4 are connected in series and then connected to the vacuum pool 6. Figure 3 This is a flow chart of the method for determining the vacuum degree of an X-ray vacuum interrupter based on random forest-GAN-XGBoost.

[0166] During the experimental preparation stage, a vacuum pump group (mechanical pump and molecular pump working in series) is used to evacuate the vacuum chamber until the system reaches the basic vacuum limit. Inert gas is injected into the cavity through a controllable air intake system (such as an N2 gas source with a precision needle valve). When the gas injection rate and the pumping rate reach a dynamic equilibrium, a stable target vacuum degree P0 is formed in the cavity. Since the arc extinguishing chamber and the vacuum chamber are connected through an air duct, the internal pressures of the two can be considered equal. Subsequently, all gas path valves are closed, the spacing between the movable electrodes is adjusted to the experimental set value, and a 60kV power frequency test voltage is gradually applied. Under this condition, field-induced electron emission occurs on the cathode surface, and high-energy electrons bombard the anode target under electric field acceleration, generating X-rays based on the bremsstrahlung effect. When the ray intensity reaches the detectable threshold, a calibrated X-ray dosimeter (1 meter away from the arc extinguishing chamber) is used to measure the radiation intensity P d0 , complete the first data calibration. By adjusting the intake parameters to change the system vacuum to P1, repeat the above measurement process to obtain P d1 , and establish characteristic data sets of vacuum degree-X-ray intensity in turn.

[0167] Figure 4 This is a schematic diagram of an X-ray measurement, including a vacuum interrupter 8, an X-ray dosimeter probe 9, a power-frequency voltage source 10, and a computer 11. The vacuum interrupter 8 and the power-frequency voltage source 10 are connected in series to form a loop. The X-ray dosimeter is placed 1 meter away from the vacuum interrupter 8 and then connected to the computer 11 to collect data in real time.

[0168] After data acquisition is completed, the vacuum degree and X-ray intensity are logarithmically transformed to construct the electron quantity characteristics and collision frequency characteristics; at the same time, a critical region asymmetric noise model is set to enhance the data.

[0169] A hybrid segmented model is then established, dynamically determining the critical region based on the standard deviation of the training data. A total loss function is set, and physical and boundary constraints are introduced based on Bremsstrahlung theory. A generative adversarial network (GAN) discriminator is added to the original hybrid model to enhance physical constraints, ensuring that the prediction results are both credible based on physical laws and the distribution of real data. A generator and a discriminator are set up, and both generator and discriminator loss functions are introduced. Physical compliance is determined based on hard-coded physical rules, and the output is a predicted X-ray intensity value and a binary discrimination result. Figure 5 Create a flow chart for the hybrid segmentation model.

[0170] Random forest pre-training is then performed, with the feature matrix input, feature importance calculated, and critical regions determined. XGBoost phased training is then performed, with a lightweight physical network layer added before XGBoost to directly obtain an approximate analytical solution for the electron mean free path. Dynamic feature weighting is also introduced to supplement feature engineering. Finally, feature construction and vacuum degree prediction are performed, while uncertainty quantification is introduced. Monte Carlo sampling is performed on the prediction results to achieve Bayesian probability output, establishing a three-level early warning mechanism. Figure 6 Flowchart for model training and result output.

[0171] Example 3

[0172] In addition, if Figure 7 As shown, the present invention also discloses a vacuum condition evaluation system for a vacuum interrupter, comprising:

[0173] A data acquisition module, used to obtain the X-ray intensity when a field emission current exists inside the vacuum interrupter;

[0174] A first data processing module is used to preprocess the acquired X-ray intensity, wherein the preprocessing includes logarithmic transformation, physical feature construction and asymmetric noise injection;

[0175] A second data processing module is used to obtain the vacuum degree in the vacuum interrupter using the pre-processed X-ray intensity and the pre-built hybrid segmented model;

[0176] The construction process of the hybrid segmented model is specifically as follows:

[0177] Establish an X-ray-vacuum relationship including the electron mean free path and field emission current, perform logarithmic transformation on the vacuum degree and X-ray intensity, and extract the physical characteristic matrix ;

[0178] The physical feature matrix is ​​analyzed by random forest algorithm. Perform pre-training and transform the physical feature matrix The data in the image are divided into a stable region and a critical region, and asymmetric noise that is adaptively enhanced with vacuum degradation is injected into the critical region.

[0179] XGBoost is used to perform segmented training on the stable region and the critical region. During the training process, a generative adversarial network discriminator is introduced to perform physical constraint enhancement to complete the construction of the hybrid segmented model. In addition, a schematic diagram of a terminal device is provided in one embodiment of the present invention. The terminal device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above-mentioned method for evaluating the vacuum condition of a vacuum interrupter are implemented. Alternatively, when the processor executes the computer program, the functions of each module / unit in the above-mentioned device embodiments are implemented.

[0180] The computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to accomplish the present invention.

[0181] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0182] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0183] The memory may be used to store the computer programs and / or modules, and the processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory.

[0184] If the module / unit integrated in the terminal device is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention can implement all or part of the process steps in the above-mentioned method embodiments by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium. It should be noted that the content of the computer-readable medium can be appropriately increased or decreased based on the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media does not include electric carrier signals and telecommunication signals.

[0185] 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 vacuum condition of a vacuum interrupter, characterized in that: The following steps are involved: S1: Obtain the X-ray intensity when there is field emission current inside the vacuum interrupter; S2: pre-processing the acquired X-ray intensity; S3: Obtain the vacuum degree in the vacuum interrupter using the pre-processed X-ray intensity and the hybrid segmented model; The construction process of the hybrid segmented model is specifically as follows: Establish an X-ray-vacuum relationship including the electron mean free path and field emission current, perform logarithmic transformation on the vacuum degree and X-ray intensity, and extract the physical characteristic matrix ; The physical feature matrix is ​​analyzed by random forest algorithm. Perform pre-training and transform the physical feature matrix The data in the image are divided into a stable region and a critical region, and asymmetric noise that is adaptively enhanced with vacuum degradation is injected into the critical region. XGBoost is used to perform segmented training on the stable region and the critical region. During the training process, a generative adversarial network discriminator is introduced to perform physical constraint enhancement to complete the construction of the hybrid segmented model.

2. The method for evaluating the vacuum condition of a vacuum interrupter according to claim 1, wherein: The establishing of an X-ray-vacuum degree relationship including electron mean free path and field emission current includes: Based on the bremsstrahlung theory, an X-ray-vacuum relationship including the electron mean free path and the field emission current is established. The X-ray-vacuum relationship is specifically: in, is the X-ray intensity, is the material constant of the X-ray tube, is the external applied voltage, is the field emission current, is the electron mean free path, is the gas absorption coefficient, is the actual gas pressure in the system, is the reference vacuum degree, is the cathode material constant, is the electric field strength, is the Boltzmann constant, is the gas temperature, is the diameter of a gas molecule.

3. The method for evaluating the vacuum condition of a vacuum interrupter according to claim 1, wherein: The physical characteristic matrix Including electron energy characteristics and collision frequency characteristics: The electron energy characteristics are: in, is the electron energy, represents the basic charge of the elementary charge, is the external applied voltage; The collision frequency characteristics are: in, is the electron collision frequency, is the contact gap, is the electron mean free path.

4. The method for evaluating the vacuum condition of a vacuum interrupter according to claim 1, wherein: In the asymmetric noise injected into the critical region that is adaptively enhanced with vacuum degradation, the noise model is: in, is the X-ray intensity measurement value after adding noise, is the logarithmically transformed X-ray intensity measurement value, The mean is 0 and the standard deviation is Asymmetric Gaussian noise, is the skewness coefficient, is the system vacuum degree, express for function.

5. The method for evaluating the vacuum condition of a vacuum interrupter according to claim 1, wherein: The physical feature matrix is ​​analyzed by random forest algorithm Perform pre-training and transform the physical feature matrix The data in is divided into stable area and critical area, including: Through the Gini coefficient and feature importance, the physical feature matrix The data in is divided into stable area and critical area; The objective function of the stable region is: The robust loss function used in the critical region is: in, is the objective function in the stable region, For the i The true X-ray intensity of the sample, For the model i The predicted value of the sample, n is the number of samples in the stable region, is the weight coefficient of the physical constraint term, is the number of samples in the critical region, For the j The weight of the feature, To predict the results, is the gas absorption coefficient, is the X-ray intensity, is the material constant of the X-ray tube, is the accelerating voltage, is the field emission current, is the electron mean free path, Expressed as a natural constant The natural logarithm of base ; is the critical region robust loss function, For the i The standard deviation of the noise of the samples, is the L1 regularization coefficient.

6. The method for evaluating the vacuum condition of a vacuum interrupter according to claim 1, wherein: The total loss function of the hybrid segmentation model is: in, is the total loss function value, is the number of samples in the stable region, Indicates the The true value of the sample, For the model The predicted value of the sample, is the regularization coefficient, is the physical constraint term, is the boundary constraint; Expressed as a natural constant The natural logarithm of base ; is the gas absorption coefficient, For the i The measured value of X-ray intensity of the sample, is the material constant of the X-ray tube, For the i The applied voltage of the sample, is the field emission current, is the electron mean free path, Represents the maximum value function.

7. The method for evaluating the vacuum condition of a vacuum interrupter according to claim 1, wherein: The stable region and the critical region are segmented and trained using XGBoost. During the training process, a generative adversarial network discriminator is introduced to perform physical constraint enhancement to complete the construction of a hybrid segmented model, including: Setting up the generator and the discriminator ; The discriminator The loss function includes real data loss, generated data loss and physical compliance supervision; The generator The loss function includes the original MSE loss, the original physical constraint, the adversarial loss, and the physical compliance incentive.

8. The method for evaluating the vacuum condition of a vacuum interrupter according to claim 1, wherein: After obtaining the vacuum degree in the vacuum interrupter using the pre-processed X-ray intensity and the mixed segmented model, the method further includes: Uncertainty quantification is used to output prediction confidence intervals for risk warning. The warning results are output with probability through Monte Carlo sampling based on Bayesian probability, and a three-level warning mechanism is established to complete compound triggering and failure protection, and to perform adaptive updates.

9. A vacuum condition assessment system for a vacuum interrupter, characterized in that: include: A data acquisition module, used to obtain the X-ray intensity when a field emission current exists inside the vacuum interrupter; A first data processing module, configured to pre-process the acquired X-ray intensity; The second data processing module uses the pre-processed X-ray intensity and the hybrid segmentation model to obtain the vacuum degree in the vacuum interrupter; The construction process of the hybrid segmented model is specifically as follows: Establish an X-ray-vacuum relationship including the electron mean free path and field emission current, perform logarithmic transformation on the vacuum degree and X-ray intensity, and extract the physical characteristic matrix ; The physical feature matrix is ​​analyzed by random forest algorithm. Perform pre-training and transform the physical feature matrix The data in the image are divided into a stable region and a critical region, and asymmetric noise that is adaptively enhanced with vacuum degradation is injected into the critical region. XGBoost is used to perform segmented training on the stable region and the critical region. During the training process, a generative adversarial network discriminator is introduced to perform physical constraint enhancement to complete the construction of the hybrid segmented model.

10. A computer system comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the vacuum condition evaluation method of a vacuum interrupter as described in any one of claims 1 to 8.

Images