Grounding insulation state identification method and system for train-mounted power supply system

Abstract

The invention provides a ground insulation state identification method for a train-mounted power supply system, and relates to the technical field of train-mounted power supply systems, and the method comprises the steps: constructing a characteristic variable at different ground fault positions based on the obtained relation between a DC bus voltage and a ground detection voltage, respectively constructing a feature vector set and a classification enabling mechanism according to the feature variables; judging whether a train-mounted power supply system is in a fault state or not based on a classification enabling mechanism, constructing a total data set of feature vectors as input of a ground fault classification model, and outputting a ground fault occurrence position; selecting a corresponding voltage signal as an input of the fault grounding resistance real-time prediction model, and outputting a real-time prediction result of the fault grounding resistance; and identifying the grounding insulation state of the train-mounted power supply system according to the real-time prediction result of the fault grounding resistance. According to the invention, the anti-interference capability and the fault diagnosis precision are effectively improved, the resistance prediction error is reduced, and the grounding insulation state identification precision is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of train onboard power supply systems, and in particular to a method and system for identifying a grounding insulation state of a train onboard power supply system. Background Art

[0002] The train's onboard power supply system provides power for various electrical loads on the train. As the service time increases, the train's onboard power supply system is susceptible to equipment aging and environmental factors, resulting in electrical failures. Such failures may lead to a poor riding experience and even cause safety issues. In order to ensure the safe and reliable operation of the train and provide passengers with a comfortable riding experience, efficient and reliable fault diagnosis methods become particularly important.

[0003] In recent years, research on ground fault diagnosis has mainly focused on ground faults in transmission lines in power transmission systems. In contrast, the on-board power supply system of a train is directly connected to multiple converter modules with almost no transmission distance. Its ground fault is mainly caused by grounding anomalies due to insulation degradation at different locations in the converter system. Research on accurately and quickly identifying and locating ground faults in the on-board power supply system of a train is not yet mature.

[0004] The prior art discloses a method for locating and identifying ground faults in an electric locomotive power supply system based on time series features. The method uses historical system signal data to construct a feature vector with fault location discrimination, extracts feature indicators in a sliding window, attaches time series features, improves the distinguishability of fault types, reduces noise interference, and constructs a ground fault location model through machine learning. The nonlinear relationship between features and fault types is mined to achieve fault location. However, the scheme fails to further derive the magnitude of the ground insulation resistance to evaluate the severity of the system ground fault. The prior art also discloses a method for evaluating the ground insulation state. The method first obtains operating parameters and states, and then determines the ground fault based on the operating parameters and states. Whether there is a ground fault; if so, obtaining characteristic indicator data and operating condition information based on the operating parameters and status; obtaining the fault type based on the characteristic indicator and the operating condition information; obtaining the ground insulation resistance based on the fault type; evaluating the ground insulation state based on the ground insulation resistance, but this method uses a decision tree based on an inequality relationship to make a judgment in the process of fault classification judgment, which is difficult to deal with noise and uncertainty, lacks adaptive ability, and has low accuracy in real-time fault diagnosis; at the same time, the obtained ground insulation resistance is only calculated through a physical model, which is greatly affected by signal fluctuation interference, resulting in a large resistance prediction error, thereby affecting the accuracy of ground insulation state identification. Summary of the invention

[0005] In order to solve the problems that the above-mentioned prior art has poor anti-interference ability, low real-time fault diagnosis accuracy, large resistance prediction error, and poor accuracy in identifying the grounding insulation state, the present invention proposes a grounding insulation state identification method and system for a train on-board power supply system, which effectively improves the anti-interference ability and fault diagnosis accuracy, reduces the resistance prediction error, and ensures the accuracy of grounding insulation state identification.

[0006] In order to achieve the above technical effects, the technical solution of the present invention is as follows: A method for identifying the grounding insulation state of a train onboard power supply system, the method comprising the following steps: S1. Obtain the relationship between the DC bus voltage and the ground detection voltage of the train onboard power supply system under normal conditions; S2. Under different ground fault positions, based on the relationship between the DC bus voltage and the ground detection voltage, construct characteristic variables, and construct characteristic vector sets and classification enabling mechanisms according to the characteristic variables; S3. Based on the classification enabling mechanism, determine whether the train onboard power supply system is in a fault state. If so, execute S4; if not, it is in a normal state; S4. Using the feature vector set, construct a total data set of feature vectors under different ground fault conditions; S5. Using the total data set as the input of a preset ground fault classification model, and outputting the ground fault location by the ground fault classification model; S6. Based on the location of the ground fault, select the voltage signal corresponding to the location of the ground fault as the input of the preset real-time prediction model of the fault ground resistance, and output the real-time prediction result of the fault ground resistance; S7. Identify the grounding insulation status of the train's onboard power supply system based on the real-time prediction result of the fault grounding resistance.

[0007] Preferably, S1 obtains the DC bus voltage U of the train onboard power supply system in a normal state. dc With ground detection voltage U jd The calculation expression of the relationship is as follows:

[0008] S1 also includes obtaining the fault grounding resistance and DC bus voltage U under different grounding fault positions dc , ground detection voltage U jd relationship.

[0009] Preferably, in S2, based on the relationship between the DC bus voltage and the ground detection voltage, a characteristic variable is constructed. The calculation expression is as follows:

[0010] According to the characteristic variables Construct the feature vector set The calculation process includes:

[0011] in, represents the mean value of the characteristic variable F within a sliding window period at the kth moment, N represents the total number of sampling points in a sliding window, , represents the variance of the feature variable F within a sliding window period at the kth moment, Indicates the ground detection voltage U jd The complex spectrum value at frequency f, Indicates the ground detection voltage U jd The value of the i-th sampling point, represents the harmonic amplitude, Indicates the ground detection voltage U jd The complex spectrum value at m times the frequency f, Indicates total harmonic distortion.

[0012] Preferably, the classification enabling mechanism includes: Based on the characteristic variable F, the cumulative deviation S of the characteristic variable F is calculated k The following formula:

[0013] Among them, E represents the mean of the feature variable F, and V represents the variance of the feature variable F; Determine the cumulative deviation S k Is it greater than the fault threshold h? If so, the train's onboard power supply system is in a fault state; if not, the train's onboard power supply system is in a normal state.

[0014] Preferably, the calculation expression for constructing a total data set D of feature vectors under different ground fault states by using the feature vector set is as follows:

[0015] in, The characteristic vector set representing the DC link positive pole grounding fault within the sliding window period, represents the mean value of the characteristic variable F of the DC link positive pole grounding fault within the sliding window period, represents the variance of the characteristic variable F of the DC link positive pole grounding fault within the sliding window period, Indicates the harmonic amplitude of the DC link positive pole grounding fault, Indicates the total harmonic distortion of the DC link positive pole grounding fault; The characteristic vector set representing the negative pole grounding fault of the DC link within the sliding window period, represents the mean value of the characteristic variable F of the negative pole grounding fault in the DC link within the sliding window period, represents the variance of the characteristic variable F of the negative pole grounding fault in the DC link within the sliding window period, Indicates the harmonic amplitude of the negative ground fault in the DC link. Indicates the total harmonic distortion of the negative pole grounding fault in the DC link; The characteristic vector set representing the ground fault at the front end of the reactor within the sliding window period, represents the mean value of the characteristic variable F of the reactor front-end ground fault within the sliding window period, The variance of the characteristic variable F of the ground fault at the front end of the reactor within the sliding window period, Indicates the harmonic amplitude of the ground fault at the front end of the reactor. Indicates the total harmonic distortion of the ground fault at the front end of the reactor; The feature vector set representing the RC input positive ground fault within the sliding window period, represents the mean value of the characteristic variable F of the RC input positive pole grounding fault within the sliding window period, represents the variance of the characteristic variable F of the RC input positive pole grounding fault within the sliding window period, Indicates the harmonic amplitude of the RC input positive ground fault, Indicates the total harmonic distortion of the RC input positive pole grounding fault; The set of feature vectors representing the RC input negative ground fault within the sliding window period, represents the mean value of the characteristic variable F of the RC input negative pole grounding fault within the sliding window period, represents the variance of the characteristic variable F of the RC input negative pole grounding fault within the sliding window period, Indicates the harmonic amplitude of the RC input negative ground fault, Indicates the total harmonic distortion of the RC input negative pole grounding fault.

[0016] Preferably, the ground fault classification model includes an input layer, a convolution layer, a pooling layer, a first fully connected layer and an output layer connected in sequence, the convolution kernel size of the convolution layer is 3×33, the pooling layer is maximum pooling, the pooling window size is 2×22, the number of nodes in the first fully connected layer is 50, the output layer is 6-dimensional, and the activation function of the output layer is softmax.

[0017] Preferably, the ground fault classification model is trained by back propagation and gradient descent method until the training round reaches a preset threshold or the loss function converges, then the training of the ground fault classification model is completed to obtain a trained ground fault classification model.

[0018] Preferably, the real-time prediction model for fault grounding resistance includes a plurality of second fully connected layers, each of which is provided with 64 neurons and uses a ReLU activation function for output.

[0019] Preferably, the total loss function used by the real-time prediction model for fault grounding resistance in the training process is as follows:

[0020] in, represents the total loss function, Indicates data error, represents the first physical constraint error, represents the weight coefficient of the first physical constraint error, represents the second physical constraint error, represents the weight coefficient of the second physical constraint error, represents the third physical constraint error, represents the weight coefficient of the third physical constraint error, represents the fourth physical constraint error, represents the weight coefficient of the fourth physical constraint error, represents the fifth physical constraint error, Represents the weight coefficient of the fifth physical constraint error.

[0021] The present invention also proposes a grounding insulation state identification system for a train onboard power supply system, and the grounding insulation state identification system is implemented based on the grounding insulation state identification method for the train onboard power supply system, and comprises: An acquisition module is used to obtain the relationship between the DC bus voltage and the ground detection voltage under the normal state of the train onboard power supply system; A first construction module is used to construct characteristic variables based on the relationship between the DC bus voltage and the ground detection voltage at different ground fault locations, and to respectively construct a characteristic vector set and a classification enabling mechanism according to the characteristic variables; A judgment module, used to judge whether the train onboard power supply system is in a fault state based on the classification enabling mechanism; The second construction module is used to construct a total data set of feature vectors under different ground fault states using the feature vector set when the on-board power supply system of the train is in a fault state; A ground fault classification module, used for taking the total data set as an input of a preset ground fault classification model, and the ground fault classification model outputs a ground fault occurrence location; A fault grounding resistance real-time prediction module is used to select a voltage signal corresponding to the grounding fault location as an input of a preset fault grounding resistance real-time prediction model based on the grounding fault location, and output a real-time prediction result of the fault grounding resistance; The grounding insulation status identification module is used to identify the grounding insulation status of the train onboard power supply system according to the real-time prediction result of the fault grounding resistance.

[0022] Compared with the prior art, the technical solution of the present invention has the following beneficial effects: The present invention proposes a method and system for identifying the grounding insulation state of a train on-board power supply system. First, by acquiring the relationship between the DC bus voltage and the grounding detection voltage of the train on-board power supply system under a normal state, a feature vector set and a classification enabling mechanism are constructed in combination with multi-dimensional feature variable extraction under different grounding fault positions, thereby significantly improving the anti-interference ability of the system; then, after confirming the fault of the train on-board power supply system through the classification enabling mechanism, a total data set of feature vectors under different grounding fault states is constructed, and the total data set is used as the input of a preset grounding fault classification model, and the location of the grounding fault is output, thereby realizing data-driven fault classification and effectively improving the real-time diagnosis accuracy of the location of the grounding fault; then, by selecting the voltage signal corresponding to the location of the grounding fault as the input of a preset real-time prediction model of fault grounding resistance, a real-time prediction result of the fault grounding resistance is output, thereby effectively reducing the prediction error of the resistance and ensuring the accuracy of the real-time prediction result of the fault grounding resistance; further, according to the real-time prediction result of the fault grounding resistance, the grounding insulation state of the train on-board power supply system is identified, thereby effectively improving the accuracy of grounding insulation state identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A flowchart showing a method for identifying the ground insulation state of a train onboard power supply system proposed in an embodiment of the present invention; Figure 2 A circuit diagram showing a train onboard power supply system proposed in an embodiment of the present invention; Figure 3 A block diagram showing the principle of identifying the grounding insulation state of a train onboard power supply system proposed in an embodiment of the present invention; Figure 4 The present invention shows a structural block diagram of a grounding insulation state identification system for a train onboard power supply system proposed in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The drawings are for illustrative purposes only and should not be construed as limiting the present patent; It is understandable to those skilled in the art that some well-known contents may be omitted in the drawings; The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0025] Example 1 like Figure 1 As shown, this embodiment proposes a method for identifying the ground insulation state of a train onboard power supply system, and the method includes the following steps: S1. Obtain the relationship between the DC bus voltage and the ground detection voltage of the train onboard power supply system under normal conditions; In S1, see Figure 2 The train onboard power supply system includes a vacuum circuit breaker, a traction transformer, a synchronous transformer TV, a current transformer TA, a first grounding resistor , Second grounding resistance , the third grounding resistance , the fourth grounding resistor , Fifth grounding resistance , first diode VD1, second diode VD2, first thyristor VT3, second thyristor VT4, inductor L, capacitor C, first detection resistor , the second detection resistor , the third detection resistor , DC bus voltage sensor , Ground detection circuit voltage sensor and control system, the external power supply network is connected to one end of the vacuum circuit breaker, the other end of the vacuum circuit breaker is connected to the primary side of the traction transformer, one end of the secondary side of the traction transformer is respectively connected to one end of the synchronous transformer TV and the current transformer TA, the other end of the secondary side of the traction transformer is respectively connected to the other end of the synchronous transformer TV, the fifth grounding resistor one end, the positive terminal of the first thyristor VT3 and the negative terminal of the second thyristor VT4, the fifth grounding resistor The other end of the grounded current transformer TA is connected to the fourth grounding resistor one end of the first diode VD1, the positive terminal of the second diode VD2, and the fourth grounding resistor The other end of the first diode VD1 is connected to the negative end of the first thyristor VT3 and the third grounding resistor One end of the inductor L and one end of the third grounding resistor The other end of the inductor L is connected to one end of the capacitor C and the first grounding resistor One end of the first sense resistor One end of the first grounding resistor The other end of the first detection resistor is grounded. The other end of the second detection resistor One end and the third sense resistor One end of the third sense resistor The other end of the second diode VD2 is grounded. The positive end of the second diode VD2 is connected to the positive end of the second thyristor VT4, the other end of the capacitor C, and the second grounding resistor The other end of the second sense resistor The other end of the DC bus voltage sensor The first sense resistor is connected in parallel and the third sense resistor Both sides of the circuit; ground detection circuit voltage sensor The first sense resistor is connected in parallel and the second sense resistor The control system is electrically connected to the synchronous transformer TV, the current transformer TA, the DC bus voltage sensor and ground detection circuit voltage sensor .

[0026] Among them, the first detection resistor , the second detection resistor , the third detection resistor , DC bus voltage sensor and ground detection circuit voltage sensor The ground detection circuit is composed of R1 equal to R2, the inductor L and the capacitor C form an intermediate DC circuit, the first diode VD1, the second diode VD2, the first thyristor VT3 and the second thyristor VT4 form a single-phase half-controlled rectifier; DC bus voltage sensor The collected voltage is the DC bus voltage U dc , ground detection circuit voltage sensor The collected voltage is the ground detection voltage U jd ; The working principle of the train onboard power supply system is as follows: The train onboard power supply system obtains single-phase 25kV AC power from the external power supply network and converts it into 600V DC power to power the electrical loads in the train compartment. The train onboard power supply system consists of a traction transformer, a single-phase half-controlled rectifier, an intermediate DC circuit and a ground detection circuit. The single-phase half-controlled rectifier can convert AC power into DC power. In order to obtain a stable 600V DC voltage, the control system collects the secondary side AC voltage U s (sampled by synchronous transformer TV), secondary side AC current I s (measured by current transformer TA) and DC bus voltage U dc (measured by DC bus voltage sensor VS1) for closed loop control. At the same time, the parallel reactance L and supporting capacitor C are used for filtering. The first grounding resistor at each location , Second grounding resistance , the third grounding resistance , the fourth grounding resistor and the fifth grounding resistor The range is "megohm level". Due to the harsh working environment, the resistance value of the grounding resistor at each location will always deteriorate. In engineering practice, when the resistance value is greater than 20000Ω, the grounding is considered normal.

[0027] S1 obtains the DC bus voltage U of the train onboard power supply system in a normal state dc With ground detection voltage U jd The calculation expression of the relationship is as follows:

[0028] S1 also includes obtaining the fault grounding resistance and DC bus voltage U under different grounding fault positions dc , ground detection voltage U jd The relationship includes: See also Figure 2 At point ①, when the first grounding resistance When the resistance value decreases due to degradation, the DC link positive pole grounding fault occurs in the train power supply system, where U jd Yes with U dc As the ground insulation damage increases, when the first grounding resistance R g1 When it gradually decreases to 0, U jd Will be from 0.5U dc Gradually decreases to 0; according to the circuit principle, R g1 with U dc and U jd The relationship between them is expressed as:

[0029] See also Figure 2 At point ②, when the second grounding resistor R g2 When the resistance value decreases due to degradation, the DC link negative pole grounding fault occurs in the train power supply system, where U jd The same is true with U dc As the degree of ground insulation damage increases, when the second grounding resistance R g2 When it gradually decreases to 0, U jd Will be from 0.5U dc Gradually increase to U dc ; According to the circuit principle, R g2 with U dc and U jd The relationship between them is expressed as:

[0030] See also Figure 2 At ③, when R g3 When the resistance value decreases due to degradation, a ground fault occurs at the front end of the reactor in the train power supply system. Us is the secondary side AC voltage of the traction transformer, I s is the secondary side AC current of the traction transformer, U ab It is related to the secondary side AC voltage Us and the secondary side AC current I s Strongly correlated first AC component, when the grounding resistance R g3 When U is a fixed value, only jd It is also the exchange quantity that can be connected with U ab phase offset; according to the circuit principle, R g3 with U dc and U jd The relationship between them is expressed as:

[0031] The first AC component U ab The calculation expression is:

[0032] See also Figure 2 At ④, when R g4 When the resistance value decreases due to degradation, the train power supply system will have an RC input positive grounding fault, where U cb1 is the secondary side AC voltage U s and the secondary side AC current I s Strongly correlated second AC component; when R g4 When U is a fixed value, only jd It is also the exchange quantity that can be connected with U cb1 phase offset; according to the circuit principle, R g4 with U dc and U jd The relationship between them is expressed as:

[0033] The second AC component U cb1 The calculation expression is:

[0034] See also Figure 2 At ⑤, when R g5 When the resistance value decreases due to degradation, the train power supply system will have an RC input negative grounding fault, where U cb2 is the secondary side AC voltage U s and the secondary side AC current I s Strongly correlated third AC component, when R g5When U is a fixed value, only jd It is also the exchange quantity that can be connected with U cb2 phase offset; according to the circuit principle, R g5 with U dc and U jd The relationship between can be expressed as: According to the circuit principle, U dc and U jd The relationship between can be expressed as:

[0035] The third AC component The calculation expression is:

[0036] S2. Under different ground fault positions, based on the relationship between the DC bus voltage and the ground detection voltage, construct characteristic variables, and construct characteristic vector sets and classification enabling mechanisms according to the characteristic variables; In S2, the ground fault mechanism is analyzed at different ground fault locations, and the characteristic variable is constructed based on the relationship between the DC bus voltage and the ground detection voltage. The calculation expression is as follows:

[0037] According to the characteristic variables Construct the feature vector set The calculation process includes: For the normal operation state, the DC link positive pole grounding fault and the DC link negative pole grounding fault type, the mean E(k) and variance V(k) of a sliding window F at the kth moment are used to distinguish them. The calculation expressions of the mean E(k) and variance V(k) are as follows:

[0038] For the three types of reactor front end grounding fault, RC input positive pole grounding fault and RC input negative pole grounding fault, the U jd Contains DC and AC components, through the mean E and variance V of F, combined with the harmonic amplitude H m And total harmonic distortion THD. jd Perform fast Fourier transform to extract each harmonic component and calculate the mth harmonic amplitude H m And total harmonic distortion THD are as follows:

[0039] According to the characteristic variables Construct the feature vector set , FV is an m+3-dimensional feature vector set composed of time domain and frequency domain features. The mathematical expression is as follows:

[0040] in, represents the mean value of the characteristic variable F within a sliding window period at the kth moment, N represents the total number of sampling points in a sliding window, , It represents the variance of the feature variable F within a sliding window period at the kth moment, where k is the index at a certain moment. Indicates the ground detection voltage U jd The complex spectrum value at frequency f contains the amplitude and phase information of the frequency component; Indicates the ground detection voltage U jd The value of the i-th sampling point represents the sampling data of the original sampling signal in the sliding window; represents the harmonic amplitude, Indicates the ground detection voltage U jd The complex spectrum value at m times the frequency f, Indicates total harmonic distortion.

[0041] S3. Based on the classification enabling mechanism, determine whether the train onboard power supply system is in a fault state. If so, execute S4; if not, it is in a normal state; In S3, the classification enabling mechanism includes: The characteristic variable F can be used to detect whether a fault occurs in the train's onboard power supply system, so as to determine whether to enable the subsequent ground fault classification model. According to formula (7), the value of F is 0 under normal circumstances. However, due to the presence of field noise and measurement errors, the characteristic variable F may fluctuate around 0, affecting fault detection. In order to eliminate the interference of noise on fault detection, based on the characteristic variable F, the CUSUM control chart method is used to determine whether a fault occurs, and the cumulative deviation S of the characteristic variable F is calculated. k The following formula:

[0042] Among them, E represents the mean of the feature variable F, and V represents the variance of the feature variable F; it is used as a smoothing constant; Determine the cumulative deviation S k Is it greater than the fault threshold h? If so, the train's onboard power supply system is in a faulty state, and the subsequent grounding fault classification model is started; if not, the train's onboard power supply system is in a normal state.

[0043] S4. Using the feature vector set, construct a total data set of feature vectors under different ground fault conditions; In S4, the train onboard power supply system fault simulation is performed through MATLAB-Simulink, and the obtained data of the normal state and the fault of the system are used as the original data set, which includes the normal state and five types of fault states, a total of six groups of data; the feature vectors FV1 to FV5 of the five types of fault states in each sliding window cycle are extracted from the feature vector set FV, and the calculation expression of the total data set D of the feature vectors under different grounding fault states is constructed as follows:

[0044] in, The characteristic vector set representing the DC link positive pole grounding fault within the sliding window period, represents the mean value of the characteristic variable F of the DC link positive pole grounding fault within the sliding window period, represents the variance of the characteristic variable F of the DC link positive pole grounding fault within the sliding window period, Indicates the harmonic amplitude of the DC link positive pole grounding fault, Indicates the total harmonic distortion of the DC link positive pole grounding fault; The characteristic vector set representing the negative pole grounding fault of the DC link within the sliding window period, represents the mean value of the characteristic variable F of the negative pole grounding fault in the DC link within the sliding window period, represents the variance of the characteristic variable F of the negative pole grounding fault in the DC link within the sliding window period, Indicates the harmonic amplitude of the negative ground fault in the DC link. Indicates the total harmonic distortion of the negative pole grounding fault in the DC link; The characteristic vector set representing the ground fault at the front end of the reactor within the sliding window period, represents the mean value of the characteristic variable F of the reactor front-end ground fault within the sliding window period, The variance of the characteristic variable F of the ground fault at the front end of the reactor within the sliding window period, Indicates the harmonic amplitude of the ground fault at the front end of the reactor. Indicates the total harmonic distortion of the ground fault at the front end of the reactor; The feature vector set representing the RC input positive ground fault within the sliding window period, represents the mean value of the characteristic variable F of the RC input positive pole grounding fault within the sliding window period, represents the variance of the characteristic variable F of the RC input positive pole grounding fault within the sliding window period, Indicates the harmonic amplitude of the RC input positive ground fault, Indicates the total harmonic distortion of the RC input positive pole grounding fault; The set of feature vectors representing the RC input negative ground fault within the sliding window period, represents the mean value of the characteristic variable F of the RC input negative pole grounding fault within the sliding window period, represents the variance of the characteristic variable F of the RC input negative pole grounding fault within the sliding window period, Indicates the harmonic amplitude of the RC input negative ground fault, Represents the total harmonic distortion of the RC input negative pole grounding fault; the total data set D contains five types of fault data feature vector indicators, namely FV1 to FV5. The total data set D describes different types of fault characteristics, provides the signal characteristics of the system under different faults, and provides training samples for the subsequent grounding fault classification model.

[0045] S5. Using the total data set as the input of a preset ground fault classification model, and outputting the ground fault location by the ground fault classification model; S6. Based on the location of the ground fault, select the voltage signal corresponding to the location of the ground fault as the input of the preset real-time prediction model of the fault ground resistance, and output the real-time prediction result of the fault ground resistance; S7. Identify the grounding insulation status of the train's onboard power supply system based on the real-time prediction result of the fault grounding resistance.

[0046] In this embodiment, firstly, by acquiring the relationship between the DC bus voltage and the grounding detection voltage of the train on-board power supply system under normal conditions, a feature vector set and a classification enabling mechanism are constructed in combination with multi-dimensional feature variable extraction under different grounding fault positions, thereby significantly improving the anti-interference ability of the system; then, after confirming the fault of the train on-board power supply system through the classification enabling mechanism, a total data set of feature vectors under different grounding fault states is constructed, and the total data set is used as the input of a preset grounding fault classification model, and the location of the grounding fault is output, thereby realizing data-driven fault classification and effectively improving the real-time diagnosis accuracy of the location of the grounding fault; then, by selecting the voltage signal corresponding to the location of the grounding fault as the input of a preset real-time prediction model of fault grounding resistance, a real-time prediction result of the fault grounding resistance is output, thereby effectively reducing the prediction error of the resistance and ensuring the accuracy of the real-time prediction result of the fault grounding resistance; further, according to the real-time prediction result of the fault grounding resistance, the grounding insulation state of the train on-board power supply system is identified, thereby effectively improving the accuracy of grounding insulation state identification.

[0047] Example 2 See also Figure 1 and Figure 3In this embodiment, the ground fault classification model described in S5 is further described. The ground fault classification model described in S5 is constructed using a convolutional neural network (CNN). The ground fault classification model includes an input layer, a convolution layer, a pooling layer, a first fully connected layer, and an output layer connected in sequence. The characteristic index in the total data set D constructed in step S4 is used as the input of the preset ground fault classification model. The characteristic vector FV of each fault state i Contains m+3 features; the convolution kernel size of the convolution layer is 3×33, the pooling layer is maximum pooling, the pooling window size is 2×22, the number of nodes in the first fully connected layer is 50, the output layer is 6-dimensional, and the activation function of the output layer is softmax. Cross entropy is used as the loss function, Adam optimizer is used, the batch size is 32 during training, and the number of training rounds is 100.

[0048] The training set and the test set of the ground fault classification model come from the total data set D constructed in step S4. The ground fault classification model is trained by back propagation and gradient descent. During the training process of the ground fault classification model, the convolution kernel weights and the fully connected layer weights are optimized by back propagation and gradient descent until the training round reaches a preset threshold or the loss function converges. The training of the ground fault classification model is completed to obtain a trained ground fault classification model. The trained ground fault classification model is used to accurately identify different types of ground faults. After the training is completed, the trained ground fault classification model is verified by using the test set to evaluate the accuracy and robustness of the trained ground fault classification model in predicting the location of the ground fault.

[0049] Example 3 See also Figure 1 and Figure 3 In this embodiment, the ground fault classification model described in S6 is further described. The real-time prediction model for fault ground resistance described in S6 is constructed based on a physical information neural network (PINN). The real-time prediction model for fault ground resistance is used to predict the fault ground resistance R g Real-time prediction; S6 combines the real-time prediction model of fault grounding resistance built based on physical information neural network (PINN) with the voltage signal corresponding to the location of the grounding fault, and can accurately output the real-time prediction results of the fault grounding resistance, thereby measuring the degree of the grounding fault.

[0050] Training a real-time prediction model for fault ground resistance based on physical information neural network (PINN) The ground fault classification model in step S5 is used to identify the location of the ground fault, and based on the location of the ground fault, a voltage signal (such as U dc , U jd , Uab , U cb1 , U cb2 ) is used as the input of the preset real-time prediction model of fault grounding resistance, and the corresponding fault grounding resistance R g As the output of the real-time prediction model of fault ground resistance.

[0051] The real-time prediction model for fault grounding resistance includes a plurality of second fully connected layers, each of which is provided with 64 neurons and outputted using a ReLU activation function; The original data set described in S4 in the above embodiment is divided into a training set and a data set of the real-time prediction model of the fault grounding resistance, and the network weight coefficient is set using an optimization algorithm including but not limited to a back propagation algorithm and a gradient descent method. , minimize the total loss function. Use the training set of the real-time prediction model of fault grounding resistance to train the real-time prediction model of fault grounding resistance until the total loss function converges and the training is completed, so that the real-time prediction model of fault grounding resistance can reasonably learn the physical laws according to different situations and obtain accurate real-time prediction results of fault grounding resistance; input the test set of the real-time prediction model of fault grounding resistance into the trained grounding fault location and recognition model for verification, and obtain the trained grounding fault location and recognition model; The total loss function used in the training process of the real-time prediction model for fault grounding resistance combines data error and physical constraint error. The data error is used to measure the difference between the output of the real-time prediction model for fault grounding resistance and the actual grounding resistance value. The physical constraint error is used to ensure that the output of the real-time prediction model for fault grounding resistance satisfies the physical formula. For the function of physical constraints, the fault grounding resistance under five ground fault locations is calculated. With DC bus voltage U dc , ground detection voltage U jd , five physical constraints are introduced, namely formulas (2), (3), (4), (5), and (6), corresponding to five ground fault conditions respectively; during the training process of the real-time prediction model for fault grounding resistance, for each physical constraint condition, the real-time prediction model for fault grounding resistance calculates the corresponding fault grounding resistance R according to the input voltage signal through the physical formula. g , and compared with the estimated value to obtain the physical constraint error. Therefore, the total loss function used by the real-time prediction model of fault grounding resistance in the training process is as follows:

[0052] in, represents the total loss function, Indicates data error, represents the first physical constraint error, represents the weight coefficient of the first physical constraint error, represents the second physical constraint error, represents the weight coefficient of the second physical constraint error, represents the third physical constraint error, represents the weight coefficient of the third physical constraint error, represents the fourth physical constraint error, represents the weight coefficient of the fourth physical constraint error, represents the fifth physical constraint error, Represents the weight coefficient of the fifth physical constraint error.

[0053] The trained ground fault location recognition model can estimate the fault ground resistance R according to the ground fault location identified by the ground fault classification model in step S5 and the voltage signal corresponding to the ground fault location. g Through offline testing and maintenance experience data, the grounding resistance thresholds of different health status levels (normal, degraded, abnormal, protection) are calculated and compared with the fault grounding resistance R g The real-time prediction results are compared to estimate the health status and formulate corresponding fault handling strategies.

[0054] Here, the present invention proposes grounding insulation state identification based on data-driven algorithm, and constructs grounding fault classification model and fault grounding resistance real-time prediction model through efficient data-driven algorithm, so as to realize accurate identification of grounding insulation state of train onboard power supply system; the present invention generates and extracts multi-state simulation data, uses MATLAB-Simulink simulation technology to generate sample data under normal state and five types of fault state, and combines the sliding window method to extract key feature variables, so as to ensure the comprehensiveness and high quality of training data of grounding fault classification model and fault grounding resistance real-time prediction model; the present invention also establishes grounding fault classification and grounding insulation state identification mechanism, and optimizes algorithm selection and parameter adjustment of grounding fault classification model on the basis of feature vector set constructed based on feature vector, so as to realize high-precision estimation of grounding fault category and state.

[0055] It should be noted that the existing grounding diagnosis of train onboard power supply systems mostly adopts a model-based construction method, and most of them can only diagnose inverter overcurrent of a single fault source type. The construction of the train onboard power supply system is complex and difficult to represent with a single mathematical model. A small interference will have a great impact on the diagnosis result. The existing calculation of the grounding resistance of the train onboard power supply system mostly adopts a calculation method based on the model derivation formula, which is based on a series of simplified assumptions and idealized conditions. When the actual situation is very different from the assumed conditions, the calculation method based on the model derivation formula may not accurately reflect the actual behavior of the system, resulting in a deviation between the calculation result and the actual situation.

[0056] In view of the above shortcomings in the prior art, the purpose of the present invention is to improve the existing train onboard power supply system fault location and ground resistance estimation method by the following means: First, improve the adaptability to complex interference; the present invention proposes a ground fault location framework for a train onboard power supply system based on data-driven, which includes a classification enabling mechanism and a ground fault classification model based on a convolutional neural network. Under normal circumstances, only the classification enabling mechanism is run, and after enabling, the ground fault classification model based on feature vector learning is run, which improves the computational efficiency and the anti-interference ability.

[0057] Second, the real-time prediction model of fault grounding resistance driven by physical information and data is combined to improve the accuracy of resistance estimation; the present invention uses a real-time prediction model of fault grounding resistance built based on physical information neural network (PINN) to estimate the fault grounding resistance. Physical information neural network (PINN) combines physical laws and data-driven learning. By using physical formulas as constraints, the real-time prediction model of fault grounding resistance can ensure that the estimation results conform to physical laws while learning data patterns. This method can more accurately reflect the behavior of the actual system and overcome the limitations of traditional model derivation formula assumptions and idealized conditions.

[0058] Third, it provides more accurate and real-time fault diagnosis results. The present invention can accurately judge the fault type of the train's onboard power supply system through fault classification based on a data-driven grounding fault classification model, and further estimate the fault grounding resistance in combination with a real-time prediction model of the fault grounding resistance. The combination of these technical means improves the accuracy and real-time performance of fault diagnosis, ensuring that the system can respond quickly in actual operation, detect potential faults in a timely manner, and avoid major safety hazards.

[0059] The method for identifying the grounding insulation state of a train onboard power supply system proposed by the present invention also has the following advantages: 1. Based on data-driven grounding insulation status identification, the advantage of the present invention is that the grounding fault of the train onboard power supply system is automatically evaluated by using a data-driven algorithm. By training and learning the system data under different states, the fault can be accurately detected and classified. The system analyzes the operating data of the electric traction system in real time, detects grounding faults in time and provides early warnings, effectively improving the safety and reliability of the system.

[0060] 2. Multi-dimensional feature fusion and data enhancement technology are used; the present invention combines time domain and frequency domain feature extraction technology to extract features with significant discrimination from the original data of the train onboard power supply system, and uses data enhancement technology to enhance the robustness of the data-driven ground fault classification model. By fusing multiple features and constructing a comprehensive feature vector set FV, the accuracy and generalization ability of the ground fault classification model fault detection are improved, enabling it to accurately identify the location of the ground fault in the complex and dynamically changing train onboard circuits.

[0061] Example 4 See also Figure 4 This embodiment proposes a grounding insulation state identification system for a train onboard power supply system. The grounding insulation state identification system implements the grounding insulation state identification method for the train onboard power supply system described in the above embodiment, including: An acquisition module is used to obtain the relationship between the DC bus voltage and the ground detection voltage under the normal state of the train onboard power supply system; A first construction module is used to construct characteristic variables based on the relationship between the DC bus voltage and the ground detection voltage at different ground fault locations, and to respectively construct a characteristic vector set and a classification enabling mechanism according to the characteristic variables; A judgment module, used to judge whether the train onboard power supply system is in a fault state based on the classification enabling mechanism; The second construction module is used to construct a total data set of feature vectors under different ground fault states using the feature vector set when the on-board power supply system of the train is in a fault state; A ground fault classification module, used for taking the total data set as an input of a preset ground fault classification model, and the ground fault classification model outputs a ground fault occurrence location; A fault grounding resistance real-time prediction module is used to select a voltage signal corresponding to the grounding fault location as an input of a preset fault grounding resistance real-time prediction model based on the grounding fault location, and output a real-time prediction result of the fault grounding resistance; The grounding insulation status identification module is used to identify the grounding insulation status of the train onboard power supply system according to the real-time prediction result of the fault grounding resistance.

[0062] In this embodiment, firstly, by acquiring the relationship between the DC bus voltage and the grounding detection voltage of the train on-board power supply system under normal conditions, a feature vector set and a classification enabling mechanism are constructed in combination with multi-dimensional feature variable extraction under different grounding fault positions, thereby significantly improving the anti-interference ability of the system; then, after confirming the fault of the train on-board power supply system through the classification enabling mechanism, a total data set of feature vectors under different grounding fault states is constructed, and the total data set is used as the input of a preset grounding fault classification model, and the location of the grounding fault is output, thereby realizing data-driven fault classification and effectively improving the real-time diagnosis accuracy of the location of the grounding fault; then, by selecting the voltage signal corresponding to the location of the grounding fault as the input of a preset real-time prediction model of fault grounding resistance, a real-time prediction result of the fault grounding resistance is output, thereby effectively reducing the prediction error of the resistance and ensuring the accuracy of the real-time prediction result of the fault grounding resistance; further, according to the real-time prediction result of the fault grounding resistance, the grounding insulation state of the train on-board power supply system is identified, thereby effectively improving the accuracy of grounding insulation state identification.

[0063] Obviously, the above embodiments of the present invention are only examples for clearly illustrating the present invention, and are not intended to limit the implementation methods of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all implementation methods here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for identifying the grounding insulation state of a train onboard power supply system, characterized in that: The method comprises the following steps: S1. Obtain the relationship between the DC bus voltage and the ground detection voltage of the train onboard power supply system under normal conditions; S2. Under different ground fault positions, based on the relationship between the DC bus voltage and the ground detection voltage, construct characteristic variables, and construct characteristic vector sets and classification enabling mechanisms according to the characteristic variables; S3. Based on the classification enabling mechanism, determine whether the train onboard power supply system is in a fault state. If so, execute S4; if not, it is in a normal state; S4. Using the feature vector set, construct a total data set of feature vectors under different ground fault conditions; S5. Using the total data set as the input of a preset ground fault classification model, and outputting the ground fault location by the ground fault classification model; S6. Based on the location of the ground fault, select the voltage signal corresponding to the location of the ground fault as the input of the preset real-time prediction model of the fault ground resistance, and output the real-time prediction result of the fault ground resistance; S7. Identify the grounding insulation status of the train's onboard power supply system based on the real-time prediction result of the fault grounding resistance.

2. The method for identifying the grounding insulation status of a train onboard power supply system according to claim 1, characterized in that: S1 obtains the DC bus voltage U of the train onboard power supply system in a normal state dc With ground detection voltage U jd The calculation expression of the relationship is as follows: S1 also includes obtaining the fault grounding resistance and DC bus voltage U under different grounding fault positions dc , ground detection voltage U jd relationship.

3. The method for identifying the grounding insulation state of a train onboard power supply system according to claim 2, characterized in that: S2: constructing a characteristic variable based on the relationship between the DC bus voltage and the ground detection voltage The calculation expression is as follows: According to the characteristic variables Construct the feature vector set The calculation process includes: in, represents the mean value of the characteristic variable F within a sliding window period at the kth moment, N represents the total number of sampling points in a sliding window, , represents the variance of the feature variable F within a sliding window period at the kth moment, Indicates the ground detection voltage U jd The complex spectrum value at frequency f, Indicates the ground detection voltage U jd The value of the i-th sampling point, represents the harmonic amplitude, Indicates the ground detection voltage U jd The complex spectrum value at m times the frequency f, Indicates total harmonic distortion.

4. The method for identifying the grounding insulation state of a train onboard power supply system according to claim 3, characterized in that: The classification enabling mechanism includes: Based on the characteristic variable F, the cumulative deviation S of the characteristic variable F is calculated k The following formula: Among them, E represents the mean of the feature variable F, and V represents the variance of the feature variable F; Determine the cumulative deviation S k Is it greater than the fault threshold h? If so, the train's onboard power supply system is in a fault state; if not, the train's onboard power supply system is in a normal state.

5. The method for identifying the grounding insulation state of a train onboard power supply system according to claim 4, characterized in that: The calculation expression for constructing the total data set D of the feature vectors under different ground fault states by using the feature vector set is as follows: in, The characteristic vector set representing the DC link positive pole grounding fault within the sliding window period, represents the mean value of the characteristic variable F of the DC link positive pole grounding fault within the sliding window period, represents the variance of the characteristic variable F of the DC link positive pole grounding fault within the sliding window period, Indicates the harmonic amplitude of the DC link positive pole grounding fault, Indicates the total harmonic distortion of the DC link positive pole grounding fault; The characteristic vector set representing the negative pole grounding fault of the DC link within the sliding window period, represents the mean value of the characteristic variable F of the negative pole grounding fault in the DC link within the sliding window period, represents the variance of the characteristic variable F of the negative pole grounding fault in the DC link within the sliding window period, Indicates the harmonic amplitude of the negative ground fault in the DC link. Indicates the total harmonic distortion of the negative pole grounding fault in the DC link; The characteristic vector set representing the ground fault at the front end of the reactor within the sliding window period, represents the mean value of the characteristic variable F of the reactor front-end ground fault within the sliding window period, The variance of the characteristic variable F of the ground fault at the front end of the reactor within the sliding window period, Indicates the harmonic amplitude of the ground fault at the front end of the reactor. Indicates the total harmonic distortion of the ground fault at the front end of the reactor; The feature vector set representing the RC input positive ground fault within the sliding window period, represents the mean value of the characteristic variable F of the RC input positive pole grounding fault within the sliding window period, represents the variance of the characteristic variable F of the RC input positive pole grounding fault within the sliding window period, Indicates the harmonic amplitude of the RC input positive ground fault, Indicates the total harmonic distortion of the RC input positive pole grounding fault; The set of feature vectors representing the RC input negative ground fault within the sliding window period, represents the mean value of the characteristic variable F of the RC input negative pole grounding fault within the sliding window period, represents the variance of the characteristic variable F of the RC input negative pole grounding fault within the sliding window period, Indicates the harmonic amplitude of the RC input negative ground fault, Indicates the total harmonic distortion of the RC input negative pole grounding fault.

6. The method for identifying the grounding insulation state of a train onboard power supply system according to claim 1, characterized in that: The ground fault classification model includes an input layer, a convolution layer, a pooling layer, a first fully connected layer and an output layer connected in sequence, the convolution kernel size of the convolution layer is 3×33, the pooling layer is maximum pooling, the pooling window size is 2×22, the number of nodes in the first fully connected layer is 50, the output layer is 6-dimensional, and the activation function of the output layer is softmax.

7. The method for identifying the grounding insulation state of a train onboard power supply system according to claim 6, characterized in that: The ground fault classification model is trained by using back propagation and gradient descent method until the number of training rounds reaches a preset threshold or the loss function converges, then the training of the ground fault classification model is completed and a trained ground fault classification model is obtained.

8. The method for identifying the grounding insulation state of a train onboard power supply system according to claim 1, characterized in that: The real-time prediction model for fault grounding resistance includes multiple second fully connected layers, each of which is provided with 64 neurons and uses a ReLU activation function for output.

9. The method for identifying the grounding insulation state of a train onboard power supply system according to claim 7, characterized in that: The total loss function used by the real-time prediction model for fault grounding resistance in the training process is as follows: in, represents the total loss function, Indicates data error, represents the first physical constraint error, represents the weight coefficient of the first physical constraint error, represents the second physical constraint error, represents the weight coefficient of the second physical constraint error, represents the third physical constraint error, represents the weight coefficient of the third physical constraint error, represents the fourth physical constraint error, represents the weight coefficient of the fourth physical constraint error, represents the fifth physical constraint error, Represents the weight coefficient of the fifth physical constraint error.

10. A grounding insulation state identification system for a train onboard power supply system, the grounding insulation state identification system is implemented based on the grounding insulation state identification method for a train onboard power supply system according to any one of claims 1 to 9, characterized in that: include: An acquisition module is used to obtain the relationship between the DC bus voltage and the ground detection voltage under the normal state of the train onboard power supply system; A first construction module is used to construct characteristic variables based on the relationship between the DC bus voltage and the ground detection voltage at different ground fault locations, and to respectively construct a characteristic vector set and a classification enabling mechanism according to the characteristic variables; A judgment module, used to judge whether the train onboard power supply system is in a fault state based on the classification enabling mechanism; The second construction module is used to construct a total data set of feature vectors under different ground fault states using the feature vector set when the on-board power supply system of the train is in a fault state; A ground fault classification module, used for taking the total data set as an input of a preset ground fault classification model, and the ground fault classification model outputs a ground fault occurrence location; A fault grounding resistance real-time prediction module is used to select a voltage signal corresponding to the grounding fault location as an input of a preset fault grounding resistance real-time prediction model based on the grounding fault location, and output a real-time prediction result of the fault grounding resistance; The grounding insulation status identification module is used to identify the grounding insulation status of the train onboard power supply system according to the real-time prediction result of the fault grounding resistance.

Images