A bridge arm reactor fault diagnosis method, device and equipment based on PINN-LSTM and a medium

Abstract

The application discloses a bridge arm reactor fault diagnosis method and device based on PINN-LSTM, equipment and medium, belongs to the field of power system equipment state monitoring. The method collects the bridge arm reactor package current voltage time series data through high-precision sensors, combines the geometric parameters of each turn in the package, and forms a fusion feature vector after preprocessing; a fault diagnosis model of fusion physical information neural network and long short-term memory network is constructed, and a supervised learning mechanism is introduced; according to the fault diagnosis model trained, the fault degree of the target bridge arm reactor is evaluated and analyzed, the electrical feature vector and the geometric feature vector are taken as inputs, and the fault diagnosis result and the accurate position of the fault turn in the package are output. The application solves the bottleneck of low fault diagnosis rate and the inability to locate the fault turn of the traditional method, is suitable for online fault monitoring of the bridge arm reactor in the high-voltage direct-current transmission system, and significantly improves the accuracy and reliability of fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of power system equipment condition monitoring technology, and in particular to a method, device, equipment and medium for fault diagnosis of bridge arm reactors based on PINN-LSTM. Background Technology

[0002] Bridge arm reactors are core equipment in high-voltage direct current transmission systems, primarily used to limit commutation current, suppress harmonic components, and ensure stable system operation. Their complex internal structure and long-term operation under high voltage and strong magnetic field conditions make them prone to inter-turn short-circuit faults.

[0003] Currently, common fault diagnosis methods for bridge arm reactors mainly include manual periodic inspections and single-threshold-based diagnostic protection. However, manual inspections have operational limitations and relatively low efficiency; while single-threshold-based diagnostic protection methods lack sensitivity to minor inter-turn short-circuit faults, resulting in high false alarm and false negative rates. Besides manual inspections, the reactor's operating status can be evaluated using neural networks, but single data-driven models (such as traditional LSTM) lack physical constraints, resulting in poor adaptability to different operating conditions and structural equipment, and relying on a large number of fault samples for training. In actual engineering, the scarcity of fault samples leads to insufficient model training. Furthermore, most existing technologies only focus on electrical characteristics such as current and voltage, neglecting the influence of turn-level geometric parameters within the enclosure on the electromagnetic characteristics of the fault, making it difficult to locate the specific faulty turn within the enclosure.

[0004] Therefore, the fault diagnosis method for bridge arm reactors needs further research to improve the accuracy and reliability of fault diagnosis. Summary of the Invention

[0005] To address the technical problems of low fault diagnosis rate and inability to locate faulty turns in traditional methods, this invention proposes a fault diagnosis method, device, equipment, and medium for bridge arm reactors based on PINN-LSTM. The PINN constraint ensures that the diagnosis results conform to physical laws, the LSTM captures the temporal characteristics of electrical features, the influence of the reactor's historical operating state on the current fault level is considered, and the generalization ability of the model is improved through supervised learning, thereby achieving accurate assessment of the operating state of the bridge arm reactor and precise location of faulty turns.

[0006] This invention adopts the following technical solution: a fault diagnosis method for bridge arm reactors based on PINN-LSTM, comprising the following steps:

[0007] Acquire the timing data of the current transformer of the target bridge arm reactor, and extract the timing electrical characteristic quantities based on the timing data;

[0008] Obtain the geometric parameters of each inner turn of the target bridge arm reactor, and preprocess the geometric parameters to extract geometric feature vectors;

[0009] The time-series electrical features and the geometric feature vectors are vectorized and concatenated to construct a fused feature vector;

[0010] Construct a fault diagnosis model that integrates Physical Information Neural Network (PINN) and Long Short-Term Memory Network (LSTM);

[0011] The fused feature vector is used as the model input, and the fault diagnosis model is trained by supervised learning based on the labeled fault sample set.

[0012] The operating status of the target bridge arm reactor is evaluated and analyzed based on the trained fault diagnosis model, and the fault diagnosis results and the precise location of the faulty turn within the enclosure are output.

[0013] Specifically, the timing data of the current transformer measurement of the target bridge arm reactor is obtained. A high-precision Rogowski coil current sensor and a capacitive voltage divider are used to synchronously collect the current and voltage timing data of each package of the bridge arm reactor. The sampling frequency is set to 20kHz to capture high-order harmonic signals.

[0014] Based on the time series data, time series electrical features are extracted. Empirical wavelet transform is used to denoise the current and voltage time series data. The amplitude, phase, effective value, peak value, kurtosis and other time domain features of the 2nd, 3rd, 5th and 7th harmonics are extracted by fast Fourier transform (FFT) to form encapsulated current and voltage feature vectors.

[0015] Furthermore, the geometric parameters of each inner turn of the target bridge arm reactor are obtained, and the geometric parameters are preprocessed to extract geometric feature vectors. The geometric parameters include the encapsulation radius, total number of turns, and turn spacing.

[0016] Furthermore, a fault diagnosis model integrating physical information neural networks and long short-term memory networks is constructed, including:

[0017] Input layer: Receives the spliced ​​encapsulated current and voltage feature vectors and geometric feature vectors, with a total dimension of 2068, of which the current and voltage time-series features account for 2048 dimensions and the geometric features account for 20 dimensions;

[0018] LSTM Feature Extraction Layer: Contains two LSTM network layers, each with 128 neurons, using the ReLU activation function to capture long-term dependencies and fault evolution trends in current and voltage time series data;

[0019] PINN Physical Constraint Layer: Embedded with the physical equations for inter-turn short circuits of bridge arm reactors derived from the laws of electromagnetic induction and Kirchhoff's laws, including the voltage and current coupling equations at the encapsulation ends of the bridge arm reactors, the equations for changes in inductance during inter-turn short circuits, and the equations for conservation of harmonic energy. The residuals of the physical equations are calculated through automatic differentiation.

[0020] Output layer: A fully connected layer is used to output the fault type and the precise location coordinates of the fault turn.

[0021] Furthermore, the time-series electrical features and the geometric feature vectors are vectorized and concatenated to construct a fused feature vector, which is then used as input to the PINN-LSTM fault diagnosis model.

[0022] The fault diagnosis model is trained using supervised learning based on a set of labeled fault samples. The training process is designed to extract temporal features using LSTM, calculate physical residuals by embedding physical constraint equations using PINN, and optimize model parameters by a total loss function that integrates data loss, physical loss, and supervised loss.

[0023] The fault diagnosis model, once trained, is used to assess and analyze the fault severity of the target bridge arm reactor, and outputs the fault diagnosis results and the precise location of the faulty turn within the enclosure.

[0024] Furthermore, the Long Short-Term Memory (LSTM) network uses a special gating mechanism to regulate the information flow, effectively solving the gradient vanishing problem in training long sequence data.

[0025] The intraneural gating network structure of LSTM includes a forget gate, an input gate, and an output gate, and its calculation formula at each time step is specifically expressed as follows:

[0026] ;

[0027] in, Let the input vector be the input vector at the current time. This is the hidden state from the previous moment. This is the unit state; The Gate of Oblivion determines which historical information to discard; As an input gate, it determines which new information to update to the cell state; This is the output gate, which determines the output at the current moment. It is the Sigmoid activation function. and These are the weight matrix and the bias term, respectively.

[0028] Furthermore, the Physical Information Neural Network (PINN) guides network parameter optimization through physical constraint equations, thereby improving the model's interpretability and generalization ability at the mechanistic level. The specific physical guidance mechanism is as follows:

[0029] Physical equations based on the laws of electromagnetic induction and Kirchhoff's laws are constructed, including the voltage-current coupling equation at the end of the bridge arm reactor enclosure, the equation for the change in inductance between turns short circuits, and the equation for the conservation of harmonic energy. The residuals of the physical equations are calculated by automatic differentiation, and the derivatives of the network output with respect to the input variables are calculated using automatic differentiation technology, thereby obtaining the residual values ​​of the physical equations.

[0030] By adding the physical residual as a regularization term to the total loss function, the gradient generated by the physical residual during backpropagation will force the weights of the neural network to converge toward the solution space that conforms to physical laws.

[0031] This process ensures that the diagnostic model not only numerically fits the measured electrical data, but also follows the electromagnetic operation mechanism of the bridge arm reactor in its internal logic. This enables the model to make accurate diagnoses based on physical laws even when fault samples are scarce, significantly enhancing the physical interpretability of the results.

[0032] Furthermore, the voltage-current coupling equation at the encapsulated end of the bridge arm reactor is as follows:

[0033] ;

[0034] in, For the terminal voltage, and For the first Layer encapsulation and the first Encapsulation current of layered encapsulation For the first Layer encapsulation and the first Mutual inductance between layers of encapsulation For the first Equivalent resistance of the encapsulated layer, This is the operating angular frequency of the reactor. For the first Equivalent inductance of layered encapsulation, This represents the number of encapsulation layers.

[0035] Furthermore, the equation describing the change in short-circuit inductance between turns describes the nonlinear relationship between the number of short-circuit turns and the inductance:

[0036] ;

[0037] in, For the first Package No. Inductance value after short circuit of one turn For the first Encapsulation under normal operating conditions self-induction, For the first Total number of turns of the package This represents the number of turns in a short circuit.

[0038] Furthermore, the harmonic energy conservation equation describes the energy balance relationship between the fundamental wave and the 2nd, 3rd, 5th, and 7th harmonics:

[0039] ;

[0040] in, For total power, , The voltage and current amplitudes of the nth harmonic are... The power factor.

[0041] Furthermore, the loss function is:

[0042] ;

[0043] in, , , , The optimal values ​​for these weighting coefficients, determined through cross-validation, are 0.3, 0.4, 0.2, and 0.1. The mean square error between the measured current and voltage characteristics and the model prediction characteristics; This is the sum of squares of the residuals from the physical equations; The cross-entropy loss is calculated between the fault location and severity labels and the model output. This is the L2 regularization term for the network weights, used to prevent overfitting.

[0044] The present invention also provides: a bridge arm reactor fault diagnosis device based on PINN-LSTM, comprising:

[0045] The data acquisition and preprocessing module is used to acquire the time-series data of the bridge arm reactor transformer and the geometric parameters of the inner turn level, and to complete the noise reduction, feature extraction and vectorized splicing processing.

[0046] The model building module is used to build a fault diagnosis model that integrates PINN and LSTM, and defines a total loss function that includes data loss, physical loss, supervision loss and regularization loss.

[0047] The supervised training module is used to optimize model parameters based on a labeled fault sample set, thereby achieving supervised learning training.

[0048] The fault diagnosis and location module is used to evaluate and analyze the operating status of the target bridge arm reactor based on the trained fault diagnosis model, and output the fault diagnosis results and the precise location of the fault turn within the enclosure.

[0049] The present invention also provides: a computer device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the PINN-LSTM-based bridge arm reactor fault diagnosis method as described above.

[0050] The present invention also provides: a computer-readable storage medium storing a computer program, wherein when the device containing the computer-readable storage medium executes the computer program, it implements the PINN-LSTM-based bridge arm reactor fault diagnosis method as described above.

[0051] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0052] 1. The method of this invention integrates PINN and LSTM, embedding physical laws such as the law of electromagnetic induction and Kirchhoff's laws into the model in the form of constraint equations. This effectively makes up for the lack of physical interpretability in traditional neural networks and greatly improves the generalization ability of the model when fault samples are scarce.

[0053] 2. The method of this invention introduces a supervised learning mechanism and turn-level geometric parameters, which can not only diagnose fault types, but also achieve precise fault location from the encapsulation level to the turn level, thus significantly improving maintenance efficiency. Attached Figure Description

[0054] Figure 1 This is a flowchart of the steps in the bridge arm reactor fault diagnosis method of the present invention;

[0055] Figure 2 This is a diagram of the PINN-LSTM neural network architecture provided in an embodiment of the present invention;

[0056] Figure 3 This is a schematic diagram of the LSTM intranet connection structure provided in an embodiment of the present invention;

[0057] Figure 4 The neuronal gating network structure of LSTM provided in the embodiments of the present invention;

[0058] Figure 5 This is a structural block diagram of the bridge arm reactor fault diagnosis device provided in an embodiment of the present invention;

[0059] Figure 6 A structural diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. Furthermore, the step numbers in the embodiments of this invention are only for ease of explanation and do not limit the order of the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0061] Example 1

[0062] A fault diagnosis method for bridge arm reactors based on PINN-LSTM is provided. For details, please refer to [link to relevant documentation]. Figure 1 , Figure 1 The flowchart shown is a step-by-step flowchart of a bridge arm reactor fault diagnosis method based on PINN-LSTM in one embodiment of the present invention, including steps S11 to S16:

[0063] Step S11: Obtain the timing data of the current transformer measurement of the target bridge arm reactor, and extract the timing electrical characteristic quantities based on the timing data.

[0064] In this embodiment, a high-precision Rogowski coil current sensor and a capacitive voltage divider sensor are used to synchronously acquire the current and voltage timing data of each package of the bridge arm reactor. To capture the 2nd to 7th harmonic signals, the sampling frequency is set to 20kHz, and the acquisition duration covers 10 power frequency cycles to ensure data integrity.

[0065] To address the problems of traditional denoising algorithms, this embodiment employs empirical wavelet transform to denoise current and voltage time-series data. The empirical wavelet function is defined as follows:

[0066] ;

[0067] in, This is the operating angular frequency of the reactor. For the first The middle frequency of each frequency band This is a transition parameter (value 0.2). This is a transition function.

[0068] Fast Fourier Transform (FFT) is performed on the denoised time series data to extract the amplitude, phase, effective value of the fundamental wave, peak value, kurtosis and other time-domain features of the 2nd, 3rd, 5th and 7th harmonics, forming encapsulated current and voltage feature vectors.

[0069] Step S12: Obtain the geometric parameters of each inner turn of the target bridge arm reactor, and preprocess the geometric parameters to extract geometric feature vectors.

[0070] Using a laser rangefinder and a winding parameter tester, the geometric parameters of each turn level within each enclosure are obtained, including the radius r of the k-th enclosure, the total number of turns N, the turn spacing d, the conductor diameter Φ, and the number of winding layers I.

[0071] In this embodiment, the maximum-minimum normalization method is used to normalize the geometric parameters to the [0,1] interval. The normalization formula is as follows:

[0072] ;

[0073] Where x is the original data, and These are the minimum and maximum values ​​for that feature dimension, respectively.

[0074] Step S13: The time-series electrical feature quantity and the geometric feature vector are vectorized and concatenated to construct a fused feature vector.

[0075] The extracted time-series electrical features are concatenated with the normalized geometric feature vectors to form a fused feature vector, which is then used as input for subsequent models.

[0076] Step S14: Construct a fault diagnosis model that integrates Physical Information Neural Network (PINN) and Long Short-Term Memory Network (LSTM).

[0077] In this embodiment, the PINN-LSTM neural network architecture, as follows: Figure 2 As shown, the fault diagnosis model includes an input layer, an LSTM temporal feature extraction layer, a PINN physical constraint layer, and an output layer.

[0078] The LSTM temporal feature extraction layer adopts a stacked 2-layer LSTM network structure, with 128 neurons in each layer.

[0079] Specifically, the LSTM network connection structure is as follows: Figure 3 As shown, a special gating mechanism is used to regulate the information flow, effectively solving the gradient vanishing problem in long sequence data training.

[0080] The intraneural gating network structure of LSTM is as follows: Figure 4 As shown, each neuron contains a three-gate structure consisting of a forget gate, an input gate, and an output gate, which effectively balances the influence of historical data and current data on the network's judgment.

[0081] The output of the first layer is processed by a Dropout layer with a Dropout rate of 0.2 before being input into the second layer to avoid overfitting and ensure stable extraction of time-series evolution features. The ReLU activation function is chosen to effectively uncover subtle patterns in the time-series data, such as gradual changes in harmonic amplitudes and the temporal correlation of current and voltage fluctuations.

[0082] The advantage of the LSTM temporal feature extraction layer lies in its ability to process long time-series data, accurately capture the dynamic process of fault development, and provide temporal basis for fault severity classification.

[0083] The PINN layer compensates for the "black box" defect of data-driven models, improves the physical interpretability of the model, and makes the model prediction results conform to objective physical laws such as electromagnetic induction and energy conservation, thus avoiding diagnostic results that contradict common sense.

[0084] Adding a PINN layer can significantly improve learning efficiency and prediction accuracy in situations with scarce sample data. When fault samples are scarce, traditional data-driven models struggle to learn effective features due to a lack of sufficient samples. However, PINN can compensate for the lack of samples by using physical equations to deduce parameter changes under fault conditions using known physical laws, guiding the model to converge quickly to the optimal solution. This reduces the reliance on a large number of labeled samples and ensures high-precision diagnosis even in scenarios with scarce samples.

[0085] The output layer is a fully connected layer, responsible for outputting the fault type, fault severity, and fault location coordinates.

[0086] Step S15: Supervised learning training is performed on the fault diagnosis model based on the labeled fault sample set.

[0087] Construct a supervised learning sample set, including normal operating conditions and fault operating conditions of short circuits at different turns (e.g., turns 1-20) in each package, and label the fault type, severity and fault turn location.

[0088] Define the total loss function It integrates data loss, physical loss, supervision loss, and regularization loss:

[0089] ;

[0090] Where α, β, γ, and λ are weighting coefficients, and their optimal values ​​were determined to be 0.3, 0.4, 0.2, and 0.1, respectively, through cross-validation. The mean square error between the measured current and voltage characteristics and the model prediction characteristics; This is the sum of squares of the residuals from the physical equations; The cross-entropy loss is calculated between the fault location and severity labels and the model output. This is the L2 regularization term for the network weights, used to prevent overfitting.

[0091] Furthermore, the Adam optimizer is used, with an initial learning rate and batch size set, and the model parameters are iteratively updated using the gradient descent algorithm until the loss function converges.

[0092] Step S16: Evaluate and analyze the operating status of the target bridge arm reactor based on the trained fault diagnosis model, and output the fault diagnosis results and the precise location of the fault turn within the enclosure.

[0093] The fused feature vector to be diagnosed is input into the trained model. The model output layer is a fully connected layer, which outputs the fault type (normal / inter-turn short circuit), fault severity, and precise location coordinates of the fault turn (such as the package number and inter-turn offset), thereby guiding accurate maintenance.

[0094] Example 2

[0095] A fault diagnosis device for a bridge arm reactor based on PINN-LSTM is provided for performing the fault diagnosis method described above, such as... Figure 5 As shown, the device includes:

[0096] The data acquisition and preprocessing module is used to acquire the time-series data of the bridge arm reactor transformer and the geometric parameters of the inner turn level, and to complete the noise reduction, feature extraction and vectorized splicing processing.

[0097] The model building module is used to build a fault diagnosis model that integrates PINN and LSTM, and defines a total loss function that includes data loss, physical loss, supervision loss and regularization loss.

[0098] The supervised training module is used to optimize model parameters based on a labeled fault sample set, thereby achieving supervised learning training.

[0099] The fault diagnosis and location module is used to evaluate and analyze the operating status of the target bridge arm reactor based on the trained fault diagnosis model, and output the fault diagnosis results and the precise location of the fault turn within the enclosure.

[0100] The technical features and effects of the device proposed in the embodiments of the present invention are the same as those of the method proposed in the embodiments of the present invention, and will not be repeated here. Each module in the above-described device can be implemented entirely or partially through software, hardware, or a combination thereof.

[0101] Example 3

[0102] A computer-readable storage medium is provided, the computer-readable storage medium including a stored computer program; wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the PINN-LSTM-based bridge arm reactor fault diagnosis method as described above.

[0103] Example 4

[0104] A computer device is provided. Figure 6 This is a structural block diagram of a preferred embodiment of a computer device provided by the present invention. The computer device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the PINN-LSTM-based bridge arm reactor fault diagnosis method as described above.

[0105] In summary, the PINN-LSTM-based bridge arm reactor fault diagnosis method, device, equipment, and medium provided in this invention, by integrating PINN and LSTM, introducing a supervised learning mechanism, and combining electrical and geometric quantities, achieves high-accuracy fault diagnosis of bridge arm reactors and precise location of faulty turns within the enclosure. This solves the technical bottleneck of traditional methods being unable to locate faulty turns and has significant engineering application value.

[0106] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for fault diagnosis of bridge arm reactors based on PINN-LSTM, characterized in that, Includes the following steps: Acquire the timing data of the current transformer of the target bridge arm reactor, and extract the timing electrical characteristic quantities based on the timing data; Obtain the geometric parameters of each inner turn of the target bridge arm reactor, and preprocess the geometric parameters to extract geometric feature vectors; The time-series electrical features and the geometric feature vectors are vectorized and concatenated to construct a fused feature vector; Construct a fault diagnosis model that integrates physical information neural networks and long short-term memory networks; The fused feature vector is used as the model input, and the fault diagnosis model is trained by supervised learning based on the labeled fault sample set. The operating status of the target bridge arm reactor is evaluated and analyzed based on the trained fault diagnosis model, and the fault diagnosis results and the precise location of the faulty turn within the enclosure are output.

2. The method for fault diagnosis of bridge arm reactors based on PINN-LSTM according to claim 1, characterized in that, The acquisition of the timing data measured by the current transformer of the target bridge arm reactor includes: High-precision Rogowski coil current sensors and capacitive voltage divider sensors are used to synchronously collect current and voltage timing data of each package of the bridge arm reactor. The sampling frequency is set to 20kHz to capture high-order harmonic signals.

3. The method for fault diagnosis of bridge arm reactors based on PINN-LSTM according to claim 1, characterized in that, The step of extracting time-series electrical features based on the time-series data includes: Empirical wavelet transform is used to denoise the current and voltage time series data; the amplitude, phase, effective value, peak value, kurtosis and other time-domain features of the 2nd, 3rd, 5th and 7th harmonics are extracted by fast Fourier transform to form encapsulated current and voltage feature vectors.

4. The method for fault diagnosis of bridge arm reactors based on PINN-LSTM according to claim 1, characterized in that, The construction of the fault diagnosis model integrating physical information neural networks and long short-term memory networks includes: Input layer: Receives the spliced ​​encapsulated current and voltage feature vectors and geometric feature vectors, with a total dimension of 2068, of which the current and voltage time-series features account for 2048 dimensions and the geometric features account for 20 dimensions; LSTM Feature Extraction Layer: Contains two LSTM network layers, each with 128 neurons, using the ReLU activation function to capture long-term dependencies and fault evolution trends in current and voltage time series data; PINN Physical Constraint Layer: Embedded with the physical equations for inter-turn short circuits of bridge arm reactors derived from the laws of electromagnetic induction and Kirchhoff's laws, including the voltage and current coupling equations at the encapsulation ends of the bridge arm reactors, the equations for changes in inductance during inter-turn short circuits, and the equations for conservation of harmonic energy. The residuals of the physical equations are calculated through automatic differentiation. Output layer: A fully connected layer is used to output the fault type and the precise location coordinates of the fault turn.

5. The method for fault diagnosis of bridge arm reactors based on PINN-LSTM according to claim 1, characterized in that, The fault diagnosis model is trained using supervised learning based on a labeled fault sample set. The total loss function used in the training process is: ； in, , , , These are the weighting coefficients; The mean square error between the measured current and voltage characteristics and the model prediction characteristics; This is the sum of squares of the residuals from the physical equations; The cross-entropy loss is calculated between the fault location and severity labels and the model output. This is the L2 regularization term for the network weights.

6. The method for fault diagnosis of bridge arm reactors based on PINN-LSTM according to claim 4, characterized in that, The voltage-current coupling equation at the encapsulated end of the bridge arm reactor is as follows: ； in, For the terminal voltage, and For the first Layer encapsulation and the first Encapsulation current of layered encapsulation For the first Layer encapsulation and the first Mutual inductance between layers of encapsulation For the first Equivalent resistance of the encapsulated layer, This is the operating angular frequency of the reactor. For the first Equivalent inductance of layered encapsulation, This represents the number of encapsulation layers.

7. The method for fault diagnosis of bridge arm reactors based on PINN-LSTM according to claim 4, characterized in that, The equation for the change in inter-turn short-circuit inductance is as follows: ； in, For the first Package No. Inductance value after short circuit of one turn For the first Encapsulation under normal operating conditions self-induction, For the first Total number of turns of the package This represents the number of turns in a short circuit.

8. The method for fault diagnosis of bridge arm reactors based on PINN-LSTM according to claim 4, characterized in that, The harmonic energy conservation equation is as follows: ； in, For total power, , The voltage and current amplitudes of the nth harmonic are... The power factor.

9. A bridge arm reactor fault diagnosis device based on PINN-LSTM, used to implement the method of any of claims 1 to 8, characterized in that, include: The data acquisition and preprocessing module is used to acquire the time-series data of the bridge arm reactor transformer and the geometric parameters of the inner turn level, and to complete the noise reduction, feature extraction and vectorized splicing processing. The model building module is used to build a fault diagnosis model that integrates physical information neural networks and long short-term memory networks, and defines a total loss function that includes data loss, physical loss, supervision loss and regularization loss. The supervised training module is used to optimize model parameters based on a labeled fault sample set using the gradient descent algorithm, thereby achieving supervised learning training. The fault diagnosis and location module is used to evaluate and analyze the operating status of the target bridge arm reactor based on the trained fault diagnosis model, and output the fault diagnosis results and the precise location of the fault turn within the enclosure.

10. An electronic device, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the PINN-LSTM-based bridge arm reactor fault diagnosis method as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein when the device containing the computer-readable storage medium executes the computer program, it implements the PINN-LSTM-based bridge arm reactor fault diagnosis method as described in any one of claims 1 to 8.

Images