Artificial intelligence cooperates with data knowledge driven direct current power distribution system fault location method

Abstract

This application belongs to the field of power system relay protection technology and provides an artificial intelligence collaborative data knowledge-driven fault location method for DC distribution systems. The method includes: firstly, constructing a signal injection fault location system at one end of the DC line; after fault isolation and deionization delay, switching the circuit impedance to perform two active injections, simultaneously acquiring two sets of transient current sequences with different attenuation characteristics. Then, using the Prony algorithm, first-order exponential decay fitting is performed on the two sets of sequences to extract two initial attenuation time constants. Simultaneous analytical models of fault distance and time constants are established for single-pole grounding and inter-pole short circuits, respectively. Next, the current sequence and initial time constants are input into a pre-trained intelligent correction model, outputting two corrected time constants. Finally, the corrected values ​​are substituted into the analytical model to calculate the fault distance, achieving accurate location. This application can overcome high impedance and strong noise interference, achieving high-precision single-end fault location in DC distribution systems.

Description

Technical Field

[0001] This application belongs to the field of power system relay protection technology, and in particular relates to a fault location method for DC power distribution systems driven by artificial intelligence collaborative data knowledge. Background Technology

[0002] DC power distribution systems have become an important development direction due to their advantages of flexible control and high efficiency. However, DC power distribution systems have low inertia, and the current rises extremely rapidly during short-circuit faults, resulting in a very limited effective transient location time window. Currently, DC fault location often draws on the traveling wave ranging method, but the short distances and complex topologies of DC power distribution lines make the traditional traveling wave method difficult to meet accuracy requirements. Signal injection-based methods achieve location by actively injecting excitation signals and detecting the oscillation decay response, possessing strong resistance to transition resistance. Currently commonly used signal injection location methods mainly employ the Prony numerical identification algorithm to fit the transient discharge current with first-order exponential decay, extract the time constant, and then input it into analytical formulas to calculate the fault distance.

[0003] However, current signal injection-based localization methods heavily rely on Prony numerical identification to extract time constants. Under conditions of high transition resistance and strong noise interference, this algorithm is prone to significant deviations. Furthermore, the nonlinear fault distance analytical model can drastically amplify even small time constant errors, leading to severe distortion of localization results. Summary of the Invention

[0004] This application provides a fault location method for DC power distribution systems driven by artificial intelligence collaborative data knowledge. It can solve the problems of current signal injection-based location methods that rely heavily on Prony numerical identification to extract time constants. Under high transition resistance and strong noise interference, this algorithm is prone to significant deviations, and the nonlinear fault distance analytical model will drastically amplify small time constant errors, resulting in serious distortion of the location results.

[0005] In a first aspect, embodiments of this application provide an artificial intelligence-assisted data knowledge-driven method for fault location in a DC power distribution system, comprising: S1, constructing a signal injection fault location system at a single end of a DC power distribution line; after a short-circuit fault occurs in the DC power distribution system and the faulty line is isolated, after a preset deionization delay, by switching the equivalent impedance of the signal injection loop, performing two active signal injections to the faulty line, and simultaneously acquiring two sets of transient discharge current sequences with different exponential decay characteristics; S2, based on the two sets of transient discharge current sequences, using the Prony numerical identification algorithm to analyze the two sets of transient discharge current sequences. The transient discharge current sequences are fitted with first-order exponential decay to extract two corresponding initial decay time constants. Based on the equivalent topology of the discharge circuits injected by the two signals, a joint physical analytical mapping model between the fault distance and the two decay time constants is established for both single-pole grounding faults and inter-pole short-circuit faults. S3. The two sets of transient discharge current sequences and the two initial decay time constants are input into the pre-trained intelligent time constant correction model to output two corrected time constants. S4. The two corrected time constants are substituted into the physical analytical mapping model to calculate the fault distance and complete the fault location of the DC distribution system.

[0006] In one possible implementation of the first aspect, S1 above involves constructing a signal injection fault location system at one end of a DC distribution line. When a short-circuit fault occurs in the DC distribution system and the faulty line is isolated, after a preset deionization delay, the equivalent impedance of the signal injection loop is switched, and two active signal injections are performed sequentially onto the faulty line. Simultaneously, two sets of transient discharge current sequences with different exponential decay characteristics are acquired, including:

[0007] S11. A single-end signal injection fault location system is constructed by configuring a DC circuit breaker with an integrated signal injection module and a controllable current-limiting reactor at one end of a DC distribution line; wherein, the signal injection module of the DC circuit breaker consists of a grounding branch, a current-limiting resistor and an auxiliary switch group, and is coupled with the commutation branch of the DC circuit breaker; the controllable current-limiting reactor consists of a current-limiting inductor and a fully controlled power device connected in parallel;

[0008] S12. When a short circuit fault occurs in the DC power distribution system, the main circuit of the control DC circuit breaker is activated to disconnect the faulty line. After waiting for a 200ms deionization delay, the signal injection stage is entered.

[0009] S13. Perform the first active signal injection: Control the fully controlled power device of the controllable current-limiting reactor to conduct to bypass the current-limiting inductor, utilize the residual energy stored in the commutation capacitor inside the DC circuit breaker to discharge to the faulty line, and collect the transient discharge current sequence that follows a first-order exponential decay law in real time to obtain the first set of transient discharge current sequences. ;

[0010] S14. Perform the second active signal injection: Turn off the fully controlled power device of the controllable current-limiting reactor, connect the current-limiting inductor in series with the discharge circuit, change the equivalent impedance of the signal injection circuit, trigger the commutation capacitor to discharge to the faulty line again, and collect the transient discharge current sequence that follows the first-order exponential decay law in real time to obtain the second set of transient discharge current sequences. .

[0011] Optionally, in another possible implementation of the first aspect, S2 above, based on two sets of transient discharge current sequences, performs first-order exponential decay fitting on the two sets of transient discharge current sequences using the Prony numerical identification algorithm to extract two corresponding initial decay time constants. Based on the equivalent topology of the discharge loops from the two signal injections, a simultaneous physical analytical mapping model between the fault distance and the two decay time constants is established for both single-pole grounding faults and inter-pole short-circuit faults, including:

[0012] S21. For the first set of transient discharge current sequences, based on the Prony linear equations, the parameters are fitted by solving the equations using the least squares method, and the initial decay time constant corresponding to the first set of transient discharge current sequences is extracted. ;

[0013] S22. For the second set of transient discharge current sequences, based on the Prony linear equations, the parameters are fitted by solving the equations using the least squares method, and the initial decay time constant corresponding to the second set of transient discharge current sequences is extracted. ;

[0014] S23. For unipolar ground faults, based on the equivalent topology of the discharge loop with two signal injections, establish the mapping equation between the time constant and the system physical parameters:

[0015] (1)

[0016] (2)

[0017] in, The time constant of the first discharge current. The time constant of the second discharge current. For the commutation branch inductor, The equivalent inductance of the faulty line. For current limiting resistor, The equivalent resistance of the faulty line. For fault transition resistance, It is a current-limiting inductor;

[0018] By combining formulas (1) and (2) and eliminating the unknowns, the fault distance can be derived. and fault transition resistance The physical mapping equations are as follows:

[0019] (3)

[0020] (4)

[0021] in, and These represent the distributed resistance and inductance per unit length of the line, respectively.

[0022] S24. For inter-electrode short-circuit faults, based on the equivalent topology of the discharge loop with two signal injections, a mapping equation between the time constant and the system physical parameters is established:

[0023] (5)

[0024] (6)

[0025] By combining formulas (5) and (6) and eliminating the unknowns, the fault distance can be derived. and fault transition resistance The physical mapping equations are as follows:

[0026] (7)

[0027] (8)

[0028] Optionally, in another possible implementation of the first aspect, the intelligent time constant correction model includes a dual-channel gated recurrent unit (GRU) based on parameter sharing, a feature fusion layer, and a multilayer perceptron (MLP). In step S3, two sets of transient discharge current sequences and two initial decay time constants are input into the pre-trained intelligent time constant correction model, outputting two corrected time constants, including:

[0029] S31. After normalizing the two sets of transient discharge current sequences, they are input into two parallel channels in the GRU. The update and reset gates of the GRU are used to coordinate the updating and propagation of the hidden state, and the deep temporal feature vectors are extracted respectively. and The hidden state update process of GRU satisfies the following formula:

[0030] Update Gate: (9)

[0031] Reset Door: (10)

[0032] Candidate hidden state: (11)

[0033] Current hidden state: (12)

[0034] in, Enter the current time. and For activation function, and This is the weight matrix. This represents element-wise multiplication. To update the door, To reset the door.

[0035] S32. Transfer deep temporal feature vectors and With two initial decay time constants and The features are concatenated to form a fused feature vector. ;

[0036] S33. Input the fused feature vector into the MLP, and perform high-dimensional nonlinear adaptive mapping through multiple fully connected layers to compensate for the identification error of the initial decay time constant, finally outputting two corrected time constants. and .

[0037] Optionally, in another possible implementation of the first aspect, the training process of the pre-trained intelligent time constant correction model is as follows:

[0038] Constructing a time constant loss function With distance loss function The details are as follows:

[0039] (13)

[0040] (14)

[0041] in, For the sample size, and These are the true values ​​of the time constant and fault distance for the corresponding samples, respectively. The calculated value of the time constant for the corresponding sample. The calculated fault distance value for the corresponding sample;

[0042] Based on the time constant loss function With distance loss function The comprehensive joint loss function is constructed as follows:

[0043] (15)

[0044] in, and The weighting coefficient for the time constant error. and This is the weighting coefficient for the distance error. The rounds in which the training phase transitions. This represents the maximum number of training iterations.

[0045] In the initial stage of training, that is At that time, set The model is forced to prioritize learning the physical mapping relationship between transient current waveforms and decay time constants;

[0046] In the later stages of training, that is When switching weighting coefficients, use... By shifting the optimization focus to the fault distance error term, the joint optimization training of the model can be completed.

[0047] Optionally, in another possible implementation of the first aspect, S4 above, substituting the two corrected time constants into the physical analytical mapping model to calculate the fault distance, includes:

[0048] S41. Adjust the two time constants and Directly replace the initial decay time constant and ;

[0049] S42, Correct the time constant and As an input variable, it is substituted into the physical analytical mapping model, and the fault distance is obtained through analytical calculation.

[0050] Beneficial Effects: The AI-driven data knowledge-based method for precise fault location in DC distribution systems proposed in this application has significant beneficial effects. First, by constructing an intelligent correction model based on a dual-channel gated cyclic unit and a multilayer perceptron, it can deeply mine long-term features in transient discharge current sequences, effectively compensating for the systematic biases generated by the traditional Prony numerical algorithm when extracting time constants under high transition resistance and strong noise interference environments, significantly improving parameter identification accuracy under complex and harsh operating conditions. Second, the pioneering multi-objective phased collaborative training strategy, through dynamic weighted joint optimization of time constant loss and fault distance loss, fundamentally suppresses the inherent error amplification effect in nonlinear physical analytical models, avoiding drastic jumps and distortions in the location results. Third, this method only requires two active signal injections at a single end of the distribution network to coordinate the control of circuit breakers and current-limiting reactors, achieving high-precision location, completely eliminating the dependence on cross-end high-speed communication and microsecond-level time synchronization equipment, significantly reducing the difficulty and cost of engineering implementation. Fourth, this method combines the interpretability of physical mechanisms with the data-driven advantages of deep learning. While maintaining the traditional analytical localization framework, it achieves adaptive anti-interference capability, providing a highly reliable and economical fault protection solution for DC power distribution systems. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart illustrating an artificial intelligence collaborative data knowledge-driven fault location method for DC power distribution systems, provided in one embodiment of this application.

[0053] Figure 2 This is an embodiment of a signal injection type DC circuit breaker provided in this application;

[0054] Figure 3 This is a circuit topology diagram of a single-ended signal injection fault location system provided in an embodiment of this application;

[0055] Figure 4 This is a schematic diagram of a single-pole grounding fault signal injection circuit provided in an embodiment of this application;

[0056] Figure 5 This is a schematic diagram of an inter-pole short-circuit fault signal injection circuit provided in an embodiment of this application;

[0057] Figure 6This is a schematic diagram of the structure of a time constant intelligent correction model provided in an embodiment of this application. Detailed Implementation

[0058] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0059] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0060] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0061] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0062] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0063] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0064] The following is a detailed description of an artificial intelligence collaborative data knowledge-driven fault location method for DC power distribution systems provided in this application, with reference to the accompanying drawings.

[0065] Figure 1 The illustration shows a flowchart of a fault location method for a DC power distribution system driven by artificial intelligence collaborative data knowledge, provided in an embodiment of this application.

[0066] like Figure 1 As shown, this AI-driven, data-knowledge-based fault location method for DC power distribution systems includes the following steps:

[0067] S1. Construct a signal injection fault location system at one end of the DC distribution line. When a short circuit fault occurs in the DC distribution system and the fault line is isolated, after a preset deionization delay, the equivalent impedance of the signal injection circuit is switched, and two active signal injections are performed on the fault line in succession. Two sets of transient discharge current sequences with different exponential decay characteristics are collected simultaneously.

[0068] For example, the preset deionization delay can be 200ms.

[0069] Furthermore, in this embodiment of the application, step S1 includes:

[0070] S11. A single-end signal injection fault location system is constructed by configuring a DC circuit breaker with an integrated signal injection module and a controllable current-limiting reactor at one end of the DC distribution line; wherein, for example... Figure 2 As shown, the signal injection module of the DC circuit breaker consists of a grounding branch, a current-limiting resistor, and an auxiliary switch group, and is coupled to the commutation branch of the DC circuit breaker; the controllable current-limiting reactor consists of a current-limiting inductor and a fully controlled power device connected in parallel.

[0071] S12. When a short circuit fault occurs in the DC power distribution system, the main circuit of the control DC circuit breaker is activated to disconnect the faulty line. After waiting for a 200ms deionization delay, the signal injection stage is entered.

[0072] S13. Perform the first active signal injection: Control the fully controlled power device of the controllable current-limiting reactor to conduct to bypass the current-limiting inductor, utilize the residual energy stored in the commutation capacitor inside the DC circuit breaker to discharge to the faulty line, and collect the transient discharge current sequence that follows a first-order exponential decay law in real time to obtain the first set of transient discharge current sequences. ;

[0073] S14. Perform the second active signal injection: Turn off the fully controlled power device of the controllable current-limiting reactor, connect the current-limiting inductor in series with the discharge circuit, change the equivalent impedance of the signal injection circuit, trigger the commutation capacitor to discharge to the faulty line again, and collect the transient discharge current sequence that follows the first-order exponential decay law in real time to obtain the second set of transient discharge current sequences. .

[0074] In this embodiment of the application, in order to construct transient echoes with different attenuation characteristics, the application implements two active signal injections under different loop impedances.

[0075] In one embodiment, such as Figure 3 The circuit topology diagram of the single-ended signal injection fault location system shown is mainly composed of a DC circuit breaker with signal injection function and a line-end controllable current-limiting reactor. During signal injection, the fault current flows along the loops corresponding to different fault types. The blue dashed line in the diagram represents the current path for a single-pole ground fault, and the red dashed line represents the current path for an inter-pole short-circuit fault. Due to the differences in the loop structures of the two types of faults, their transient response characteristics also differ, thus providing a basis for fault location. This scheme only requires configuring the above-mentioned device at one end of the line, without relying on measurement information from the other end of the line, to achieve effective fault location identification and reduces the requirements for two-end communication and high-precision time synchronization.

[0076] S2. Based on two sets of transient discharge current sequences, the Prony numerical identification algorithm is used to fit the two sets of transient discharge current sequences with first-order exponential decay, and two corresponding initial decay time constants are extracted. Based on the equivalent topology of the discharge loop of the two signal injections, a simultaneous physical analytical mapping model between the fault distance and the two decay time constants is established for single-pole grounding fault and inter-pole short-circuit fault, respectively.

[0077] Furthermore, in this embodiment of the application, step S2 includes:

[0078] S21. For the first set of transient discharge current sequences, based on the Prony linear equations, the parameters are fitted by solving the equations using the least squares method, and the initial decay time constant corresponding to the first set of transient discharge current sequences is extracted. ;

[0079] S22. For the second set of transient discharge current sequences, based on the Prony linear equations, the parameters are fitted by solving the equations using the least squares method, and the initial decay time constant corresponding to the second set of transient discharge current sequences is extracted. ;

[0080] S23. For single-pole grounding faults (such as...) Figure 4As shown), based on the equivalent topology of the discharge loop with two signal injections, a mapping equation between the time constant and the system physical parameters is established:

[0081] (1)

[0082] (2)

[0083] in, The time constant of the first discharge current. The time constant of the second discharge current. For the commutation branch inductor, The equivalent inductance of the faulty line. For current limiting resistor, The equivalent resistance of the faulty line. For fault transition resistance, It is a current-limiting inductor;

[0084] By combining formulas (1) and (2) and eliminating the unknowns, the fault distance can be derived. and fault transition resistance The physical mapping equations are as follows:

[0085] (3)

[0086] (4)

[0087] in, and These represent the distributed resistance and inductance per unit length of the line, respectively.

[0088] S24. For inter-pole short circuit faults (such as...) Figure 5 As shown), based on the equivalent topology of the discharge loop with two signal injections, a mapping equation between the time constant and the system physical parameters is established:

[0089] (5)

[0090] (6)

[0091] By combining formulas (5) and (6) and eliminating the unknowns, the fault distance can be derived. and fault transition resistance The physical mapping equations are as follows:

[0092] (7)

[0093] (8)

[0094] S3. Input the two sets of transient discharge current sequences and the two initial decay time constants into the pre-trained intelligent time constant correction model, and output two corrected time constants;

[0095] Furthermore, in the embodiments of this application, such as Figure 6 As shown, the above-mentioned intelligent time constant correction model includes a dual-channel gated recurrent unit (GRU) based on parameter sharing, a feature fusion layer, and a multilayer perceptron (MLP). Step S3 includes:

[0096] S31. After normalizing the two sets of transient discharge current sequences, they are input into two parallel channels in the GRU. The update and reset gates of the GRU are used to coordinate the updating and propagation of the hidden state, and the deep temporal feature vectors are extracted respectively. and The hidden state update process of GRU satisfies the following formula:

[0097] Update Gate: (9)

[0098] Reset Door: (10)

[0099] Candidate hidden state: (11)

[0100] Current hidden state: (12)

[0101] in, Enter the current time. and For activation function, and This is the weight matrix. This represents element-wise multiplication. To update the door, To reset the door.

[0102] S32. Transfer deep temporal feature vectors and With two initial decay time constants and The features are concatenated to form a fused feature vector. ;

[0103] S33. Input the fused feature vector into the MLP, and perform high-dimensional nonlinear adaptive mapping through multiple fully connected layers to compensate for the identification error of the initial decay time constant, finally outputting two corrected time constants. and .

[0104] Furthermore, in this embodiment, the training process of the pre-trained intelligent time constant correction model is as follows:

[0105] Constructing a time constant loss function With distance loss function The details are as follows:

[0106] (13)

[0107] (14)

[0108] in, For the sample size, and These are the true values ​​of the time constant and fault distance for the corresponding samples, respectively. The calculated value of the time constant for the corresponding sample. The calculated fault distance value for the corresponding sample;

[0109] Based on the time constant loss function With distance loss function The comprehensive joint loss function is constructed as follows:

[0110] (15)

[0111] in, and The weighting coefficient for the time constant error. and This is the weighting coefficient for the distance error. The rounds in which the training phase transitions. This represents the maximum number of training iterations.

[0112] In the initial stage of training, that is At that time, set The model is forced to prioritize learning the physical mapping relationship between transient current waveforms and decay time constants;

[0113] In the later stages of training, that is When switching weighting coefficients, use... By shifting the optimization focus to the fault distance error term, the joint optimization training of the model can be completed.

[0114] S4. Substitute the two corrected time constants into the physical analytical mapping model to calculate the fault distance and complete the fault location of the DC power distribution system.

[0115] Furthermore, in this embodiment of the application, step S4 includes:

[0116] S41. Adjust the two time constants and Directly replace the initial decay time constant and ;

[0117] S42, Correct the time constant and As an input variable, it is substituted into the physical analytical mapping model, and the fault distance is obtained through analytical calculation.

[0118] This application provides an AI-driven, data-driven fault location method for DC power distribution systems. First, a signal injection fault location system is constructed at one end of the DC line. After fault isolation and deionization delay, the loop impedance is switched to perform two active injections, simultaneously acquiring two sets of transient current sequences with different attenuation characteristics. Then, the Prony algorithm is used to fit the two sets of sequences with first-order exponential decay, extracting two initial attenuation time constants. Simultaneous analytical models of fault distance and time constants are established for single-pole grounding and inter-pole short circuits, respectively. Next, the current sequence and initial time constants are input into a pre-trained intelligent correction model, which outputs two corrected time constants. Finally, the corrected values ​​are substituted into the analytical model to calculate the fault distance, achieving accurate fault location. This application can overcome high impedance and strong noise interference, achieving high-precision single-end fault location in DC power distribution systems.

[0119] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0120] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A fault location method for DC power distribution systems driven by artificial intelligence collaborative data knowledge, characterized in that, Includes the following steps: S1. Construct a signal injection fault location system at one end of the DC distribution line. When a short circuit fault occurs in the DC distribution system and the fault line is isolated, after a preset deionization delay, the equivalent impedance of the signal injection circuit is switched, and two active signal injections are performed on the fault line in succession. Two sets of transient discharge current sequences with different exponential decay characteristics are collected simultaneously. S2. Based on two sets of transient discharge current sequences, the Prony numerical identification algorithm is used to fit the two sets of transient discharge current sequences with first-order exponential decay, and two corresponding initial decay time constants are extracted. Based on the equivalent topology of the discharge loop of the two signal injections, a simultaneous physical analytical mapping model between the fault distance and the two decay time constants is established for single-pole grounding fault and inter-pole short-circuit fault, respectively. S3. Input the two sets of transient discharge current sequences and the two initial decay time constants into the pre-trained intelligent time constant correction model, and output two corrected time constants; S4. Substitute the two corrected time constants into the physical analytical mapping model to calculate the fault distance and complete the fault location of the DC power distribution system.

2. The method as described in claim 1, characterized in that, S1 involves constructing a signal injection fault location system at one end of a DC distribution line. When a short-circuit fault occurs in the DC distribution system and the faulty line is isolated, after a preset deionization delay, the equivalent impedance of the signal injection loop is switched, and two active signal injections are performed sequentially onto the faulty line. Simultaneously, two sets of transient discharge current sequences with different exponential decay characteristics are acquired, including: S11. A single-end signal injection fault location system is constructed by configuring a DC circuit breaker with an integrated signal injection module and a controllable current-limiting reactor at one end of a DC distribution line; wherein, the signal injection module of the DC circuit breaker consists of a grounding branch, a current-limiting resistor and an auxiliary switch group, and is coupled to the commutation branch of the DC circuit breaker; the controllable current-limiting reactor consists of a current-limiting inductor and a fully controlled power device connected in parallel; S12. When a short circuit fault occurs in the DC power distribution system, the main circuit of the control DC circuit breaker is activated to disconnect the faulty line. After waiting for a 200ms deionization delay, the signal injection stage is entered. S13. Perform the first active signal injection: control the fully controlled power device of the controllable current-limiting reactor to conduct to bypass the current-limiting inductor, utilize the residual energy stored in the commutation capacitor inside the DC circuit breaker to discharge to the faulty line, and collect the transient discharge current sequence that follows the first-order exponential decay law in real time to obtain the first set of transient discharge current sequences. S14. Perform the second active signal injection: Control the fully controlled power device of the controllable current-limiting reactor to turn off, insert the current-limiting inductor into the discharge circuit, change the equivalent impedance of the signal injection circuit, trigger the commutation capacitor to discharge to the faulty line again, and collect the transient discharge current sequence that follows the first-order exponential decay law in real time to obtain the second set of transient discharge current sequences.

3. The method as described in claim 2, characterized in that, S2, based on two sets of transient discharge current sequences, uses the Prony numerical identification algorithm to perform first-order exponential decay fitting on the two sets of transient discharge current sequences respectively, extracting two corresponding initial decay time constants. Based on the equivalent topology of the discharge loops from the two signal injections, a simultaneous physical analytical mapping model between the fault distance and the two decay time constants is established for both single-pole grounding faults and inter-pole short-circuit faults, including: S21. For the first set of transient discharge current sequences, based on the Prony linear equations, the parameters are fitted by solving the equations using the least squares method, and the initial decay time constant corresponding to the first set of transient discharge current sequences is extracted. ; S22. For the second set of transient discharge current sequences, based on the Prony linear equations, the parameters are fitted by solving the equations using the least squares method, and the initial decay time constant corresponding to the second set of transient discharge current sequences is extracted. ; S23. For unipolar ground faults, based on the equivalent topology of the discharge loop with two signal injections, establish the mapping equation between the time constant and the system physical parameters: ； （1） ； （2） in, The time constant of the first discharge current. The time constant of the second discharge current. For the commutation branch inductor, The equivalent inductance of the faulty line. For current limiting resistor, The equivalent resistance of the faulty line. For fault transition resistance, It is a current-limiting inductor; By combining formulas (1) and (2) and eliminating the unknowns, the fault distance can be derived. and fault transition resistance The physical mapping equations are as follows: ； （3） ； （4） in, and These represent the distributed resistance and inductance per unit length of the line, respectively. S24. For inter-electrode short-circuit faults, based on the equivalent topology of the discharge loop with two signal injections, a mapping equation between the time constant and the system physical parameters is established: ； （5） ； （6） By combining formulas (5) and (6) and eliminating the unknowns, the fault distance can be derived. and fault transition resistance The physical mapping equations are as follows: ； （7） （8）。 4. The method as described in claim 3, characterized in that, The intelligent time constant correction model includes a dual-channel gated recurrent unit (GRU) based on parameter sharing, a feature fusion layer, and a multilayer perceptron (MLP). In step S3, two sets of transient discharge current sequences and two initial decay time constants are input into the pre-trained intelligent time constant correction model, outputting two corrected time constants, including: S31. After normalizing the two sets of transient discharge current sequences, they are input into two parallel channels in the GRU. The update and reset gates of the GRU are used to coordinate the updating and propagation of the hidden state, and the deep temporal feature vectors are extracted respectively. and The hidden state update process of GRU satisfies the following formula: Update Gate: ; (9) Reset Door: ; (10) Candidate hidden state: ; (11) Current hidden state: ;(12) in, Enter the current time. and For activation function, and This is the weight matrix. This represents element-wise multiplication. To update the door, To reset the door; S32. Transfer deep temporal feature vectors and With two initial decay time constants and The features are concatenated to form a fused feature vector. ; S33. Input the fused feature vector into the MLP, and perform high-dimensional nonlinear adaptive mapping through multiple fully connected layers to compensate for the identification error of the initial decay time constant, finally outputting two corrected time constants. and .

5. The method as described in claim 4, characterized in that, The training process of the pre-trained intelligent time constant correction model is as follows: Constructing a time constant loss function With distance loss function The details are as follows: ； （13） ； （14） in, For the sample size, and These are the true values ​​of the time constant and fault distance for the corresponding samples, respectively. The calculated value of the time constant for the corresponding sample. The calculated fault distance value for the corresponding sample; Based on the time constant loss function With distance loss function The comprehensive joint loss function is constructed as follows: ； （15） in, and The weighting coefficient for the time constant error. and This is the weighting coefficient for the distance error. The rounds in which the training phase transitions. This represents the maximum number of training iterations. In the initial stage of training, that is At that time, set The model is forced to prioritize learning the physical mapping relationship between transient current waveforms and decay time constants; In the later stages of training, that is When switching weighting coefficients, use By shifting the optimization focus to the fault distance error term, the joint optimization training of the model can be completed.

6. The method as described in claim 5, characterized in that, S4 involves substituting the two corrected time constants into the physical analytical mapping model to calculate the fault distance, including: S41. Adjust the two time constants and Directly replace the initial decay time constant and ; S42, Correct the time constant and As an input variable, it is substituted into the physical analytical mapping model, and the fault distance is obtained through analytical calculation.

Images