Power distribution network fault positioning system and method

Abstract

The invention relates to the technical field of smart power grids, and discloses a power distribution network fault positioning system and method, and the system comprises a causal generation module, an event inversion module, a dynamic pruning module, a causal collaborative verification module, and a meta-cognitive analysis module. The method comprises the following steps: iteratively calling a causal generation module through an event inversion module to reversely solve an optimal event parameter containing a fault position, and eliminating a physically unreasonable search space by utilizing a dynamic pruning module in iteration; then, the causal collaborative verification module verifies the result and calculates a diagnosis confidence coefficient, and when the confidence coefficient is lower than a threshold value, the event model is upgraded to reconstruct a complex fault; further, the system may generate a diagnosability map offline to identify a diagnostic blind spot. Through a generative inversion and model upgrading mechanism, accurate positioning of a fault and accurate reconstruction of a complex accident sequence are realized, a quantitative evaluation capability of a self-diagnosis capability is possessed, and an intelligent level of fault processing is substantially improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart grid, in particular to a power distribution network fault positioning system and method. BACKGROUND

[0002] As the terminal link of the power system, the power distribution network has a complex network structure, widely distributed lines and a variable operating environment, which makes it a part of the power system with a high failure rate. Therefore, after a fault occurs, it is of great significance to quickly and accurately locate the fault point in order to shorten the user outage time, reduce economic losses and improve power supply reliability.

[0003] To achieve rapid fault positioning, the research direction of the prior art is mainly focused on using physical quantity information such as traveling wave and impedance, or using artificial intelligence methods based on data driving. However, these technologies still face common technical bottlenecks in practical application. Based on the traditional physical mechanism model, the positioning accuracy is highly dependent on the accurate grasp of the power grid topology parameters and the fine setting of the protection value, and it is not strong in adaptability in the power distribution network with frequent network structure changes.

[0004] And the existing artificial intelligence method mostly depends on the learning of a large number of fault sample libraries to build a discriminant model from fault features to fault location. This technical paradigm severely limits the completeness of the sample library, and its generalization ability is insufficient for new or rare fault types that the sample library fails to cover, and it is prone to misjudgment. In addition, the existing technology has limited ability to handle complex cascading faults or concurrent faults, and can usually only give a single diagnostic result, but cannot quantitatively evaluate the physical self-consistency and reliability of the result, and also lacks the ability to actively optimize the positioning ability of the system itself under a specific sensor layout, resulting in a lack of ability to deal with complex scenarios and diagnostic blind spots.

[0005] Therefore, the present application proposes a power distribution network fault positioning system and method to solve the deficiencies of the prior art. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a power distribution network fault positioning system and method, which solves the problem that the power distribution network fault positioning technology has difficulty in evaluating the reliability of the diagnostic result, has insufficient generalization ability and lacks active awareness and optimization of its own positioning ability when facing complex fault scenarios.

[0007] To achieve the above purpose, the present application realizes the following technical scheme: a power distribution network fault positioning system, the system comprises:

[0008] a causal generation module, configured to input fault event parameters into a pre-trained Fourier neural operator proxy model to generate a corresponding spatio-temporal transient physical field covering the entire power distribution network in response to the fault event parameters;

[0009] an event inversion module, configured to receive actual fault observation waveforms of the power distribution network, and to iteratively call the causal generation module to reversely solve optimal fault event parameters containing fault location information that can optimally generate the actual fault observation waveforms;

[0010] a dynamic pruning module, configured to analyze the spatio-temporal transient physical field during the iterative solving process of the event inversion module, and to remove unreasonable fault location hypothesis regions from a subsequent search space of the event inversion module according to preset physical rules;

[0011] a causal collaborative verification module, configured to call the causal generation module to generate a final spatio-temporal transient physical field after the event inversion module solves the optimal fault event parameters, and to calculate a final diagnostic confidence associated with the optimal fault location information according to consistency between the final spatio-temporal transient physical field and the actual fault observation waveforms in multiple physical dimensions;

[0012] a meta-cognition analysis module, configured to run in an offline state, to inject virtual fault event parameters, and to call the causal generation module, the event inversion module and the causal collaborative verification module to perform closed-loop simulation to generate a diagnosability map evaluating fault location capability of the current power distribution network sensor layout, and to identify a diagnostic blind area.

[0013] Preferably, the causal generation module is configured to:

[0014] The Fourier neural operator proxy model is optimized by a composite loss function during pre-training to learn a mapping relationship from the fault event parameters to the spatio-temporal transient physical field;

[0015] When defining the composite loss function, a corresponding spatio-temporal transient physical field is generated by calling the Fourier neural operator proxy model based on a set of training fault event parameters;

[0016] Based on the generated spatio-temporal transient physical field, a data fidelity term minimizing the difference between the spatio-temporal transient physical field and the result of the high-precision simulation software, and a physical residual term minimizing the violation degree of the spatio-temporal transient physical field to the basic physical equation of the power distribution network are calculated;

[0017] The composite loss function is obtained based on the sum of the data fidelity term and the physical residual term, and the calculation formula of the composite loss function is:

[0018] ;

[0019] wherein, is a composite loss function; is a data fidelity term; is a physical residual term; and is a weight coefficient balancing the data fidelity term and the physical residual term.

[0020] Preferably, the event inversion module is configured to:

[0021] solve the optimal fault event parameters containing fault location information by constructing and iteratively solving a minimization objective function;

[0022] In constructing the objective function, the causal generation module is invoked to generate corresponding waveforms according to a set of fault event parameters to be evaluated, and the similarity between the generated waveforms and the actual fault observation waveforms is calculated based on the dynamic time warping algorithm, and the calculation formula of the objective function is:

[0023]

[0024] wherein, is an objective function, is a fault event parameter to be solved containing fault location information; is the number of sensors receiving actual fault observation waveforms; is the actual fault observation waveform collected by the th sensor; is the waveform generated by the causal generation module according to the fault event parameter at the th sensor position ; is the distance calculated by the dynamic time warping algorithm; is the weight of the observation data of the th sensor in the objective function;

[0025] The event inversion module is also configured to have a solving unit based on a Bayesian optimization algorithm, which searches in the fault event parameter space to obtain the optimal fault event parameter with the minimum value for the purpose of minimizing the objective function.

[0026] Preferably, the dynamic pruning module is configured to:

[0027] a qualitative feature analysis unit extracts the waveform polarity and transient frequency component of the spatiotemporal transient physical field at the sensor position generated by the causal generation module in the iteration;

[0028] ​a space pruning unit, which removes the fault location hypothesis region corresponding to the spatio-temporal transient physical field from the subsequent search space of the event inversion module when the waveform polarity or the transient frequency component is fundamentally physically inconsistent with the actual fault observation waveform.

[0029] Preferably, the cause-effect collaborative verification module is configured to:

[0030] When performing the consistency judgment in multiple physical dimensions, the multiple physical dimensions include time verification, amplitude verification, and shape verification, wherein,

[0031] The time verification judges whether the disturbance arrival time sequence of the actual fault observation waveform is consistent with the propagation time sequence predicted according to the optimal fault location information.

[0032] The amplitude verification judges whether the relative intensity of the actual fault observation waveform is consistent with the relationship of attenuation with propagation distance predicted according to the optimal fault location.

[0033] The shape verification judges whether the waveform characteristics of the actual fault observation waveform are consistent with the shape predicted according to the optimal fault location after propagation on the corresponding line and passing through refraction and reflection.

[0034] Preferably, the cause-effect collaborative verification module is further configured to:

[0035] When the final diagnostic confidence is higher than or equal to the preset confidence threshold, it is judged that the current optimal fault event parameter is sufficient to reasonably explain the received actual fault observation waveform, the diagnostic process is ended, and the result is output.

[0036] When the final diagnostic confidence is lower than the preset confidence threshold, it is judged that the current event model describing a single fault is insufficient to explain the actual fault observation waveform, and the event inversion module is triggered to re-solve the optimal fault event parameter in the complex event model space describing a sequence fault or a concurrent fault.

[0037] Preferably, the meta-cognition analysis module is configured to:

[0038] a virtual diagnosis circulation unit, which, for each injected virtual fault event parameter, calls the cause-effect generation module to generate a virtual observation waveform, and then inputs the virtual observation waveform and calls the event inversion module and the cause-effect collaborative verification module to perform a complete diagnostic process to obtain an inferred fault event parameter and an expected diagnostic confidence.

[0039] a map generation unit, which inputs all the inference results and expected diagnostic confidences obtained after the virtual diagnosis circulation unit traverses all the virtual fault event parameters, integrates and generates a diagnosability map.

[0040] Preferably, the data structure of the fault event parameters is defined as an extensible event grammar, which is used for:

[0041] a single event model describing a single, independent, single fault event;

[0042] a sequential event model describing a cascading fault scenario triggered by an initial fault after a certain time delay, causing another secondary fault.

[0043] The application also provides a power distribution network fault locating method, comprising the following steps:

[0044] S1, using a pre-trained Fourier neural operator proxy model, generating a time-space transient physical field covering the entire power distribution network corresponding to the input fault event parameters;

[0045] S2, after receiving the actual fault observation waveform of the power distribution network, taking the actual fault observation waveform as the target, iteratively solving the optimal fault event parameters containing fault location information that can optimally generate the actual fault observation waveform by calling step S1;

[0046] S3, in the iterative solving process of step S2, analyzing the time-space transient physical field, and removing unreasonable fault location hypothesis areas from the subsequent search space of step S2 according to the preset physical rules;

[0047] S4, after solving the optimal fault event parameters in step S2, calling step S1 to generate the final time-space transient physical field, and calculating the final diagnostic confidence associated with the optimal fault location information according to the consistency between the final time-space transient physical field and the actual fault observation waveform in multiple physical dimensions;

[0048] S5, in the offline state, by injecting virtual fault event parameters and calling steps S1, S2 and S4 for closed-loop simulation, to generate a diagnosability map evaluating the system fault locating capability under the current power distribution network sensor layout, and identify the diagnostic blind area.

[0049] Preferably, after step S4, it further comprises:

[0050] S4-1: determining whether the final diagnostic confidence calculated in step S4 is lower than the preset confidence threshold;

[0051] S4-2: if the final diagnostic confidence is lower than the preset confidence threshold, upgrading the event model describing the fault from the single event model to the sequential event model, and returning to re-execute step S2.

[0052] The application provides a power distribution network fault locating system and method.

[0053] 1、The application converts the fault locating problem from the traditional classification matching mode to the generative inversion mode based on the physical process by constructing the cause-effect generation module and combining the event inversion module to solve the fault.

[0054] 2、The application extends the locating ability from a single fault point to the reconstruction of a complex accident sequence by setting the cause-effect coordination verification module and the extensible event model.

[0055] 3、The application has the ability to comprehensively evaluate and review the self-positioning ability of the system in the offline stage by introducing the meta-cognition analysis module.

[0056] 4、The application improves the solving efficiency and accuracy of fault locating by designing the dynamic pruning module. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 It is a structural schematic diagram of the system function module of the application.

[0058] Figure 2An online fault diagnosis flowchart of the present application;

[0059] Figure 3 An offline meta-cognition analysis and diagnosability map generation flowchart of the present application;

[0060] Figure 4 A diagnosability map schematic diagram of the present application.

[0061] Wherein, 10, a causal generation module; 20, an event inversion module; 30, a dynamic pruning module; 40, a causal collaborative corroboration module; 50, a meta-cognition analysis module; 60, a diagnosability map; 61, a diagnostic blind area. DETAILED DESCRIPTION

[0062] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0063] Referring to Figure 1 The present application provides a power distribution network fault positioning system, which can be deployed in a server or a dedicated edge computing device. The system comprises a causal generation module 10, an event inversion module 20, a dynamic pruning module 30, a causal collaborative corroboration module 40, and a meta-cognition analysis module 50.

[0064] The causal generation module 10 is configured to establish a deterministic mapping between fault event parameters and the whole network physical response caused by the fault event parameters. The module uses a pre-trained Fourier neural operator proxy model to generate a spatiotemporal transient physical field covering the whole network of the power distribution network as output when receiving a set of fault event parameters as input.

[0065] The event inversion module 20 is connected with the causal generation module 10. After the system receives actual fault observation waveforms collected by one or more sensors in the power distribution network, the event inversion module 20 takes the actual fault observation waveforms as the target, iteratively calls the causal generation module 10 for iterative calculation, and inversely solves a set of optimal fault event parameters that can optimally generate the actual fault observation waveforms and contain fault location information.

[0066] The dynamic pruning module 30 is connected with the event inversion module 20 and the cause generation module 10. During the iterative solving process of the event inversion module 20, the dynamic pruning module 30 synchronously analyzes the spatio-temporal transient physical field generated by the cause generation module 10 in the iteration, and identifies and removes the physically unreasonable fault location hypothesis region from the subsequent search space of the event inversion module 20 according to the preset physical rules.

[0067] The cause coordination verification module 40 is connected with the output end of the event inversion module 20. After the event inversion module 20 solves the optimal fault event parameters, the cause coordination verification module 40 first calls the cause generation module 10 to generate a final spatio-temporal transient physical field with the optimal fault event parameters. Subsequently, the module calculates a final diagnostic confidence associated with the positioning result according to the consistency between the final spatio-temporal transient physical field and the actual fault observation waveform in multiple physical dimensions.

[0068] The metacognition analysis module 50 is connected with the cause generation module 10, the event inversion module 20 and the cause coordination verification module 40, and operates in an offline state. The module injects a series of virtual fault event parameters into the system, and calls other modules for complete closed-loop simulation, to generate a diagnosability map 60 as shown in Figure 4 , which evaluates the fault location capability of the current power distribution network sensor layout, and identifies the diagnostic blind area 61.

[0069] Referring to Figure 2 , the application also provides a power distribution network fault location method. The method is an online diagnosis process, and specifically includes the following steps:

[0070] Step S1, the cause generation module 10 is used to generate a corresponding spatio-temporal transient physical field according to the input fault event parameters.

[0071] Step S2, after receiving the actual fault observation waveform, the event inversion module 20 takes the waveform as the target, and iteratively solves the optimal fault event parameters containing fault location information by calling step S1.

[0072] Step S3, during the iterative solving process of step S2, the dynamic pruning module 30 analyzes the intermediate generated spatio-temporal transient physical field, and removes unreasonable fault location hypothesis regions from the subsequent search space of step S2 according to the physical rules.

[0073] Step S4, after the optimal fault event parameters are solved in step S2, the cause coordination verification module 40 calls step S1 to generate a final spatio-temporal transient physical field, and calculates a final diagnostic confidence associated with the positioning result.

[0074] Step S4-1, determine whether the final diagnostic confidence calculated in step S4 is lower than a preset confidence threshold. If the result is no, output the optimal fault event parameter containing the optimal fault location as the final diagnosis result, and the flow ends. If the result is yes, execute step S4-2.

[0075] Step S4-2, upgrade the event model used to describe the fault from a single event model to a sequence event model, and return to re-execute step S2 to search in a higher dimensional parameter space.

[0076] Referring to Figure 3 , the metacognition analysis module 50 performs the metacognition analysis step S5 in an offline state. This step performs a closed-loop virtual diagnosis flow including steps S1, S2, S4 once for each parameter in a set of preset virtual fault event parameters, to obtain an inferred fault event parameter and an expected diagnostic confidence. After traversing all virtual fault event parameters, integrate all results to generate a diagnosability map 60 as shown in Figure 4 .

[0077] The system and method of the embodiments of the present application can be deployed in a computing environment including but not limited to a server, a workstation, a cloud platform or an embedded computing unit. The computing environment includes a processor, a memory, a communication interface and the like hardware. The memory stores a computer program which is executed by the processor to implement the steps of the aforementioned method. The communication interface is used to interact with a power grid automation system, a wide area measurement system (WAMS) or other data acquisition terminals to obtain actual fault observation waveforms.

[0078] The specific modules and steps in the system and method shown in Figures 1 to 4 will be described in detail below.

[0079] In order to uniformly and structurally describe a single fault or even a complex cascading fault scenario, the present application defines an extensible event grammar (Event-Grammar) for specifying the data structure of the fault event parameter. The event grammar is the technical basis for the entire system to perform causal generation, event inversion and model upgrading. The event grammar specifically includes:

[0080] Single event model (Single-Event-Model): used to describe an independent, single fault event. Its data structure is a parameter set containing all key physical properties describing the fault. A typical single event model can be defined as:

[0081] ;

[0082] In the formula, Define the fault location, consisting of line identifier and relative distance (e.g. percentage) from the start of the line; Define the fault type, e.g. single-phase-to-ground, two-phase short circuit, three-phase short circuit, etc. enumerated values; Define the initial phase angle of the fault occurrence, which affects the initial state of the transient process; Define the transition resistance at the fault point, which affects the magnitude and waveform characteristics of the fault current.

[0083] Sequential-Event-Model: used to describe the cascading fault scenario that is triggered by an initial fault after a certain time delay. This model is built on top of the single-event model, by combining two single-event models and a time parameter. Its data structure can be defined as:

[0084]

[0085] where, is a parameter set following the structure of single-event model, used to describe the initial fault; is another parameter set following the structure of single-event model, used to describe the secondary fault triggered by ; is a time interval parameter, representing the time elapsed from occurrence to occurrence.

[0086] This extensible event syntax design makes the solving target of event inversion module 20 parameterized and structured, and provides a clear data structure basis for the automatic upgrade of cause-effect collaborative validation module 40 from single-event model to sequential-event model, thus realizing the reconstruction of the whole picture of complex accidents.

[0087] The detailed implementation of cause-effect generation module 10 is described with reference to Figure 1 . The function of this module is to build an efficient and accurate proxy model from fault event parameters to the whole network transient physical field, providing a basis for subsequent event inversion and validation. This module is pre-trained and executed in step S1 in the online diagnosis process of the system.

[0088] ​The core of the causal generation module 10 is a Fourier neural operator surrogate model. The reason for choosing this model structure is that the transient traveling wave propagation process triggered by a fault in the distribution network can be described by a partial differential equation (specifically, the telegraph equation). The Fourier neural operator surrogate model can process distribution networks with complex topologies without relying on specific mesh discretization by directly learning to solve the operator in the Fourier domain. When solving physical processes defined by partial differential equations, the computational efficiency is significantly improved compared to traditional time-domain finite difference or finite element numerical simulation methods. This feature enables the model to meet the technical requirements of generating a time-space transient physical field covering the entire distribution network.

[0089] When pre-training the Fourier neural operator surrogate model, the goal is to learn an accurate mapping from fault event parameters to time-space transient physical fields. Fault event parameters are a set of structured data describing the cause of the fault, which can include: fault location (defined by line identifier and relative distance percentage from the line head), fault type (such as single-phase ground, phase-to-phase short circuit), fault initial phase angle, and fault transition resistance value. The time-space transient physical field is a data structure that represents the change of physical quantities (such as voltage or current) at all discrete points in the distribution network model over time within a preset time window.

[0090] The training process is optimized by a composite loss function. In calculating the composite loss function, first, based on a set of training fault event parameters, the Fourier neural operator surrogate model is called to generate a corresponding time-space transient physical field; then, based on the generated time-space transient physical field, a data fidelity term and a physical residual term are calculated. Finally, based on the data fidelity term and the physical residual term, the composite loss function is constructed, and its calculation formula is:

[0091] ;

[0092] In the formula, is the composite loss function; is the data fidelity term; is the physical residual term; and are weight coefficients for balancing the data fidelity term and the physical residual term.

[0093] Data fidelity term The calculation process of the data fidelity term is as follows: for the same set of input training fault event parameters, on the one hand, a space-time transient physical field is generated by the Fourier neural operator proxy model, and on the other hand, a space-time transient physical field serving as a benchmark is generated by a high-precision electromagnetic transient simulation software (for example, PSCAD or EMTP). The data fidelity term is obtained by calculating the norm difference (for example, mean square error) between the two physical fields. The purpose of this term is to ensure the numerical accuracy of the model generation result.

[0094] The physical residual term The calculation process of the data fidelity term is as follows: for the same set of input training fault event parameters, on the one hand, a space-time transient physical field is generated by the Fourier neural operator proxy model, and on the other hand, a space-time transient physical field serving as a benchmark is generated by a high-precision electromagnetic transient simulation software (for example, PSCAD or EMTP). The data fidelity term is obtained by calculating the norm difference (for example, mean square error) between the two physical fields. The purpose of this term is to ensure the numerical accuracy of the model generation result.

[0095] Reference Figure 1 and Figure 2 The detailed implementation of the event inversion module 20 is described. This module performs step S2 in the online diagnosis process of the system, and its core task is to solve an inverse problem:

[0096] After receiving the actual fault observation waveform collected by the power distribution network sensor, the cause leading to this result, i.e., a set of optimal fault event parameters containing fault location information, is inferred and solved in reverse.

[0097] To achieve this reverse solving, the event inversion module 20 first constructs an objective function and minimizes the value of the objective function through iterative calculation. In constructing the objective function, the module calls the causal generation module 10 to generate a set of corresponding transient waveforms according to a set of fault event parameters to be evaluated. Subsequently, the module calculates the similarity between the generated waveform and the actual fault observation waveform. The similarity calculation here uses the dynamic time warping algorithm. The reason for using this algorithm is that there are slight deviations in the data collection start times of different sensors, and the waveform shape of the transient traveling wave will be stretched or compressed to some extent during propagation. The dynamic time warping algorithm can measure the similarity in shape by calculating the optimal matching path between two time series, rather than strictly calculating the point-to-point Euclidean distance, thereby reducing the requirement for high-precision strict clock synchronization of all sensors by the system.

[0098] The specific calculation formula of the objective function is:

[0099] ;

[0100] wherein, is the objective function, whose value represents the degree of dissimilarity between the generated waveform and the actual observed waveform, the fault event parameters to be solved, is the objective function, whose value represents the degree of dissimilarity between the generated waveform and the actual observed waveform, is the objective function, whose value represents the degree of dissimilarity between the generated waveform and the actual observed waveform, is the objective function, whose value represents the degree of dissimilarity between the generated waveform and the actual observed waveform, is the objective function, whose value represents the degree of dissimilarity between the generated waveform and the actual observed waveform, is the objective function, whose value represents the degree of dissimilarity between the generated waveform and the actual observed waveform, is the objective function, whose value represents the degree of dissimilarity between the generated waveform and the actual observed waveform, is the objective function, whose value represents the degree of dissimilarity between the generated waveform and the actual observed waveform, is the objective function, whose value represents the degree of dissimilarity between the generated waveform and the actual observed waveform, is the objective function, whose value represents the degree of dissimilarity between the generated waveform and the actual observed waveform, is the objective function, whose value represents the degree of dissimilarity between the generated waveform and the actual observed waveform, is the objective function, whose value represents the degree of dissimilarity between the generated waveform and the actual observed waveform,

[0101] The event inversion module 20 further comprises a solving unit for searching in the fault event parameter space to find the optimal fault event parameters that make the objective function reach the minimum value. The solving unit adopts a Bayesian optimization algorithm. The reason for choosing this algorithm is that each evaluation of the objective function needs to call the causal generation module 10 once, and the calculation cost is high, which is a typical expensive black-box function optimization problem. The Bayesian optimization algorithm can find the global optimal solution in a few iterations by constructing a probability proxy model of the objective function and using the acquisition function to guide the sampling position of the next iteration, thereby improving the overall calculation efficiency of the event inversion process while ensuring the solution accuracy.

[0102] The detailed implementation of the dynamic pruning module 30 is described with reference to Figure 1 and Figure 2 . The search space of the event inversion module 20 when solving the optimal fault event parameters has high dimension and wide range. If global optimization is performed without constraints, it will lead to large consumption of calculation resources and the solution process is easy to converge to a local optimal solution that is not physically valid. The dynamic pruning module 30 is set up to solve this problem. This module runs in parallel with the iterative solution process of the event inversion module 20, executes step S3, and removes unreasonable fault location hypothesis areas from the subsequent search space using prior physical knowledge, thereby improving the solution efficiency and the accuracy of the results.

[0103] The dynamic pruning module 30 comprises a qualitative feature analysis unit and a spatial pruning unit. The qualitative feature analysis unit is used to extract the waveform qualitative features at each sensor location, such as the initial polarity of transient voltage or current and the transient frequency component, of the spatio-temporal transient physical field generated by the causal generation module 10 in each iteration.

[0104] Taking the waveform polarity as an example, for a single-phase ground fault occurring in the distribution network, based on the traveling wave theory, the initial polarity of the transient traveling wave current detected by all the measurement points upstream of the fault point is the same, while the initial polarity of the transient traveling wave current detected by all the measurement points downstream of the fault point is opposite to that of the upstream. In a certain iteration of the event inversion module 20, if a hypothetical fault location is included in the fault event parameters to be evaluated, the qualitative feature analysis unit first extracts the initial polarity of the waveform generated by the hypothesis at each sensor location, and compares it with the initial polarity of the actual fault observation waveform. If it is found that the generated waveform polarity distribution is inconsistent with this physical law, the spatial pruning unit determines that the fault location of the hypothesis is not established, and removes the location and its adjacent line region from the subsequent search space of the event inversion module 20.

[0105] Taking the transient frequency component as another example, due to the distributed capacitance effect of the cable line in the distribution network, the transient traveling wave propagating on it usually contains more transient high-frequency components than the overhead line. If the main energy of the actual fault observation waveform is concentrated in the low frequency band after analysis, and the fault location hypothesis currently evaluated by the event inversion module 20 is located in or adjacent to a section of cable line, there is a physical contradiction between the hypothesis and the observation. The spatial pruning unit judges accordingly to remove the fault location region containing the section of cable line from the subsequent search space. In this way, the dynamic pruning module 30 introduces physical constraints into the mathematical optimization process, avoiding the invalid iterative calculation of the solving algorithm in the unsolved region.

[0106] Reference Figure 1 and Figure 2 The detailed implementation of the causal collaborative verification module 40 is described. After the event inversion module 20 solves the optimal fault event parameters, the traditional diagnostic method usually outputs it as the final result, but cannot evaluate the reliability of the positioning result. The causal collaborative verification module 40 is set to solve this technical problem, which is used to verify the physical self-consistency of the solution output by the event inversion module 20 in step S4, and calculate a quantitative final diagnostic confidence to reflect the reliability of the positioning result.

[0107] When the module makes a consistency judgment, it verifies the final spatio-temporal transient physical field generated by the optimal fault event parameters and the actual fault observation waveform from multiple physical dimensions. The specific verification dimensions include:

[0108] Temporal validation: According to the optimal fault location calculated by the event inversion module 20, combined with the pre-stored power distribution network topology data (including the length and connection relationship of each line) and the propagation speed of the transient traveling wave in the corresponding medium, the theoretical propagation time from the fault location to each sensor installation location is calculated. Sort all the theoretical propagation times of the sensors to get a theoretical disturbance arrival time sequence. Compare this theoretical time sequence with the arrival time sequence of the disturbance signal detected by each sensor extracted from the actual fault observation waveform. The higher the consistency between the two, the higher the credibility of the solution.

[0109] Amplitude validation: The amplitude of the transient traveling wave will decrease as the propagation distance increases. According to the calculated optimal fault location and the distribution of each sensor on the line, the relative intensity relationship that the observed transient waveforms at each sensor should satisfy can be predicted. Compare this theoretical relative amplitude relationship with the actual relative amplitude relationship of each waveform measured from the actual fault observation waveform. The higher the consistency between the two, the higher the credibility of the solution.

[0110] Morphology validation: When the transient traveling wave encounters impedance discontinuities such as line branches, ends, and different types of cable connections in the propagation path, refraction and reflection occur, which are superimposed on the original traveling wave, resulting in specific morphological characteristics of the waveform observed at the sensor, such as oscillation or attenuation characteristics. The cause-and-effect coordination validation module 40 judges whether the morphological characteristics of the actual fault observation waveform are consistent with the predicted waveform morphology after propagation on the corresponding line and reflection at the impedance discontinuity according to the optimal fault location. The higher the consistency between the two, the higher the credibility of the solution.

[0111] The cause-and-effect coordination validation module 40 comprehensively calculates a final diagnostic confidence value based on the consistency in the above-mentioned multiple physical dimensions.

[0112] Referring to Figure 2 After step S4, the module further performs step S4-1, comparing the calculated final diagnostic confidence with a pre-set confidence threshold. If the diagnostic confidence is higher than or equal to the threshold, it is considered that the current optimal fault event parameters are sufficient to reasonably explain the received actual fault observation waveform, and the diagnostic process ends and outputs the result.

[0113] If the diagnosis confidence is below the threshold, step S4-2 is performed. At this point, the module determines that the single-event model currently employed to describe the single fault is not sufficient to explain the complex observation waveform. Consequently, the module upgrades the event model used to describe the fault from a single-event model to a complex event model, such as a sequence event model defined in the system to describe a sequence of faults induced by one fault after a certain time interval. The parameter structure of the sequence event model can be defined to contain two sets of independent fault event parameters and one time interval parameter. After the model is upgraded, the module triggers the event inversion module 20 to return and re-performs step S2 to search for a solution in a higher-dimensional parameter space defined by the sequence event model, thereby enabling the analysis and localization of complex accident scenarios such as cascading faults or concurrent faults.

[0114] Reference is made to Figure 1 , Figure 3 and Figure 4 for a detailed implementation of the meta-cognitive analysis module 50. The module operates in an offline state of the system, and performs step S5, which is to quantify the overall fault localization capability of the current system under a certain power distribution network topology and sensor layout, and identify areas of weak or non-coverage (i.e. localization blind areas).

[0115] The meta-cognitive analysis module 50 includes a virtual diagnosis loop unit and a map generation unit. The operation flow of the module is as follows:

[0116] First, in the offline state, the established digital model of the power distribution network is discretized into a series of virtual fault points along all lines thereof at a pre-set spatial step (e.g. every 50 meters).

[0117] Subsequently, the virtual diagnosis loop unit injects one or more pre-set virtual fault event parameters representing typical fault types for each virtual fault point. For each injected virtual fault event parameter, the unit performs a complete closed-loop virtual diagnosis flow:

[0118] First, the causal generation module 10 is invoked to generate a set of virtual observation waveforms at the existing sensor layout positions of the power distribution network, with the virtual fault event parameter as input;

[0119] Next, the virtual observation waveforms are inputted and the event inversion module 20 and the causal synergy validation module 40 are invoked to perform a complete online diagnosis flow including steps S2 to S4.

[0120] The output of this flow, i.e. an inferred fault event parameter and a corresponding expected diagnosis confidence, is recorded and associated with the injected virtual fault point.

[0121] After the virtual diagnosis cycle unit traverses all the preset virtual fault points and virtual fault event parameters, the map generation unit integrates all the recorded expected diagnosis confidence data. This unit presents the expected diagnosis confidence value of each location on the map through a visual method (for example, a color-coded heat map) against the background of the topology structure map of the power distribution network, thereby generating a diagnosability map 60 as shown in Figure 4 .

[0122] In this diagnosability map 60, different colors or gray scales in the map represent the expected diagnosis confidence level of the system when a fault occurs at the location. Among them, the area where the expected diagnosis confidence is significantly lower than the preset threshold is clearly marked as a diagnosis blind area 61. The diagnosability map provides an intuitive and quantitative evaluation result for the sensor network diagnosis capability for the planner, and provides a direct technical decision basis for subsequent sensor optimization layout, sensor increase or installation position adjustment, etc., the purpose of which is to eliminate the diagnosis blind area and improve the overall positioning coverage and reliability of the system.

[0123] In order to more clearly illustrate the cooperative workflow of the system and method provided by the present application, a hypothetical fault scenario is described below.

[0124] Referring to Figure 1 and Figure 2 , it is assumed that a single-phase ground fault occurs on a line of the power distribution network. The multiple sensors installed in the power grid capture the transient traveling wave triggered by this fault and transmit the collected actual fault observation waveform data to the system. After receiving the data, the event inversion module 20 starts the iterative solving process of step S2 to search for the optimal fault event parameters containing the fault location in the preset single-event model parameter space. In the early stage of iteration, when the assumed fault location evaluated by the event inversion module 20 deviates far from the true location, the dynamic pruning module 30 performs step S3 in parallel to quickly remove a large area of physically impossible lines from the subsequent search space by comparing the initial polarity of the intermediate waveform generated by the cause-effect generation module 10 with the actual observation waveform. Thereafter, the event inversion module 20 continues to search in the reduced space using the Bayesian optimization algorithm and finally converges to a set of optimal fault event parameters containing the optimal fault location.

[0125] Subsequently, the process enters step S4, and the cause-effect cooperative validation module 40 receives the set of parameters. The module first calls the cause-effect generation module 10 to generate the final spatiotemporal transient physical field and validates it from the time, amplitude, and shape dimensions with the actual observation waveform to calculate the final diagnosis confidence. If the confidence is higher than the preset threshold, the system outputs the set of optimal fault event parameters containing the optimal fault location as the final diagnosis result.

[0126] In another complex scenario, a secondary fault is induced at another location shortly after the initial fault occurs. At this time, any set of optimal parameters solved by the event inversion module 20 based on the single-event model, which contains a single fault location, cannot simultaneously satisfy the physical self-consistency requirement of all sensor observation waveforms under the verification of the cause-effect collaborative validation module 40, resulting in a final diagnosis confidence lower than the preset threshold. The cause-effect collaborative validation module 40 accordingly performs step S4-2, upgrades the event model from the single-event model to the sequence-event model, and triggers the event inversion module 20 to return to step S2. The event inversion module 20 then re-searches and solves in a higher-dimensional sequence-event parameter space, and finally outputs a diagnosis result containing two fault locations, fault types, and time intervals that can reconstruct the chain fault process.

[0127] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A power distribution network fault location system characterized by, The system comprises: a causal generation module, which adopts a pre-trained Fourier neural operator proxy model to input fault event parameters to generate a corresponding spatio-temporal transient physical field covering the entire power distribution network; an event inversion module, which, after receiving an actual fault observation waveform of the power distribution network, takes the actual fault observation waveform as a target, iteratively calls the causal generation module, and reversely solves optimal fault event parameters containing fault location information that can optimally generate the actual fault observation waveform; a dynamic pruning module, which analyzes the spatio-temporal transient physical field during the iterative solving process of the event inversion module, and removes unreasonable fault location hypothesis areas from the subsequent search space of the event inversion module according to preset physical rules; a causal collaborative verification module, which, after the event inversion module solves the optimal fault event parameters, calls the causal generation module to generate a final spatio-temporal transient physical field, and calculates a final diagnostic confidence associated with the optimal fault location information according to the consistency between the final spatio-temporal transient physical field and the actual fault observation waveform in multiple physical dimensions; a meta-cognition analysis module, which runs in an offline state, injects virtual fault event parameters, and calls the causal generation module, the event inversion module, and the causal collaborative verification module to perform closed-loop simulation to generate a diagnosability map evaluating the fault location capability of the system under the current sensor layout of the power distribution network, and identify diagnostic blind areas.

2. The power distribution network fault location system of claim 1, wherein, The causal generation module is used for: During pre-training of the Fourier neural operator proxy model, a composite loss function is used for optimization to learn the mapping relationship from fault event parameters to spatio-temporal transient physical fields; When defining the composite loss function, a set of training fault event parameters are used to call the Fourier neural operator proxy model to generate corresponding spatio-temporal transient physical fields; Based on the generated spatio-temporal transient physical fields, a data fidelity term that minimizes the difference between the spatio-temporal transient physical fields and the results of high-precision simulation software, and a physical residual term that minimizes the violation degree of the spatio-temporal transient physical fields to the basic physical equations of the power distribution network are calculated; The sum of the data fidelity term and the physical residual term is used to obtain the composite loss function, and the calculation formula of the composite loss function is: ； wherein is a composite loss function; is a data fidelity term; is a physical residual term; and is a weight coefficient balancing the data fidelity term and the physical residual term.

3. The power distribution network fault location system of claim 1, wherein, The event inversion module is used for: To reversely solve the optimal fault event parameters containing fault location information, a target function is constructed and iteratively solved in a minimization manner; When constructing the target function, the causal generation module is called to generate corresponding waveforms according to a set of fault event parameters to be evaluated, and the similarity between the generated waveforms and the actual fault observation waveform is calculated based on the generated waveforms and the actual fault observation waveform by using a dynamic time warping algorithm, and the calculation formula of the target function is: ； wherein, is the objective function, is the fault event parameter to be solved, which contains the fault location information; is the number of sensors receiving the actual fault observation waveform; is the actual fault observation waveform collected by the sensor; is the waveform generated by the causal generation module according to the fault event parameter at the sensor location ; is the distance calculated by the dynamic time warping algorithm; is the weight of the observation data of the sensor in the objective function; The event inversion module is also used for a solving unit based on a Bayesian optimization algorithm, which searches in the fault event parameter space to obtain the optimal fault event parameters with the minimum value for the purpose of minimizing the target function.

4. The power distribution network fault location system of claim 1, wherein, The dynamic pruning module is used for: a qualitative feature analysis unit, which extracts the waveform polarity and transient frequency component of the spatio-temporal transient physical field at the sensor location generated by the causal generation module during iteration; The space pruning unit removes the fault location hypothesis region corresponding to the time-space transient physical field from the subsequent search space of the event inversion module when there is a fundamental physical contradiction between the waveform polarity or transient frequency component and the actual fault observation waveform.

5. The power distribution network fault location system of claim 1, wherein, The causal collaborative verification module is configured to: When making a consistency judgment in multiple physical dimensions, the multiple physical dimensions include time verification, amplitude verification, and shape verification, wherein, The time verification judges whether the disturbance arrival time sequence of the actual fault observation waveform is consistent with the propagation time sequence predicted according to the optimal fault location information; The amplitude verification judges whether the relative intensity of the actual fault observation waveform is consistent with the relationship between the attenuation and the propagation distance predicted according to the optimal fault location; The shape verification judges whether the waveform characteristics of the actual fault observation waveform are consistent with the shape predicted according to the optimal fault location after propagation on the corresponding line and passing through the reflection.

6. The power distribution network fault location system of claim 1, wherein, The causal collaborative verification module is further configured to: When the final diagnostic confidence is higher than or equal to the preset confidence threshold, it is judged that the current optimal fault event parameter is sufficient to reasonably explain the received actual fault observation waveform, and the diagnostic process ends and the result is output; When the final diagnostic confidence is lower than the preset confidence threshold, it is judged that the event model describing a single fault is insufficient to explain the actual fault observation waveform, and the event inversion module is triggered to solve the optimal fault event parameter in the complex event model space describing a sequence fault or a concurrent fault again.

7. The power distribution network fault location system of claim 1, wherein, The metacognition analysis module is configured to: A virtual diagnosis loop unit, for each injected virtual fault event parameter, calls the causal generation module to generate a virtual observation waveform, and then inputs the virtual observation waveform and calls the event inversion module and the causal collaborative verification module to perform a complete diagnostic process to obtain inferred fault event parameters and expected diagnostic confidence; A map generation unit takes all the inference results and expected diagnostic confidence obtained after the virtual diagnosis loop unit traverses all the virtual fault event parameters as input, integrates and generates a diagnosability map.

8. A power distribution network fault location system according to claim 7, characterised in that, The data structure of the fault event parameter is defined as an extensible event syntax, and the event syntax includes: A single event model for describing an independent single fault event; And a sequence event model for describing a cascading fault scenario in which an initial fault triggers another secondary fault after a certain time delay.

9. A method for fault location in a power distribution network, applied to the system of any of claims 1-8, characterized by, The method comprises the following steps: S1, using a pre-trained Fourier neural operator proxy model, generating a time-space transient physical field corresponding to the fault event parameter according to the input fault event parameter, covering the entire power distribution network; S2, after receiving the actual fault observation waveform of the power distribution network, taking the actual fault observation waveform as the target, iteratively solving the optimal fault event parameter containing the fault location information that can optimally generate the actual fault observation waveform by calling step S1. S3. During the iterative solving process of step S2, analyze the spatio-temporal transient physical field, and remove unreasonable fault location hypothesis regions from the subsequent search space of step S2 according to pre-set physical rules; S4. After the optimal fault event parameters are solved in step S2, call step S1 to generate the final spatio-temporal transient physical field, and calculate the final diagnostic confidence associated with the optimal fault location information according to the consistency between the final spatio-temporal transient physical field and the actual fault observation waveform in multiple physical dimensions; S5. In the offline state, by injecting virtual fault event parameters, call steps S1, S2 and S4 to perform closed-loop simulation to generate a diagnosability map for evaluating the system fault location capability under the current power distribution network sensor layout, and identify the diagnostic blind area.

10. The method of claim 9, wherein, After step S4, further comprising: S4-1: judge whether the final diagnostic confidence calculated in step S4 is lower than a pre-set confidence threshold; S4-2: if the final diagnostic confidence is lower than the pre-set confidence threshold, upgrade the event model describing the fault from a single event model to a sequence event model, and return to re-execute step S2.

Images