A low-voltage power distribution network gray region situation awareness method, system, device and medium

Abstract

The present application belongs to the technical field of low-voltage power distribution network monitoring and fault diagnosis, and discloses a low-voltage power distribution network gray domain situation awareness method, system, device and medium to solve the problem that the prior art cannot simultaneously consider multi-rate measurement collaborative utilization and feeder-level accurate positioning. The method comprises: acquiring a transformer area topology and a multi-rate measurement sequence; determining candidate feeders and node topology position characteristics according to the topology; using a root node time sequence diagnosis model to encode a high-resolution sequence to output fault state, mechanism and phase probability; if the fault condition is met, generating a node initial representation by combining a topology position characteristic with snapshots before and after a sampling interval; obtaining a topology-aware node representation through topology propagation; for each candidate feeder, generating a candidate feeder representation from the node representation and a non-candidate feeder representation from the remaining nodes, and constructing a comparison difference representation based on the difference between the two; fusing the above representations to output a feeder fault positioning probability, and finally obtaining a situation awareness result.

Description

Technical Field

[0001] This invention belongs to the field of low-voltage distribution network monitoring and fault diagnosis technology, specifically relating to a method, system, equipment and medium for gray-domain situational awareness of low-voltage distribution networks. Background Technology

[0002] Low-voltage distribution network areas often suffer from unbalanced measurement configurations. While high-time-resolution three-phase electrical quantities at the millisecond or cycle level can be obtained from the low-voltage side or main outgoing line of the transformer in the distribution area, measurement devices on the user side or branch nodes often only provide aggregated measurement snapshots at intervals of several minutes or even longer. This measurement discrepancy means that although local disturbances are quickly observable at the root node, their distribution location and impact range in the feeder space can only be observed with delay through low-time-resolution snapshots at the node side, creating an incomplete information state where local observations are possible but the overall situation remains unclear.

[0003] To address the aforementioned issues, existing fault diagnosis methods typically resample measurements at different rates to the same time scale or rely solely on measurements from a single side for classification. The former inevitably introduces interpolation errors, especially during fault transients, which can smooth out or even weaken abrupt changes in fault characteristics; the latter struggles to simultaneously meet the dual requirements of rapid fault identification and feeder-level spatial location. Furthermore, low-voltage feeders exhibit a radial topology, with fault effects propagating along the power supply path. Measurement responses from adjacent nodes on the same feeder exhibit strong spatial correlation, but existing methods often treat each node as an independent sample, failing to fully utilize the spatial correlation information formed by topological connections, thus limiting the accuracy of feeder-level location. Simultaneously, purely data-driven models, lacking physical constraints, may output diagnostic results that do not conform to basic electrical principles, such as identifying poor contact faults as multi-phase faults, complicating operation and maintenance assessments.

[0004] Therefore, there is an urgent need for a gray-domain situational awareness method that can assign diagnostic responsibilities based on the physical observability of different measurement rates without changing the existing measurement configuration, and integrate evidence of rapid perturbation of root nodes, low temporal resolution spatial response of nodes, and feeder topology constraints. Summary of the Invention

[0005] Based on the aforementioned shortcomings and deficiencies in the existing technology, one of the objectives of this invention is to at least solve one or more of the aforementioned problems in the existing technology. In other words, one of the objectives of this invention is to provide a method, system, device, and medium for gray-domain situational awareness of low-voltage distribution networks that meets one or more of the aforementioned requirements. This aims to achieve the goal of allocating diagnostic responsibilities based on the physical observability of different measurement rates without changing the existing measurement configuration, and to integrate evidence of rapid disturbances at root nodes with low temporal resolution spatial responses of nodes and feeder topology constraints, thereby realizing the structured identification of fault states, fault mechanisms, fault phase sets, and affected feeders.

[0006] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for gray-domain situational awareness in low-voltage distribution networks, comprising the following steps: Acquire the feeder topology of the low-voltage distribution area, the high time resolution sequence of the three-phase electrical quantities at the root node, and the low time resolution snapshot sequence of the three-phase measurements at the node side; The relationship between nodes and feeders, the adjacency relationship between nodes, the topological position characteristics of nodes relative to the root node, and all candidate feeders are determined based on the feeder topology. The root node time-series diagnostic model is used to encode the three-phase electrical quantities of the root node with high time resolution, obtain the root node disturbance characterization, and output the diagnostic probability of fault state, fault mechanism and fault phase set. The fault determination condition is determined based on the diagnostic probability of the fault state. If the condition is not met, the normal state is output. If the condition is met, the sampling interval is determined. Before and after the interval, low time resolution snapshots of three-phase measurements on the node side are selected and fused with the topological location features to generate the initial node characterization. Based on the feeder topology, the initial node representation is propagated through topology messages to obtain the topology-aware node representation. For each candidate feeder, a candidate feeder representation is generated based on the topology-aware node representation of the nodes contained in the candidate feeder, a non-candidate feeder representation is generated based on the node representation of the non-candidate feeders, and a contrast difference representation relative to other feeders is constructed based on the difference between the candidate feeder representation and the non-candidate feeder representation. By integrating candidate feeder characterization, comparative difference characterization, and root node disturbance characterization, the fault location probability of each candidate feeder is output. Based on the fault status, fault mechanism, fault phase set, and fault location probability, the gray-domain situational awareness results are obtained.

[0007] In a second aspect, the present invention provides a low-voltage distribution network gray-domain situational awareness system for implementing the method described in the first aspect, comprising: The data acquisition module is used to acquire the feeder topology, the high time resolution sequence of the three-phase electrical quantities at the root node, and the low time resolution snapshot sequence of the three-phase measurements at the node side. The topology feature construction module generates the node-feedline affiliation relationship, node adjacency relationship, topological position features of nodes relative to the root node, and all candidate feeders based on the feeder topology. The root node rapid diagnosis module is used to encode the three-phase electrical quantities of the root node with high time resolution, and output the root node disturbance characterization, fault state, fault mechanism and diagnostic probability of fault phase set. The node graph timing coding and topology propagation module is used to determine the sampling interval and select low temporal resolution snapshots of three-phase measurements on the node side before and after it. These snapshots are then fused with topological location features to generate an initial node representation. Based on the feeder topology, the initial node representation is propagated with topological messages to output a topology-aware node representation. The multi-rate feeder positioning module generates candidate feeder representations based on the topology-aware node representations of the nodes contained in the candidate feeders and generates non-candidate feeder representations based on the node representations contained in the non-candidate feeders. Based on the differences between the candidate feeder representations and the non-candidate feeder representations, it constructs a comparative difference representation of each candidate feeder relative to other feeders. It then fuses the candidate feeder representations, comparative difference representations, and root node disturbance representations to output the fault location probability. The result decoding module outputs the results based on the fault state, fault mechanism, fault phase set, and fault location probability.

[0008] Thirdly, the present invention provides an electronic device, the electronic device including a memory, a processor and a computer program, wherein the computer program, when executed by the processor, implements the method described in the first aspect.

[0009] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0010] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention addresses the gray-domain characteristics of low-voltage distribution networks, where root nodes possess high temporal resolution measurements while node-side measurements only have low temporal resolution measurements. It decouples fault diagnosis into two independent yet coordinated stages: rapid root node identification and node-side spatial localization. This allows fast and slow measurements to each play a role within their respective observability advantages. Compared to existing technologies that forcibly resample measurements at different rates to the same time scale, this invention avoids the smoothing and weakening of transient fault characteristics by interpolation errors, achieving rapid fault identification and accurate fault localization while preserving key diagnostic information.

[0011] 2. Based on the initial node representation, this invention performs message propagation between nodes according to the feeder topology, enabling the representation of each node to incorporate abnormal information from adjacent nodes. Furthermore, it enhances the separability between feeders by constructing comparative difference features between candidate and non-candidate feeders. Compared to existing technologies that treat each node as an independent sample, this invention fully utilizes the spatial correlation information inherent in the radial topology of the low-voltage distribution network, effectively improving the accuracy of feeder-level fault location.

[0012] 3. This invention introduces legal combination constraints during model training, limiting the output of poor contact faults to single-phase, and simultaneously merging ground faults with their corresponding ungrounded faults. Compared to purely data-driven models that may output diagnostic results that do not conform to basic electrical principles, this invention, through the fusion of physical constraints and data-driven approaches, ensures that the diagnostic results always conform to electrical laws, thus improving the rationality and reliability of the diagnostic results.

[0013] Further or more detailed beneficial effects will be described in conjunction with specific embodiments in the detailed implementation. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart illustrating the gray-domain situational awareness method for low-voltage distribution networks as described in Embodiment 1 of the present invention.

[0016] Figure 2 This is a schematic diagram illustrating the working principle of the root node timing diagnosis model described in Embodiment 1 of the present invention.

[0017] Figure 3 This is a schematic diagram of the process for outputting the fault location probability of each candidate feeder as described in Embodiment 1 of the present invention.

[0018] Figure 4 This is a structural diagram of the electronic device described in Embodiment 3 of the present invention.

[0019] Icon labels: 400. Electronic devices; 401. Processor; 402. Communication bus; 403. User interface; 404. Network interface; 405. Memory. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0021] In the following description, several embodiments of the present invention are provided. Different embodiments can be substituted or combined. Therefore, the present invention can also be considered to include all possible combinations of the same and / or different embodiments described. Thus, if one embodiment includes features A, B, and C, and another embodiment includes features B and D, then the present invention should also be considered to include embodiments containing one or more other possible combinations of A, B, C, and D, even if such embodiments are not explicitly described in the following text.

[0022] The following description provides examples and does not limit the scope, applicability, or examples set forth in the claims. Changes may be made to the function and arrangement of the described elements without departing from the scope of the invention. Various processes or components may be appropriately omitted, substituted, or added to the various examples. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Furthermore, features described with respect to some examples may be combined into other examples.

[0023] To facilitate a better understanding of the embodiments of the present invention, its application scenarios will be explained before providing a detailed explanation of the specific implementation methods.

[0024] The gray-domain situational awareness method for low-voltage distribution networks described in this specification is applied to fault diagnosis scenarios in low-voltage distribution networks. In these scenarios, high temporal resolution measurements can be obtained at the low-voltage side of the transformer or at the main outgoing line, while only low temporal resolution measurements can be obtained at the user side or branch node side. The application of the gray-domain situational awareness method for low-voltage distribution networks aims to achieve structured identification of fault status, fault mechanism, fault phase set, and affected feeders by using the above-mentioned multi-rate measurement fusion.

[0025] The following is a brief explanation of the low-voltage distribution network, gray area, high time-resolution sequence of three-phase electrical quantities at the root node, low time-resolution snapshot sequence of three-phase measurements at the node side, root node timing diagnosis model, fault state, fault mechanism, fault phase set, initial node representation, topology sensing node representation, fault location probability, and gray area situational awareness results involved in several embodiments of this specification: Low-voltage distribution network refers to the power distribution network from the low-voltage side of the distribution transformer to the user terminal, including the low-voltage side busbar of the transformer, low-voltage feeders and various load nodes distributed along the line.

[0026] The gray area refers to a mixed data state where the root node can obtain fast disturbance measurements, while the node side can only obtain low temporal resolution measurements. That is, local disturbances are fast observable at the root node, while the distribution of fault effects in the feeder space can only be indirectly observed through low temporal resolution snapshots at the node side.

[0027] The high time-resolution sequence of three-phase electrical quantities at the root node refers to the high sampling rate time-series data of three-phase voltage, three-phase current, active power, and reactive power obtained at the measurement points on the low-voltage side or the total outgoing line of the transformer substation, with a sampling interval of 20 milliseconds.

[0028] The low time resolution snapshot sequence of three-phase measurements on the node side refers to the cross-sectional data of three-phase voltage, three-phase current, active power and reactive power provided by the node load side or node injection side measurement device at 15-minute intervals. Multiple consecutive cross-sections constitute a snapshot sequence.

[0029] The root node timing diagnosis model refers to a pre-trained model that performs timing encoding on the high time resolution sequence of the three-phase electrical quantities at the root node and outputs the fault state, fault mechanism, and fault phase set diagnosis probability.

[0030] Fault status refers to the binary state of whether a fault has occurred in the low-voltage distribution area, including fault status and normal status.

[0031] Fault mechanism refers to the physical type of fault, including short-circuit faults, open-circuit faults, and poor contact faults.

[0032] The fault phase set refers to the combination of phases involved in the fault, including seven categories: A, B, C, AB, BC, AC, and ABC.

[0033] Initial node representation refers to the node-level feature vector obtained by fusing low-temporal-resolution snapshots of the node with topological location features.

[0034] Topology-aware node representation refers to the node-level feature vector obtained by propagating the initial node representation through feeder topology messages and integrating neighborhood information.

[0035] Fault location probability refers to the probability value of each candidate feeder failing.

[0036] Gray-domain situational awareness results refer to structured diagnostic outputs that include whether a fault has occurred, the affected feeder, the fault mechanism, and the set of fault phases.

[0037] Example 1: This embodiment includes three low-voltage feeders. , , Taking a typical transformer substation as the research object, a feeder-level fault diagnosis task is established based on the multi-rate measurement configuration of this substation. The substation contains 26 nodes, where node 1 represents the root node on the low-voltage side of the substation transformer, and the node set of the three feeders is defined as follows: (1), (2), (3).

[0038] Full graph node set for: (4).

[0039] Feeder-level output definitions include fault states. Faulty feeder , Fault mechanism Separate sets .

[0040] Therefore, the final diagnostic output is recorded as: (5), In equation (5), the fault state, fault feeder, fault mechanism, and fault phase set respectively satisfy: (6), (7), (8), (9), In equation (7), N / A is only used as a placeholder for the feed field when a low temporal resolution snapshot has not yet been obtained in the normal sample or online stage.

[0041] For fault samples, the feeder localization output is limited to a set of three candidate feeders. The model's probability output includes fault detection probability, mechanism classification probability, phase set classification probability, and feeder classification probability: (10) in: (11), (12), (13) (14).

[0042] Regarding fault mechanism and phase set, in this embodiment, the fault type is described by two labels: fault mechanism and phase set. (15) Fault mechanism label is defined as: (16) In equation (16), These represent short-circuit faults, open-circuit faults, and poor contact faults, respectively.

[0043] The set label for the phase is defined as: (17) In equation (17), a single letter indicates a single-phase fault, two letters indicate a two-phase fault, and ABC indicates a three-phase fault.

[0044] In the defined labeling system, grounding participation is not treated as an independent output dimension, but is merged into the same phase set along with the corresponding non-grounding fault. For example, both AB short circuit and ABG short circuit are denoted as: (18).

[0045] Regarding the legal fault combinations and the short-circuit modeling range, this embodiment defines them as follows: Short-circuit and open-circuit faults allow for all seven phase sets, as shown in the following formulas: (19) (20).

[0046] In this embodiment, only single-phase contact failures are considered for contact-related faults; therefore, the allowed combinations are: (twenty one), Accordingly, the following combinations are defined as illegal combinations: (twenty two), (twenty three), (twenty four), (25).

[0047] like Figures 1 to 3 As shown in the figure, this embodiment provides a gray-domain situational awareness method for low-voltage distribution networks, including the following steps.

[0048] Step S1: Obtain the feeder topology of the low-voltage distribution area, the high time resolution sequence of the three-phase electrical quantities at the root node, and the low time resolution snapshot sequence of the three-phase measurements at the node side.

[0049] It is understandable that a low-voltage distribution area includes a root node and multiple low-voltage feeders. The root node is the low-voltage side busbar or the total outgoing line measurement point of the distribution area transformer.

[0050] Specifically, this step acquires the 20-millisecond cycle-level three-phase electrical quantity sequence at the root node and obtains low-temporal-resolution three-phase measurement snapshots for each load node at multiple 15-minute sampling sections. Each node's low-temporal-resolution snapshot includes the phase voltage, phase current, active power, and reactive power of phases A, B, and C. The node-side three-phase measurement low-temporal-resolution snapshot sequence is provided by user-side smart meters, node terminal measurement devices, edge measurement devices, or estimation fusion results.

[0051] For the root node, since it does not have subordinate user aggregation measurements, the phase voltage, phase current, active power and reactive power in the low time resolution input of the root node are obtained by averaging the high time resolution sequence of the three-phase electrical quantities of the root node within a preset short time window near the corresponding 15-minute sampling time. This preset short time window can be 1 second or set according to the measurement noise level.

[0052] The phase current in the low time resolution snapshot sequence of three-phase measurements on the node side is used to characterize the operating status of the node load side or node injection side. The node current refers to the aggregate current value of all load devices connected to the node at the same time, and does not represent the branch current on the power supply path from the root node to the node.

[0053] Furthermore, the multi-rate input tensor and nodal-side measurements can be further constructed according to the following mathematical model: Each rapid diagnostic sample is captured within a 5-second time window, which contains 250 sampling points at a 20 ms fundamental frequency level. (26) The root node input tensor is defined as: (27) The 12-dimensional channels correspond to three-phase voltage, current, active power, and reactive power.

[0054] This embodiment employs a phase-organized channel arrangement: (28) For the b-th sample and the t-th cycle sampling point, we have: (29) The root node sequence window covers the states before and after the fault; for example, a time window could be used. In this way, the timing encoder can not only observe the normal operating baseline before the fault occurs, but also capture fault abrupt changes, protection actions or continuous abnormal processes, as well as the steady-state response after the fault.

[0055] Node-level 15-minute snapshot measurement definition The root node 5 s cycle level three-phase measurement sequence is defined as follows: The four-step node-level 15-minute snapshot measurement is defined as follows: .

[0056] Node-level low temporal resolution measurements represent snapshot values ​​(instantaneous values) of the smart meter at 15-minute sampling intervals. Let the k-th sampling time be denoted as... Then a single node snapshot is recorded as: (30) In this embodiment, K=4 snapshots are used to construct low temporal resolution map samples: (31), If the fault occurs in: (32), In equations (30) to (32), , and All of these represent historical snapshots prior to the failure. This represents the first available node snapshot after the failure occurred. Since this embodiment only considers failures that will have ongoing operational consequences, it is assumed that the failure's impact is... The moment is still observable.

[0057] The node-level input tensor is: (33), Its dimensions correspond to the batch size, four 15-minute snapshots, 26 nodes, three phases (A / B / C), and four-dimensional features for each phase, respectively. .

[0058] Before entering the node-time encoder, the phase dimension and feature dimension are merged into 12-dimensional node features: (34).

[0059] Regarding node aggregation features and the construction of node 1 snapshot, for node... And parting Let the set of users connected to this node be . The node voltage is obtained by aggregating the median voltage of the terminal meters: (35), The aggregated load current, active power, and reactive power of a node are obtained by summing the measurements from the users under that node. (36) (37) (38), Therefore, the node Separation At any moment The characteristics of each phase are: (39).

[0060] Therefore, the node current referred to in this embodiment refers to the aggregated load current at the node, not the branch current flowing through the upstream line section. Node 1 has no end-user aggregated data; its 15-minute snapshot characteristics are derived from the 20-ms fundamental frequency level data of the root node. The average value is obtained from a nearby 1-second window. Let: (40) (41), For any root node variable The low temporal resolution snapshot feature of node 1 is defined as: (42).

[0061] Step S2: Determine the affiliation relationship between nodes and feeders, the adjacency relationship between nodes, the topological position characteristics of nodes relative to the root node, and all candidate feeders included in the low-voltage substation area based on the feeder topology.

[0062] Specifically, the node set, branch set, and feeder node set are determined based on the transformer area topology. In the graph neural network, the adjacency matrix is ​​undirected and self-loops are added, allowing nodes to retain their own anomaly characteristics while also receiving information from neighboring nodes. The topological location features include node number embedding, feeder affiliation one-hot encoding, normalized topological distance from the node to the root node, and a flag indicating whether it is a feeder terminal node. Candidate feeders are the feeders in the low-voltage transformer area whose status as affected by faults is to be determined; the candidate feeder set is directly determined by the feeder topology.

[0063] Furthermore, the transformer substation topology, feeder set, and adjacency matrix can be further defined as follows: For the transformer substation topology map and adjacency matrix, the low-voltage transformer substation map is defined as follows: (43), The edge sets of the three feeders are as follows: (44), (45) (46) The adjacency matrix is ​​denoted as: (47).

[0064] This embodiment uses an undirected topological graph as the graph structure, that is, if ,but: (48).

[0065] Step S3: Encode the high time resolution sequence of the three-phase electrical quantities of the root node using the pre-trained root node timing diagnosis model to obtain the root node disturbance characterization, and output the diagnostic probability of fault state, fault mechanism and fault phase set.

[0066] Specifically, the root node temporal diagnostic model employs a residual temporal convolutional network (Residual TCN). The root node fast diagnostic samples utilize a preset time window, covering the states before and after the fault at a 20-millisecond sampling interval. Each fast diagnostic sample extracts a 5-second time window, containing 250 sampling points at the 20-millisecond fundamental frequency level.

[0067] Furthermore, the root node temporal coding, attention pooling, and fast diagnostic head can be further implemented according to the following model: Regarding the overall architecture and tensor flow, the multi-rate topology-sensing feeder-level low-voltage fault diagnosis network proposed in this embodiment can be represented as the following overall mapping: (49).

[0068] The model consists of a root node fast timing branch, a node-level low temporal resolution graph branch, and a multi-rate feeder positioning branch. The root node branch receives... Output root node perturbation representation It provides fault detection, fault mechanism, and phase set probability through three task headers. Node-level low temporal resolution graph tributary reception. Adjacency Matrix First, the four-step snapshot sequence of each node is temporally encoded. Then, the static position encoding of the node is incorporated into the node representation. Subsequently, topological neighborhood information is propagated through GraphSAGE to obtain the topologically aware representation of each node. The feeder localization branch performs attention pooling on each feeder to construct relative difference features of candidate feeders, and then... The output probabilities of the three feeders are fused together.

[0069] The root encoder first converts the input tensor from The result is transposed to channel-priority form and then passed through three layers of one-dimensional dilated convolutional residual blocks. The kernel size of each of the three convolutional layers is 5, with dilation rates of 1, 2, and 4, respectively, and a total of 64 hidden channels. Each residual block consists of dilated convolution, LayerNorm, ReLU, random deactivation, and residual connections. The l-th layer can be abstractly represented as: (50), In equation (50), Includes dilated convolution, LayerNorm, ReLU, and random deactivation; when the residual branch dimensions are inconsistent. express Projection is performed if necessary, otherwise an identity mapping is used. The output of the three-layer residual TCN is projected to 128 dimensions via a linear mapping. Subsequently, the model uses attention-based temporal pooling to form the root node perturbation representation. For the t-th time step: (51), (52), The root node perturbation representation is fed into three rapid diagnostic heads.

[0070] The fault detection head uses a sigmoid output: (53), The fault mechanism identification head and the phase set identification head are respectively: (54), (55), In equations (54) and (55), The three components correspond to Short, Open, and Contact. The seven components correspond to A, B, C, AB, BC, AC, and ABC. This branch can provide rapid diagnostic output within seconds of a fault occurring.

[0071] In the diagnostic probability output of the fault phase set, ground faults and their corresponding ungrounded faults are assigned to the same phase set label. Specifically, both a phase A ground fault and a phase A unground fault are recorded as phase A faults, both an AB phase-to-phase short circuit and an AB phase-to-phase short circuit to ground fault are recorded as AB phase faults, and so on.

[0072] The training objectives of the root node time-series diagnostic model include fault detection loss, fault mechanism identification loss, phase set identification loss, feeder location loss, and legal combination constraint loss, and each loss term is weighted and summed. The legal combination constraint loss is used to constrain the fault phase set corresponding to poor contact faults to be single-phase. The training samples are constructed from a preset three-phase low-voltage distribution area sample model, historical measured samples, or a combination of both, covering normal, short-circuit, open-circuit, and poor contact samples, and incorporating operating conditions such as light load, rated load, heavy load, three-phase imbalance, distributed photovoltaic output disturbance, and electric vehicle load disturbance. Fault samples are sampled hierarchically according to fault feeder, fault section, fault location, fault mechanism, phase set, and fault parameters, and the same fault event is not repeatedly included in the sample set.

[0073] Step S4: Determine whether the fault determination condition is met based on the diagnostic probability of the fault state. If not, output the normal state. If it is met, determine the sampling interval. Select low time resolution snapshots of three-phase measurements on the node side before and after the sampling interval, and fuse the selected snapshots with the topological location features to generate the initial node representation.

[0074] Specifically, the fault determination criteria include a fault state probability greater than a preset threshold. The preset threshold can be 0.5, or it can be set by a fixed empirical value or determined by the validation samples according to the constraints of recall, precision or F1 value.

[0075] If the fault state probability does not meet the fault determination conditions, the normal state is output, and the affected feeder, fault mechanism, and fault phase set are set to empty or inapplicable.

[0076] If the fault state probability meets the fault determination criteria, the sampling interval is determined based on the 15-minute sampling interval containing the fault occurrence time or the fault determination time. The selected node-side three-phase measurement low-time-resolution snapshots include snapshots of the three sampling sections before the fault and the snapshot of the first available sampling section after the fault. When there is a delay in the time when the root node time-series diagnostic model outputs the diagnostic probability, the sampling interval to which the fault belongs is determined by backtracking from the root node disturbance start point.

[0077] Before entering the node time encoder, the low temporal resolution snapshots at the node side merge the phase dimension and feature dimension to form the node features for each snapshot section of each node. For each node, the system extracts a short historical snapshot sequence and performs time encoding using a bidirectional gated recurrent unit (bidirectional GRU) with parameters shared by all nodes. Subsequently, the short historical temporal representation of the node is fused with the topological location features to obtain the initial representation of the node.

[0078] Step S5: Based on the feeder topology, perform topology message propagation on the initial representation of the node to obtain the topology-aware node representation.

[0079] Specifically, in the low-voltage distribution area topology with self-loops, topology-aware node representations are obtained through a topology message propagation network. Topology message propagation is implemented using a graph neural network, specifically a mean-aggregation type GraphSAGE graph neural network. Each layer performs mean aggregation on the node's neighborhood representation and then linearly transforms the fused representation with the aggregated neighborhood representation.

[0080] Furthermore, node graph temporal coding, static position coding, and topology propagation can be further implemented according to the following model: The low temporal resolution node graph input is first merged with the phase dimension and the feature dimension to obtain: (56).

[0081] For each node The model extracts its four-step snapshot sequence. Time encoding is performed using a bidirectional GRU with shared parameters across all nodes. The hidden size of each direction of the bidirectional GRU is 32, and the concatenated result is: (57).

[0082] To enhance the node representation's awareness of the transformer topology location, the model adds static location encoding to each node: (58), In equation (58), Embedding 16-dimensional node numbers, for One-hot encoding For normalized topological distance, Indicates whether it is a feeder end node.

[0083] The node is initially represented as: (59).

[0084] Therefore, the initial node representation satisfies The first layer of GraphSAGE maps it to 128 dimensions. Subsequently, two layers of mean GraphSAGE propagate node anomaly information along the low-voltage substation topology. After adding a self-loop, the neighborhood aggregation of the l-th layer is: (60).

[0085] The node is updated to: (61).

[0086] With both the hidden and output dimensions of the two GraphSAGE layers set to 128, the final topology-aware node representation is obtained: (62), (63).

[0087] Step S6: For each candidate feeder, generate a candidate feeder representation based on the topology-aware node representation of the nodes contained in the candidate feeder, generate a non-candidate feeder representation based on the node representations of other feeders besides the candidate feeder, and construct a comparative difference representation of the candidate feeder relative to other feeders based on the difference between the candidate feeder representation and the non-candidate feeder representation.

[0088] Specifically, for each candidate feeder, attention-weighted pooling is performed on the topology-aware node representations in the set of feeder nodes to obtain the candidate feeder representation.

[0089] Furthermore, candidate feeder pooling, contrast difference characterization, and multi-rate fusion localization can be further implemented according to the following model: After graph encoding, the model performs attention pooling on the node set of each candidate feed. Let Indicates feeder The set of nodes. For the feeder On the node Attention weights are defined as follows: (64), The corresponding feeder is characterized as follows: (65), (66).

[0090] To highlight the anomalous differences between candidate feeds and other feeds, the model first calculates the average representation of non-candidate feeds: (67) Then, candidate feeder fusion vectors are constructed: (68) (69).

[0091] Step S7: Integrate the candidate feeder characterization, the comparison difference characterization, and the root node disturbance characterization to output the fault location probability of each candidate feeder.

[0092] Specifically, the three feeders share the same feeder positioning MLP: (70) After concatenating the logit values ​​of the three candidate feeders, we obtain: (71), And output feeder probability via softmax: (72).

[0093] Step S8: Based on the fault state, fault mechanism, fault phase set and fault location probability, obtain the gray-domain situational awareness result.

[0094] Specifically, the training samples for the root node timing diagnosis model can be constructed from a preset three-phase low-voltage distribution area sample model, historical measured samples, or a combination of both, covering normal, short-circuit, open-circuit, and poor contact samples, and incorporating operating conditions such as light load, rated load, heavy load, three-phase imbalance, distributed photovoltaic power output disturbance, and electric vehicle load disturbance. Fault samples are sampled hierarchically according to fault feeder, fault section, fault location, fault mechanism, phase set, and fault parameters, and the same fault event is not repeatedly included in the sample set.

[0095] The training objectives include fault detection loss, fault mechanism identification loss, phase set identification loss, and feeder location loss, plus a legal combination constraint loss. Fault detection loss can use binary cross-entropy, while fault mechanism identification, phase set identification, and feeder location losses can use multi-class cross-entropy. These losses are weighted and summed according to preset weights. For contact failures, the system sets legal combination constraints, allowing only single-phase sets A, B, or C for this type of fault. If the model outputs multiple-phase sets such as AB, BC, AC, or ABC for contact failure samples, the legal combination constraint generates a penalty term.

[0096] The multi-task loss function, legal combination constraints, and online decoding rules can be further defined as follows: The training objective consists of four task losses and one legal combination constraint: (73), In this embodiment, the following is taken: (74) Fault detection loss is binary cross-entropy: (75).

[0097] For healthy samples, only fault detection loss is calculated, excluding mechanism, phase, and feeder location losses. For faulty samples, the mechanism, phase, and feeder location losses are as follows: (76) (77), (79) The losses mentioned above represent the average value of valid samples within a batch; healthy samples are only included. To suppress the output of multiphase phase sets due to poor contact, the legal combination constraint is defined as follows: (80), In equation (80), , The same applies to the other phase components.

[0098] During online decoding, the fault detection threshold is fixed at [value]. : (81).

[0099] During online decoding, if the signal is determined to be normal, the output feeder field will be N / A. If the signal is determined to be faulty, then... , and Let the mechanism, phase set, and feeder tag set be represented respectively, then the decoding is: (82), (83), (84), The final results are presented in a structured format: (85), For example, when the model outputs , , At that time, the diagnosis result was described as a short-circuit fault in phase AB of feeder F2. Due to the grounding involvement being combined, this result also covers the combined short-circuit faults of AB and ABG.

[0100] During online operation, the system first performs a rapid diagnosis on the cycle-level sequence of the sliding root node. If the fault detection probability does not meet the fault determination criteria, a normal state is output, and the affected feeder fields are marked as inapplicable. If the fault detection probability is greater than a preset threshold, the root node disturbance characterization and the fault occurrence time are cached, and the rapid diagnosis results of the fault mechanism and fault phase set are output simultaneously. When the first 15-minute snapshot on the node side after the fault arrives, the system constructs a graph time series input from the historical snapshots before and after the fault, and performs multi-rate feeder localization in conjunction with the cached root node disturbance characterization, outputting complete gray-area situational awareness results.

[0101] Example 2: This embodiment provides a low-voltage distribution network gray-domain situational awareness system, characterized in that it is used to implement the method described in Embodiment 1, including: The data acquisition module is used to acquire the feeder topology, the high time resolution sequence of the three-phase electrical quantities at the root node, and the low time resolution snapshot sequence of the three-phase measurements at the node side. The topology feature construction module generates the node-feedline affiliation relationship, node adjacency relationship, topological position features of nodes relative to the root node, and all candidate feeders based on the feeder topology. The root node rapid diagnosis module is used to encode the three-phase electrical quantities of the root node with high time resolution, and output the root node disturbance characterization, fault state, fault mechanism and diagnostic probability of fault phase set. The node graph timing coding and topology propagation module is used to determine the sampling interval and select low temporal resolution snapshots of three-phase measurements on the node side before and after it. These snapshots are then fused with topological location features to generate an initial node representation. Based on the feeder topology, the initial node representation is propagated with topological messages to output a topology-aware node representation. The multi-rate feeder positioning module generates candidate feeder representations based on the topology-aware node representations of the nodes contained in the candidate feeders and generates non-candidate feeder representations based on the node representations contained in the non-candidate feeders. Based on the differences between the candidate feeder representations and the non-candidate feeder representations, it constructs a comparative difference representation of each candidate feeder relative to other feeders. It then fuses the candidate feeder representations, comparative difference representations, and root node disturbance representations to output the fault location probability. The result decoding module outputs the results based on the fault state, fault mechanism, fault phase set, and fault location probability.

[0102] Example 3: like Figure 4 As shown, this embodiment provides an electronic device, which may include: at least one processor, at least one network interface, a user interface, a memory, and at least one communication bus.

[0103] The communication bus can be used to enable communication between the various components mentioned above.

[0104] The user interface may include buttons, and optional user interfaces may also include standard wired interfaces and wireless interfaces.

[0105] The network interface may include, but is not limited to, Bluetooth modules, NFC modules, Wi-Fi modules, etc.

[0106] The processor may include one or more processing cores. It connects various parts of the electronic device via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in memory, and accessing data stored in memory to perform various functions and process data. Optionally, the processor can be implemented using at least one hardware form of DSP, FPGA, or PLA. The processor may integrate one or more of the following: CPU, GPU, and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor.

[0107] The memory may include RAM or ROM. Optionally, the memory may include a non-transitory computer-readable medium. The memory can be used to store instructions, programs, code, code sets, or instruction sets. The memory may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor. The memory, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a sensing application program. The processor can be used to call the sensing application program stored in the memory and execute the steps of the low-voltage distribution network gray area situational awareness method mentioned in the foregoing embodiments.

[0108] Example 4: This embodiment provides a computer-readable storage medium storing instructions that, when executed on a computer or processor, cause the computer or processor to perform the above-described instructions. Figure 1 One or more steps in the illustrated embodiment. If the constituent modules of the above-described electronic device are implemented as software functional units and sold or used as independent products, they can be stored in the computer-readable storage medium.

[0109] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this specification are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., Digital Versatile Discs (DVDs)), or semiconductor media (e.g., Solid State Disks (SSDs)).

[0110] Those skilled in the art will understand that all or part of the processes in the method of Embodiment 1 described above can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks. Unless otherwise specified, the technical features of this embodiment and the implementation scheme can be combined arbitrarily.

[0111] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps can be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0112] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0113] The foregoing description is merely an exemplary embodiment of the present invention and should not be construed as limiting the scope of the invention. Any equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of embodiments of the invention upon considering the specification and practicing the disclosure herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include knowledge or means in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of the invention are defined by the claims.

Claims

1. A method for gray-domain situational awareness in low-voltage distribution networks, characterized in that, Including the following steps: Acquire the feeder topology of the low-voltage distribution area, the high time resolution sequence of the three-phase electrical quantities at the root node, and the low time resolution snapshot sequence of the three-phase measurements at the node side; The relationship between nodes and feeders, the adjacency relationship between nodes, the topological position characteristics of nodes relative to the root node, and all candidate feeders are determined based on the feeder topology. The root node time-series diagnostic model is used to encode the three-phase electrical quantities of the root node with high time resolution, obtain the root node disturbance characterization, and output the diagnostic probability of fault state, fault mechanism and fault phase set. The fault determination condition is determined based on the diagnostic probability of the fault state. If the condition is not met, the normal state is output. If the condition is met, the sampling interval is determined. Before and after the interval, low time resolution snapshots of three-phase measurements on the node side are selected and fused with the topological location features to generate the initial node characterization. Based on the feeder topology, the initial node representation is propagated through topology messages to obtain the topology-aware node representation. For each candidate feeder, a candidate feeder representation is generated based on the topology-aware node representation of the nodes contained in the candidate feeder, a non-candidate feeder representation is generated based on the node representation of the non-candidate feeders, and a contrast difference representation relative to other feeders is constructed based on the difference between the candidate feeder representation and the non-candidate feeder representation. By integrating candidate feeder characterization, comparative difference characterization, and root node disturbance characterization, the fault location probability of each candidate feeder is output. Based on the fault status, fault mechanism, fault phase set, and fault location probability, the gray-domain situational awareness results are obtained.

2. The method according to claim 1, characterized in that: The training objective of the root node timing diagnosis model includes a legal combination constraint loss, which is used to constrain the fault phase set corresponding to poor contact faults to be a single phase set.

3. The method according to claim 1, characterized in that, When the root node time-series diagnostic model outputs the diagnostic probability of the fault phase set: Ground faults and their corresponding non-ground faults are assigned to the same phase set label.

4. The method according to claim 1, characterized in that: The topology message propagation is implemented using a graph neural network; The graph neural network is a mean-aggregation type graph neural network. Each layer performs mean aggregation on the representation of the node's neighborhood and then performs a linear transformation after fusing its own representation with the neighborhood aggregated representation.

5. The method according to claim 1, characterized in that: The candidate feeder representation is obtained by performing attention-weighted pooling on the topology-aware node representation of the nodes contained in the candidate feeder. The non-candidate feeder characterization is obtained by aggregating the node characterizations of other feeders besides the candidate feeders; The contrast difference is characterized as the difference between the candidate feed line characterization and the non-candidate feed line characterization.

6. The method according to claim 1, characterized in that, When the fault determination conditions are met: The selected node-side three-phase measurement low-time-resolution snapshots include snapshots of multiple sampling sections before the fault and snapshots of the first available sampling section after the fault. When there is a delay in the time when the root node time-series diagnostic model outputs the diagnostic probability, the sampling interval to which the fault belongs is determined by backtracking from the root node disturbance start point.

7. The method according to claim 1, characterized in that: The root node temporal diagnostic model is a residual temporal convolutional network; The residual temporal convolutional network includes multi-layer one-dimensional dilated convolutional residual blocks, normalization units, nonlinear activation units, random deactivation units, and residual connections.

8. The method according to claim 1, characterized in that: The topological location features include at least one of node number embedding, feeder affiliation encoding, topological distance from node to root node, and feeder end marker.

9. The method according to claim 1, characterized in that: The phase current in the low time resolution snapshot sequence of the three-phase measurements on the node side is used to characterize the operating status of the node load side or the node injection side, and the phase current is not equivalent to the branch current on the power supply path from the root node to the node.

10. The method according to claim 1, characterized in that: The gray-domain situational awareness refers to the structured identification of fault status, affected feeders, fault mechanisms, and fault phase sets in low-voltage distribution areas, under the condition that the root node can obtain rapid disturbance measurements while the node side can only obtain low temporal resolution measurements.

11. The method according to claim 1, characterized in that: The fault determination criteria include a fault state probability greater than a preset threshold. The preset threshold is set by a fixed empirical value or determined by the validation samples according to the constraints of recall, precision or F1 value.

12. The method according to claim 1, characterized in that: The sampling interval of the high time resolution sequence of the three-phase electrical quantities at the root node is 20 milliseconds, and a single rapid diagnostic sample is extracted from a preset time window before and after the fault. The node-side three-phase measurement low-time-resolution snapshot sequence consists of multiple 15-minute sampling sections.

13. The method according to claim 1, characterized in that: The phase voltage, phase current, active power, and reactive power in the low time resolution input at the root node are obtained by averaging the high time resolution sequence of the three-phase electrical quantities of the root node within a preset short time window near the corresponding sampling time.

14. A gray-domain situational awareness system for low-voltage distribution networks, characterized in that, For implementing the method as claimed in any one of claims 1 to 13, comprising: The data acquisition module is used to acquire the feeder topology, the high time resolution sequence of the three-phase electrical quantities at the root node, and the low time resolution snapshot sequence of the three-phase measurements at the node side. The topology feature construction module generates the node-feedline affiliation relationship, node adjacency relationship, topological position features of nodes relative to the root node, and all candidate feeders based on the feeder topology. The root node rapid diagnosis module is used to encode the three-phase electrical quantities of the root node with high time resolution, and output the root node disturbance characterization, fault state, fault mechanism and diagnostic probability of fault phase set. The node graph timing coding and topology propagation module is used to determine the sampling interval and select low temporal resolution snapshots of three-phase measurements on the node side before and after it. These snapshots are then fused with topological location features to generate an initial node representation. Based on the feeder topology, the initial node representation is propagated with topological messages to output a topology-aware node representation. The multi-rate feeder positioning module generates candidate feeder representations based on the topology-aware node representations of the nodes contained in the candidate feeders and generates non-candidate feeder representations based on the node representations contained in the non-candidate feeders. Based on the differences between the candidate feeder representations and the non-candidate feeder representations, it constructs a comparative difference representation of each candidate feeder relative to other feeders. It then fuses the candidate feeder representations, comparative difference representations, and root node disturbance representations to output the fault location probability. The result decoding module outputs the results based on the fault state, fault mechanism, fault phase set, and fault location probability.

15. An electronic device comprising a memory, a processor, and a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 13.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 13.

Images