Power distribution branch box monitoring system and method based on fault indicators

Abstract

The application relates to the technical field of electric power, and particularly discloses a power distribution branch box monitoring system and method based on a fault indicator, which can compensate and correct through historical load data and operation instruction information, can make up for the short-time limitation of real-time data, can make the dynamic topology structure diagram at the first moment more conform to the actual power grid operation state, can guarantee that the topology structure of the dynamic topology structure diagram is dynamically updated with the change of the power grid, the corrected dynamic topology diagram can accurately reflect the real connection relationship and electrical distance of the branch box, the quantization result of the branch correlation matrix is more conform to the correlation degree of the actual power grid, and through the dynamic construction logic of the dynamic topology structure diagram, the dynamic change process of the power distribution network topology diagram can be adapted, the effect of dynamically monitoring the fault of the power distribution branch box is achieved, and the purpose of reducing the false alarm rate is achieved.

Description

Technical Field

[0001] This invention relates to the field of power technology, and in particular to a monitoring system and method for distribution branch boxes based on fault indicators. Background Technology

[0002] A fault indicator is an electromagnetic induction device that detects the presence of short-circuit current. Installed along power distribution lines, fault indicators activate when a fault occurs and short-circuit current flows, displaying a red fault sign. A line inspection reveals that all fault indicators up to the fault point on the power supply side show red signs, while those after the fault point do not. This indicates the fault point lies between the last red-signed point and the first non-red-signed point. Distribution branch boxes, as critical nodes in the distribution network, perform vital functions such as power distribution, line protection, and fault isolation. Their operational status directly impacts the reliability and security of the power supply. With the advancement of smart grid construction, the demand for real-time monitoring of distribution branch boxes is increasingly urgent. Fault indicators, as core components for monitoring line current, voltage, and fault status, are widely used in distribution branch box monitoring systems.

[0003] Currently, monitoring of distribution substations largely relies on the traditional centralized networking model. This involves a main controller deployed in the substation or monitoring center communicating point-to-point with fault indicators in each substation to collect data and issue commands. However, this networking method has significant limitations: firstly, when the distribution network topology is complex or the number of substations is large, the communication load on the main controller increases dramatically, easily leading to data transmission delays and packet loss, resulting in decreased real-time monitoring performance; secondly, centralized networking is highly dependent on the main controller, and if the main controller fails, the entire monitoring system will be paralyzed, resulting in insufficient reliability.

[0004] In view of this, Chinese patent application CN 120824929 A discloses a method for monitoring electrical variable regulation in distribution boxes. This method combines distributed high-frequency acquisition with dynamic gradient analysis to capture subtle changes in electrical variables in real time. By comparing with historical benchmarks, it identifies potential anomalies in advance, improving the sensitivity and accuracy of anomaly detection from the source. At the same time, based on the spatiotemporal correlation analysis of the distribution network topology, it integrates scattered anomaly nodes into a correlation circle with clear boundaries through quantitative calculation of temporal concentration and spatial distribution density. This breaks through the limitations of isolated analysis and accurately delineates the propagation path and impact range of faults. Combined with the fault judgment mechanism based on branch topology and dynamic characteristics of electrical variables, it achieves a scientific distinction between local and overall power grids, improving the anomaly response speed, fault location accuracy, and processing efficiency of the distribution network.

[0005] In the above scheme, the spatiotemporal correlation analysis based on the distribution network topology map requires retrieving the distribution network topology map containing the node connections and spatial locations of distribution branch boxes from the distribution network management system. These distribution network topology maps are pre-configured, and the physical connection relationships of the distribution branch boxes are pre-entered. If line maintenance or the addition of new distribution branch boxes occurs, the distribution network topology map needs to be reconfigured or adjusted. Thus, the above scheme cannot adapt to the dynamic changes of the distribution network and is difficult to dynamically monitor the faults of distribution branch boxes. At the same time, during the dynamic operation of the distribution network, due to the correlation between distribution branch boxes such as load coupling and fault propagation, the fault data of a single node is easily interfered with. That is, the strong electromagnetic fields of equipment such as transformers and high-voltage cables can electromagnetically couple with the acquisition sensors of the fault indicator, causing the sensors to sense additional stray electromagnetic signals, which are superimposed on the actual electrical parameters, resulting in fluctuations in the effective values ​​of current and voltage. If electrical parameter data is transmitted via carrier wave or wireless communication, electromagnetic noise (such as high-frequency harmonics and radio frequency signals) can interfere with the communication link, causing distortion of the electrical parameter data during transmission. For example, the sampled value of the current mutation rate may jump due to signal interference. Various types of electromagnetic interference can directly affect the acquisition and transmission of electrical parameters, leading to parameter fluctuation errors and a high risk of false alarms.

[0006] Therefore, there is an urgent need for a distribution branch box monitoring system and method based on fault indicators that can adapt to dynamic changes in the distribution network topology, dynamically monitor faults in distribution branch boxes, and reduce false alarm rates. Summary of the Invention

[0007] This invention provides a power distribution branch box monitoring system and method based on fault indicators, which can adapt to dynamic changes in the power distribution network topology, dynamically monitor faults in power distribution branch boxes, and reduce false alarm rate.

[0008] To solve the above-mentioned technical problems, this application provides the following technical solution:

[0009] The monitoring method for distribution branch boxes based on fault indicators includes the following steps:

[0010] Step 1: Collect electrical parameter data of the branch where the fault indicator is located in real time to obtain initial monitoring data;

[0011] Step 2: Determine whether the initial monitoring data is within the preset normal range and whether the fluctuation amplitude of the initial monitoring data collected for a preset number of consecutive times is less than the preset fluctuation threshold. Also determine whether adjacent nodes meet the same requirements. If so, standardize the initial monitoring data to obtain standardized monitoring data.

[0012] Step 3: Exchange standardized monitoring data between each fault indicator through the neighbor discovery protocol, and extract neighbor association feature data based on the standardized monitoring data received by each fault indicator;

[0013] Step 4: Process the adjacent association feature data through a graph theory-based dynamic topology construction algorithm. Using the fault indicator as a node, the edge weights obtained from the adjacent association feature data between nodes generate the initial dynamic topology diagram of the distribution branch box.

[0014] Step 5: Obtain historical load data and operation instruction information for each distribution branch box, and then compensate and correct the dynamic topology diagram at the initial moment based on the historical load data and operation instruction information for each distribution branch box to obtain the dynamic topology diagram at the first moment.

[0015] Step 6: Obtain the topology connection relationship and electrical distance from the dynamic topology diagram at the first moment. Based on the topology connection relationship and electrical distance, quantify the correlation coefficient between each distribution branch box through the branch correlation degree calculation model to obtain the branch correlation degree matrix.

[0016] Step 7: Obtain standardized monitoring data at the second moment. Extract fault feature vectors for each distribution branch box based on the standardized monitoring data at the second moment and the dynamic topology diagram at the first moment. Use a multi-dimensional fault decision algorithm to fuse and analyze the branch correlation matrix and fault feature vectors. Combine the topological connection relationship of nodes in the dynamic topology diagram to determine whether the alarm signal of the current fault indicator is an interference signal caused by the fault of an adjacent node. If the correlation coefficient is less than the feature matching threshold, it is determined to be a fault of the current node and a fault alarm message is generated. If the correlation coefficient is greater than or equal to the feature matching threshold, it is determined to be an interference signal caused by the fault of an adjacent node and the alarm is blocked. Then, query and verify.

[0017] The basic principle and beneficial effects of this scheme are as follows: The dynamic topology diagram in this scheme is constructed from real-time collected electrical parameter data. Compared with directly retrieving the distribution network topology diagram from the distribution network management system, the dynamic topology diagram has dynamic characteristics and can reflect the topological connection relationship between distribution branch boxes in real time. This makes the subsequent correlation analysis more in line with the actual distribution network status. Even if the distribution network topology diagram is reconfigured or adjusted, it can adapt to the dynamic changes of the distribution network, thereby dynamically monitoring the faults of distribution branch boxes.

[0018] Meanwhile, a two-step verification process filters out outliers caused by sensor anomalies and transient electromagnetic interference, and excludes unstable data with large short-term fluctuations. The consistency of data from adjacent nodes is verified simultaneously, providing a reliable data source for the subsequent construction of the dynamic topology diagram. If the initial monitoring data fails the two-step verification, the subsequent standardization and topology construction processes will be suspended to avoid generating incorrect fault feature vectors based on invalid data, reduce false alarms caused by data distortion, and ensure the accuracy of the dynamic topology diagram.

[0019] Then, by compensating and correcting using historical load data and operation command information, the short-term limitations of real-time data can be overcome, making the dynamic topology diagram at the first moment more consistent with the actual power grid operation status. This ensures that the topology of the dynamic topology diagram is dynamically updated as the power grid changes. The corrected dynamic topology diagram can accurately reflect the actual connection relationship and electrical distance of the branch boxes, making the quantitative result of the branch correlation matrix more consistent with the correlation degree of the actual power grid. When making fault decisions, the correlation coefficient obtained based on the accurate dynamic topology diagram can more accurately distinguish between interference signals of adjacent nodes and the actual fault of the current node, avoiding misjudgment of correlation degree caused by errors in the dynamic topology diagram, reducing the situation of misjudging the actual fault as an interference signal (or vice versa), and further reducing the false alarm rate.

[0020] Furthermore, compared to analyzing local data collected from a single fault indicator, which ignores the correlations such as load coupling and fault propagation between distribution branch boxes, this solution considers the electrical correlation characteristics of the distribution branch box and its adjacent distribution branch boxes, such as the historical load fluctuation correlation and physical connection distance. This can characterize the degree of electrical coupling between different distribution branch boxes. At the same time, a multi-dimensional fault identification mechanism is constructed. When a fault occurs in a distribution branch box and causes a fault in an adjacent distribution branch box, the correlation coefficient is used to determine whether the fault is caused by interference from the adjacent node fault. If the correlation coefficient is greater than or equal to the feature matching threshold, the interference alarm is directly blocked. Only when the correlation coefficient is less than the feature matching threshold is it determined to be a real fault of the current node and an alarm is triggered. Compared to relying on a single node for judgment, this avoids the false alarm triggered by the associated fault of the adjacent distribution branch box, reduces the false alarm rate, ensures that the real fault signal is not drowned out, shortens the fault location and processing time, and ensures the continuity of power supply in the distribution network.

[0021] In summary, this invention, through the dynamic construction logic of the dynamic topology diagram, can adapt to the dynamic changes in the distribution network topology diagram; by integrating the branch correlation matrix and fault feature vector through a multi-dimensional fault decision algorithm, and combining the topological connection relationship of nodes in the dynamic topology diagram, it achieves the effect of dynamically monitoring the faults of distribution branch boxes and reduces the false alarm rate.

[0022] Furthermore, in step 7, when the multi-dimensional fault decision algorithm determines whether the alarm signal is an interference signal caused by a fault in an adjacent node, it includes: extracting the fault feature vector of the current fault indicator and the real-time fault feature vector of the distribution branch box ranked first in the correlation coefficient, and calculating the feature similarity; if the feature similarity is greater than or equal to the feature matching threshold, it is determined to be an interference signal caused by a fault in an adjacent node and the alarm is blocked; if the feature similarity is less than the feature matching threshold, the determination is made by combining the comparison result of the correlation coefficient and the feature matching threshold.

[0023] The beneficial effects are as follows: by adding feature similarity comparison in fault decision-making, the logic for eliminating interference signals is refined; based on correlation coefficient analysis and combined with the matching degree judgment of fault feature vectors, it can effectively distinguish special scenarios with high correlation but mismatched fault features, such as adjacent distribution branch boxes with similar historical loads but different current fault types, avoiding misjudgment due to a single correlation coefficient, and improving the accuracy and reliability of fault judgment.

[0024] Furthermore, in step 7, after generating fault alarm information or masking alarms, the judgment result is fed back in real time to the fault indicator of the current fault indicator and the fault indicator of the power distribution branch box ranked first in the correlation coefficient. After receiving the feedback, each fault indicator updates the adjacent correlation feature data stored locally.

[0025] The beneficial effects are as follows: self-optimization is achieved by providing real-time feedback on the judgment results and updating the adjacent related feature data. The feedback information enables adjacent distribution branch boxes to dynamically perceive the fault propagation path. The updated adjacent related feature data improves the timeliness of subsequent topology construction and correlation coefficient calculation. At the same time, the sharing of fault node identification and shielding reasons enhances the collaborative judgment capability, reduces the risk of false alarms caused by data lag, and improves the dynamic adaptability of monitoring.

[0026] Furthermore, in step 7, when generating fault alarm information or blocking alarms, a short-term prediction model is constructed using the branch correlation degree matrix and historical fault feature vectors. Based on the real-time monitoring data of each distribution branch box, the probability value of the distribution branch box with the highest correlation degree coefficient in the future preset time period is predicted. If the probability value is greater than the warning threshold, a fault warning prompt is output and synchronized to the corresponding branch box fault indicator.

[0027] The beneficial effects are as follows: real-time prediction is carried out simultaneously in the fault determination stage, realizing the integration of monitoring, judgment and early warning; the short-term prediction model is built based on the correlation matrix and historical data, without the need for additional data collection, and can predict the fault risk of adjacent distribution branch boxes in advance, thereby improving the accuracy of fault early warning.

[0028] Furthermore, in step 3, when exchanging standardized monitoring data through the neighbor discovery protocol, if the fault indicator detects that the number of signal transmission failures exceeds the preset number, it automatically switches to low-power communication mode, uses frequency hopping spread spectrum to retry the interaction, and writes the signal transmission failure status and switching mode information into the adjacent association feature data; in step 7, when judging the fault, the historical association coefficient stored locally is used to assist in the judgment, combining the signal transmission failure status and the node connection redundancy in the dynamic topology diagram.

[0029] The beneficial effects are as follows: when the signal cannot be transmitted, low-power frequency hopping spread spectrum can improve the signal transmission success rate and avoid monitoring failure due to communication interruption; combined with historical correlation coefficient to assist in the judgment, it can maintain the continuity of fault judgment when the signal interaction is not smooth and reduce the risk of missed judgment caused by poor communication.

[0030] Furthermore, in step 3, when exchanging standardized monitoring data through the neighbor discovery protocol, if the signal transmission power of the fault indicator is lower than a preset power threshold, a signal forwarding request is sent to the neighbor fault indicator with which a communication connection has been established. The forwarding request includes the standardized monitoring data to be sent and the target node identifier. After receiving the forwarding request, the neighbor fault indicator selects whether to forward the data based on its own communication load status. If the forwarding is successful, the forwarding record is synchronized to the fault indicator that initiated the request.

[0031] The beneficial effects are as follows: when the signal is weak and cannot be transmitted, the existing neighbor communication relationship can be used to realize signal relay without the need for additional equipment deployment, thus expanding the signal transmission coverage. At the same time, the forwarding record is integrated into the topology construction, so that the dynamic topology can reflect the differences in the communication capabilities of nodes. In subsequent fault judgment, the data exchanged between neighboring nodes with stable communication can be selected first, thereby improving the communication reliability and data integrity of the monitoring.

[0032] Furthermore, in step 3, when exchanging standardized monitoring data through the neighbor discovery protocol, the fault indicator synchronously detects the zero-crossing points of the current and voltage it collects. Taking the zero-crossing point of the current and voltage as the communication trigger time, it packages and sends the standardized monitoring data and the zero-crossing time marker information. After receiving the data, the neighbor fault indicator first verifies the time difference between the zero-crossing time marker information and the local zero-crossing detection result. If the time difference is less than the synchronization threshold, the data is confirmed to be valid and used to extract adjacent related feature data; otherwise, it requests retransmission.

[0033] The beneficial effects are as follows: the zero-crossing point of current and voltage is the instantaneous zero point of the grid voltage / current waveform. At this moment, the electromagnetic interference of the grid is weakest. Using this as the communication trigger moment can reduce the superposition interference of clutter on the transmitted signal in a strong electromagnetic environment, reduce the data transmission error rate, and ensure the integrity and originality of standardized monitoring data. In addition, the data carries zero-crossing time marker information. Neighboring nodes can identify invalid data that has been interfered with by verifying the consistency between the marker and the local zero-crossing detection result. If the time marker is distorted or the data is tampered with due to electromagnetic interference during data transmission, it will be rejected because the time difference exceeds the synchronization threshold, thus preventing erroneous data from entering the associated feature extraction stage and ensuring the accuracy of subsequent topology construction and fault diagnosis.

[0034] Furthermore, in step 3, each fault indicator uses carrier communication to exchange standardized monitoring data, and detects the peak and trough positions of the carrier signal in real time. When the peak amplitude is greater than a preset peak threshold or the trough amplitude is less than a preset trough threshold, the standardized monitoring data is sent, and peak and trough detection marker information is carried in the standardized monitoring data. After receiving the standardized monitoring data, the neighboring fault indicator verifies whether the carrier signal amplitude at the time of reception is in the peak and trough interval based on the peak and trough detection marker information. If so, the standardized monitoring data is confirmed to be valid and used to extract adjacent associated feature data. If not, the standardized monitoring data is discarded and retransmission is triggered.

[0035] The beneficial effects are as follows: by using the peak and trough communication logic of carrier communication, the anti-interference capability of carrier communication is improved; by selecting peak and trough periods to send data, the signal amplitude advantage can be used to reduce transmission attenuation and improve the data reception success rate; combined with the amplitude verification mechanism, interference data in non-peak and trough periods can be effectively filtered out, reducing the transmission error rate, adapting to the complex carrier communication environment of power distribution networks, and ensuring the stable interaction of standardized monitoring data. Attached Figure Description

[0036] Figure 1 This is a flowchart of Example 1 of the power distribution branch box monitoring method based on fault indicators.

[0037] Figure 2 This is a system block diagram of Embodiment 2 of a power distribution branch box monitoring system based on a fault indicator. Detailed Implementation

[0038] The following detailed description illustrates the specific implementation method:

[0039] Example 1

[0040] A method for monitoring distribution branch boxes based on fault indicators, as shown in the appendix. Figure 1 As shown, six adjacent distribution branch boxes, designated as branch boxes 1-6, are used in a 10kV urban distribution network. Each branch box is equipped with a fault indicator that features current / voltage acquisition, multi-mode communication, and edge computing capabilities. The preset parameters are configured as follows: correlation coefficient ranking preset position is 3, feature matching threshold is 0.7, early warning threshold is 70%, preset future time period is 5 minutes, preset number of signal transmission failures is 3, signal transmission power feature matching threshold is -75dBm, communication load threshold is 60%, zero-crossing synchronization threshold is 5ms, carrier peak feature matching threshold is 2.2V, and carrier trough feature matching threshold is -2.2V. The specific implementation process is as follows:

[0041] Step 1: Collect initial monitoring data. The fault indicator in each branch box collects the electrical parameter data of the branch where the fault indicator is located in real time through the built-in high-precision current transformer and voltage transformer. For example, the effective value of the current collected by branch box 1 is 110A, the effective value of the voltage is 10.3kV, and the current change rate is 3.5A / ms; the effective value of the current collected by branch box 2 is 95A, the effective value of the voltage is 10.4kV, and the current change rate is 2.8A / ms. The other branch boxes collect the corresponding parameters synchronously to form the initial monitoring data.

[0042] Step 2: Standardization Processing. The Z-score standardization algorithm is used to process the initial monitoring data. The processing formula is as follows: ,in, To standardize electrical parameter data, The original electrical parameter data, This is the average value of the parameter collected over the past 24 hours. This represents the standard deviation. For example, the effective current of branch box 1 is 110A, calculated... , The standardized data is Meanwhile, the 300A abnormal current value collected in branch box 3 due to instantaneous sensor failure was removed, and all data were uniformly converted into 64-bit floating-point format to obtain standardized monitoring data.

[0043] Step 3: Neighbor-related feature data extraction and interaction optimization. Each fault indicator uses the IEEE 802.15.4e standard neighbor discovery protocol to exchange standardized monitoring data. When the fault indicator in branch box 4 detects its own signal transmission power as -82dBm, lower than the preset power threshold of -75dBm, it automatically sends a signal forwarding request to branch boxes 3 and 5, which have established communication connections. The request includes its own standardized monitoring data and the identifier of the target node, branch box 6. Branch box 3's communication load is 45%, lower than the load threshold of 60%, so it agrees to forward the signal and synchronizes the forwarding record to branch box 4. During the interaction, the fault indicator in branch box 5 fails to transmit signals three times consecutively, exceeding the preset number of failures. It automatically switches to a low-power communication mode and re-attempts interaction using frequency hopping spread spectrum technology, with a frequency hopping range of 2.4-2.48GHz. Simultaneously, it writes the signal transmission failure status and mode switching information into the neighbor-related feature data.

[0044] All fault indicators synchronously detect the zero-crossing points of the current and voltage they collect. At the moment of voltage zero-crossing, such as when the sinusoidal voltage is 0V, data transmission is triggered, packaging standardized monitoring data with zero-crossing moment marker information. After receiving the data, neighboring fault indicators verify the time difference between the marker information and the local zero-crossing detection result. If the time difference between branch box 1 and branch box 2 is 3ms, which is less than the synchronization threshold of 5ms, the data is confirmed to be valid. When interacting using carrier communication, the peak and trough positions of the carrier signal are detected in real time. When the peak amplitude is detected to be 2.5V, which is greater than the preset peak threshold of 2.2V, data is transmitted, and the data frame carries the peak marker. After receiving the data, the neighbor verifies that the amplitude at the time of reception is in the peak range of 2.2-2.8V, confirms the data validity, and extracts adjacent correlation characteristic data such as communication strength (e.g., the communication strength between branch box 1 and branch box 2 is -68dBm), electrical parameter correlation (e.g., current correlation coefficient 0.82), and signal transmission delay (e.g., 12ms).

[0045] Step 4: Construct a dynamic topology graph. Calculate the fusion value of associated features using the following formula: ,in, The associated feature fusion value is used. A graph-based Kruskal algorithm is employed to process adjacent associated feature data. Each fault indicator is treated as a node, and the associated feature fusion value is used as the edge weight between nodes to generate a dynamic topology graph. For example, all edges between nodes are sorted from largest to smallest weight, and the edges with the largest weights are selected sequentially to connect the corresponding nodes, while avoiding loops. The final connection relationships between nodes in the graph are obtained, such as: branch box 1 is connected to branch box 2 and branch box 3; branch box 2 is connected to branch box 4; and branch box 3 is connected to branch box 5 and branch box 6, thus realizing self-networking of the monitoring system.

[0046] The specific calculation of the associated feature fusion value is as follows:

[0047] A. Calculate parameter correlation based on RMS current, RMS voltage, and current abrupt change rate. Calculate the correlation of RMS current, RMS voltage, and current abrupt change rate for two nodes (e.g., branch boxes i and j). Current RMS correlation:

[0048]

[0049] in, Correlation of current RMS value , The effective value of the current at the two nodes,

[0050] Voltage RMS correlation:

[0051]

[0052] in, For voltage RMS correlation, , These are the effective values ​​of the voltages at the two nodes.

[0053] Current mutation rate correlation:

[0054]

[0055] in, The correlation is with the rate of change of current. , Let be the current abrupt change rate between the two nodes.

[0056] The formula for parameter correlation is:

[0057]

[0058] B. Calculate the associated feature fusion value. At this point, obtain the communication strength, convert the communication strength to a value in the [0,1] interval, and obtain... The formula is as follows:

[0059]

[0060] Then calculate the ratio of the communication delay between the current nodes to the maximum delay within the monitoring range to obtain the delay / maximum delay.

[0061] Assumptions: Branch box 1: RMS current value RMS voltage Current mutation rate Branch box 2: RMS current value RMS voltage Current mutation rate With a normalized communication strength of 0.8 and a delay / maximum delay ratio of 0.2, the calculated value is... , Calculate the fusion value of related features .

[0062] Step 5: Extract fault feature vectors. Extract fault feature vectors for each branch box from the dynamic topology diagram. These feature vectors include fault current amplitude, fault duration, voltage drop amplitude, and phase change information. For example, when a short-circuit fault occurs in branch box 2, the extracted fault feature vector is [185A, 1.8s, 1.3kV, 12°].

[0063] Step 6: Construct the branch correlation matrix. Based on the historical load data, topological connection relationships, and electrical distances of each branch box, the correlation coefficient is quantified using a weighted summation algorithm through the branch correlation calculation model, as shown in the following formula: ),in, Let be the correlation coefficient between the i-th branch box and the J-th branch box. , The average load over the past 30 days for branch boxes i and j (e.g., 85kW for branch box 1 and 90kW for branch box 2). This is the topology connectivity coefficient (e.g., 1 for direct connections and 0.6 for indirect connections). For electrical distance (e.g., 450m between branch box 1 and branch box 2). Let the maximum electrical distance be 1000m. This allows us to calculate the correlation coefficient between branch box 1 and branch box 2 as 0.83, between branch box 1 and branch box 3 as 0.76, between branch box 2 and branch box 4 as 0.81, and so on. If there are N branch boxes, the branch correlation matrix is ​​an N×N matrix, where the row and column indices correspond to two different branch boxes. The matrix element Rij represents the correlation coefficient between the i-th and j-th branch boxes. This hierarchical weighted fusion algorithm, by quantifying the weight ratio of each influencing factor, allows for an objective assessment of the correlation between branch boxes.

[0064] Step 7: Fault Decision and Optimization. A multi-dimensional fault decision algorithm is used to fuse the branch correlation matrix and fault feature vectors, and to make a judgment based on the dynamic topology node connection relationship: the fault feature vector of the branch box 1 that triggered the alarm is extracted, and compared with the real-time fault feature vectors of the top 3 branch boxes 2, 3, and 4 in terms of correlation coefficient. The cosine similarity algorithm is used to calculate the feature similarity. If the similarity is 0.85, which is greater than the feature matching threshold of 0.7, it is determined to be an interference signal caused by the fault of an adjacent node and the alarm is blocked; if the similarity is 0.65, which is less than the feature matching threshold, it is determined to be a fault of the current node and a fault alarm information is generated. Specifically, the multi-dimensional fault decision algorithm uses a dual-threshold fusion decision algorithm, which combines the correlation coefficient and feature similarity to achieve fault identification. The logic is as follows: the cosine similarity algorithm is used to compare the fault feature vectors, and the formula is:

[0065]

[0066] in, This is the fault feature vector of the current node. The fault feature vectors of the associated nodes are then subjected to a double judgment; if the feature similarity is... If the feature matching threshold is used to directly identify it as an interference signal from an associated node, the alarm information will be blocked. Feature matching threshold, proceed to correlation coefficient judgment, if correlation coefficient If the correlation threshold is used to determine if a signal is interference, then... If the correlation threshold is set, the current node is determined to be faulty, and an alarm message is generated.

[0067] After generating an alarm message or disabling an alarm, the judgment result, including the fault node identifier, correlation coefficient, and disabling reason, is fed back in real time to the current fault indicator and the fault indicators of the top 3 branch boxes with the highest correlation coefficient. Each fault indicator updates its locally stored adjacent correlation feature data upon receiving the feedback. Simultaneously, an LSTM short-term prediction model is constructed using the branch correlation matrix and historical fault feature vectors. Inputting the real-time monitoring data of each distribution branch box, the model predicts the probability of a fault occurring in the top 3 branch boxes with the highest correlation coefficient within the next 5 minutes.

[0068] In this embodiment, taking eight distribution branch boxes (denoted as B1-B8) of a 10kV distribution network as an example, the implementation process of constructing an LSTM short-term prediction model is described in detail. The feature matching threshold is set to 0.7, the prediction duration is the next 5 minutes, and the correlation coefficient ranking is the top 3.

[0069] First, data preprocessing is performed. Based on the 8×8 branch correlation matrix obtained in step 6, whose element values ​​range from 0 to 1, the matrix is ​​normalized row by row to ensure that the sum of the elements in each row is 1, eliminating the influence of different dimensional units. For example, the correlation coefficient of row B1 is [1.0, 0.88, 0.82, 0.55, 0.51, 0.48, 0.45, 0.42], which becomes [0.22, 0.19, 0.18, 0.12, 0.11, 0.10, 0.10, 0.08] after normalization, resulting in the standardized correlation matrix. Historical fault feature vectors of each branch box over the past 6 months are collected, with a dimension of 4, including fault current amplitude, fault duration, voltage drop amplitude, and phase change angle. Zero feature vectors for non-fault periods are also added, and the data is organized into a time series dataset with a sampling interval of 1 minute. For example, a certain historical sequence of B2 is: T-10min, [0,0,0,0], no fault (no); T-9min, [0,0,0,0], no fault (no); T-8min, [175A, 2.1s, 1.5kV, 14°], fault (yes). The correlation coefficients of each branch box in the standardized correlation matrix are taken, and the coefficients of the top 3 adjacent nodes are fused with the historical fault feature vector of the corresponding branch box to expand the feature dimension. For example, the fused feature dimension of B1 is: its own fault feature vector (4 dimensions) + the fault feature vectors of the top 3 associated nodes (4×3=12 dimensions) + the top 3 correlation coefficients (3 dimensions), for a total dimension of 19.

[0070] Next, the LSTM model structure is built. In this embodiment, a time series prediction model is built based on the TensorFlow / PyTorch framework. The core structure is as follows: Input layer, with 19 dimensions (fused features), time step = 10 (taking the first 10 minutes of time series data); Hidden layer 1 (LSTM), with 64 neurons, activation function tanh, returning sequences (return_sequences=True); Hidden layer 2 (LSTM), with 32 neurons, activation function tanh, not returning sequences (return_sequences=False); Fully connected layer, with 8 neurons (corresponding to 8 branch bins), activation function ReLU; Output layer, with 3 neurons (the top 3 most relevant branch bins), activation function = sigmoid (outputting 0-1 probability values).

[0071] Next, model training and validation were performed. The preprocessed fused time-series data was divided into a training set (70%), a validation set (20%), and a test set (10%) in a 7:2:1 ratio. The "fault status" label was converted to a fault probability label, with fault = 1 and non-fault = 0. The training parameters were set as follows: the optimizer was Adam, the learning rate was 0.001; the loss function was binary cross-entropy; the number of iterations was 50, and the batch size was 32; an early stopping strategy was used, stopping training if the validation set loss did not decrease for 5 consecutive rounds to avoid overfitting. The model accuracy was validated using the test set. If the actual fault occurrence rate was ≥85% when the predicted fault probability was ≥0.7 (early warning threshold), the model was considered to have passed training.

[0072] Finally, real-time fault probability prediction is performed. Real-time monitoring data (electrical parameters verified and standardized in step 2) is collected from each branch box. Real-time fault feature vectors are extracted and combined with the latest standardized correlation matrix to generate 19-dimensional real-time fusion features. The time step is 10, using the data from the previous 10 minutes. From the branch correlation matrix, the top 3 branch boxes (e.g., B2, B3, B4) with the highest correlation coefficients of the current branch box (e.g., B1) are extracted. The real-time fusion features are input into the trained LSTM model, and the model outputs the fault probability values ​​for B2, B3, and B4 within the next 5 minutes. For example:

[0073] B2 Fault probability = 0.82 (≥ warning threshold 0.7), output fault warning prompt;

[0074] B3 Failure probability = 0.65 (< warning threshold 0.7), no warning;

[0075] B4 Fault probability = 0.78 (≥ warning threshold 0.7), output fault warning prompt.

[0076] The actual occurrence of faults on the day is compared with the predicted results daily, the historical fault feature vector dataset is updated, and the LSTM model is retrained every 7 days, adjusting parameters such as the number of neurons in the network layer and the learning rate to continuously improve the prediction accuracy.

[0077] If the failure probability of branch box 3 is 78%, which is greater than the warning threshold of 70%, a fault warning will be output and synchronized to the fault indicator of branch box 3. If the fault judgment process involves branch box 5, which has failed to send signals, the fault status and the node connection redundancy in the dynamic topology will be considered. If branch box 5 has two communication links, the historical correlation coefficient stored locally (for example, a historical correlation coefficient of 0.78 with branch box 3) will be used to assist in the judgment.

[0078] Example 2

[0079] Based on Embodiment 1, a power distribution branch box monitoring system based on a fault indicator is also provided, as shown in the attached figure. Figure 2 As shown, the system includes six fault indicators, a data interaction module, a topology construction module, a correlation analysis module, and a fault decision module, with each module integrated within the fault indicators. The fault indicators are deployed in six distribution branch boxes of a 10kV distribution network. The data interaction module supports neighbor discovery protocol, frequency hopping spread spectrum communication, and carrier communication, and is used to execute the standardized monitoring data interaction, signal forwarding, zero-crossing communication, and peak-valley communication logic described in Example 1. The topology construction module uses the graph-based Kruskal algorithm for generating the dynamic topology diagram described in Example 1. The correlation analysis module has a built-in branch correlation calculation model, used to execute the correlation coefficient quantification and matrix construction described in Example 1. The fault decision module integrates a multi-dimensional fault decision algorithm and a short-term prediction model, used to execute the fault judgment, real-time feedback, fault early warning, and abnormal state auxiliary judgment functions described in Example 1. Through the collaborative work of each module, the system fully executes the monitoring method described in Example 1, achieving autonomous networking of distribution branch boxes, accurate fault judgment, interference signal shielding, and fault early warning. It is suitable for monitoring needs in complex topologies and harsh communication environments of urban distribution networks.

[0080] Example 3

[0081] Based on Example 1, this example is a further development. Any details not covered herein can be found in Example 1. A method for monitoring distribution branch boxes based on fault indicators includes the following steps:

[0082] Step 1: Collect electrical parameter data of the branch where the fault indicator is located in real time to obtain initial monitoring data;

[0083] Step 2: Determine whether the initial monitoring data is within the preset normal range and whether the fluctuation amplitude of the initial monitoring data collected for a preset number of consecutive times is less than the preset fluctuation threshold. Also, determine whether adjacent nodes meet the same requirements. For example, if the preset normal range is 115-125A and the preset fluctuation threshold is 5%, extract the current data from the initial monitoring data. If the current of branch box A is 120A, 121A, and 119A for three consecutive times, it is within the preset normal range and the fluctuation amplitude is 1.6%, which is less than the preset fluctuation threshold. Then, standardize the initial monitoring data to obtain standardized monitoring data.

[0084] Step 3: Exchange standardized monitoring data between each fault indicator through the neighbor discovery protocol, and extract neighbor association feature data based on the standardized monitoring data received by each fault indicator;

[0085] Step 4: Process the adjacent association feature data through a graph theory-based dynamic topology construction algorithm. Using the fault indicator as a node, the edge weights obtained from the adjacent association feature data between nodes generate the initial dynamic topology diagram of the distribution branch box.

[0086] Step 5: Obtain historical load data and operation instruction information for each distribution branch box. Then, compensate and correct the dynamic topology diagram at the initial moment based on the historical load data and operation instruction information for each distribution branch box to obtain the dynamic topology diagram at the first moment. For example, the distribution network has 3 branch boxes A, B, and C. The initial topology is that AB and BC are directly connected. Retrieve the load records of branch boxes A, B, and C from the distribution network dispatch system database. It is found that the load of branch box A is 40% higher than other time periods from 18:00 to 22:00 every day, while the load of branch box C drops to 30% of the normal level on Sundays. At the same time, receive the instruction issued by the dispatch center to know that branch box B will be under maintenance the next day and needs to be temporarily disconnected from branch box A. At this point, compensation and correction are required. First, adjust the edge weights of nodes A and B based on the load characteristics of branch box A, and increase the correlation weight during peak load periods so that the topology can adapt to the changes in correlation caused by load changes. Then, according to the maintenance instructions of branch box B, temporarily remove the AB connection edge in the initial topology in advance, and add the logic that A is indirectly connected to B through C. After correction, the dynamic topology structure diagram at the first moment is AC and BC directly connected, ensuring that the monitoring logic is normal during maintenance.

[0087] Step 6: Obtain the topology connection relationship and electrical distance from the dynamic topology diagram at the first moment. Based on the topology connection relationship and electrical distance, quantify the correlation coefficient between each distribution branch box through the branch correlation degree calculation model to obtain the branch correlation degree matrix.

[0088] Step 7: Obtain standardized monitoring data at the second moment. Based on the standardized monitoring data at the second moment and the dynamic topology diagram at the first moment, extract the fault feature vectors of each distribution branch box. Use a multi-dimensional fault decision algorithm to fuse and analyze the branch correlation matrix and fault feature vectors. Combine this with the topological connection relationship of nodes in the dynamic topology diagram to determine whether the alarm signal of the current fault indicator is an interference signal caused by a fault in an adjacent node: if the correlation coefficient is less than the feature matching threshold, determine that the current node is faulty and generate a fault alarm message; if the correlation coefficient is greater than or equal to the feature matching threshold, determine that it is an interference signal caused by a fault in an adjacent node and block the alarm, and perform a query verification. For example, if the distribution network has 3 branch boxes A, B, and C, and the topology at the first moment is AC and BC directly connected, with AC=0.85, BC=0.83, and AB=0.6 in the branch correlation matrix and a feature matching threshold of 0.7, first obtain the standardized monitoring data at the second moment, and the fault indicator collects data in real time and processes it through Step 2. After verification and standardization, the data for branch box A are: current 0.5, voltage 0.4, and current mutation rate 0.3; for branch box B, current 1.2, voltage 0.1, and current mutation rate 1.8; and for branch box C, current 0.6, voltage 0.5, and current mutation rate 0.4. Then, fault feature vectors are extracted. Combining this with the topology at the first moment, fault feature vectors for each box are extracted: branch box A is [130A, 0s, 0kV, 0°] (no fault); branch box B is [180A, 2s, 1.2kV, 15°] (suspected fault); and branch box C is [135A, 0s, 0kV, 0°] (no fault). Finally, fusion analysis and fault judgment are performed. First, the branch correlation matrix is ​​called, and then the feature similarity is calculated using the cosine similarity algorithm. For example, the feature similarity between B and C is 0.88, and the feature similarity between B and A is 0.4. Based on the topological connection relationship, the analysis focuses on the relationship between B and its adjacent node C. The correlation coefficient (0.83 ≥ 0.7) and feature similarity (0.88 ≥ 0.7) between B and C are used to determine that the alarm of branch box B is an interference signal caused by the fault of branch box C. The alarm of branch box B is blocked, and the real-time monitoring log and historical fault records of branch box C are retrieved to confirm that there is a short circuit fault in branch box C, thus verifying the accuracy of the judgment.

[0089] The above are merely embodiments of the present invention. The invention is not limited to the fields covered by these embodiments. Commonly known structures and characteristics in the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are able to access all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for monitoring distribution branch boxes based on fault indicators, characterized in that, Includes the following steps: Step 1: Collect electrical parameter data of the branch where the fault indicator is located in real time to obtain initial monitoring data; Step 2: Determine whether the initial monitoring data is within the preset normal range and whether the fluctuation amplitude of the initial monitoring data collected for a preset number of consecutive times is less than the preset fluctuation threshold. Also determine whether adjacent nodes meet the same requirements. If so, standardize the initial monitoring data to obtain standardized monitoring data. Step 3: Exchange standardized monitoring data between each fault indicator through the neighbor discovery protocol, and extract neighbor association feature data based on the standardized monitoring data received by each fault indicator; Step 4: Process the adjacent association feature data through a graph theory-based dynamic topology construction algorithm. Using the fault indicator as a node, the edge weights obtained from the adjacent association feature data between nodes generate the initial dynamic topology diagram of the distribution branch box. Step 5: Obtain historical load data and operation instruction information for each distribution branch box, and then compensate and correct the dynamic topology diagram at the initial moment based on the historical load data and operation instruction information for each distribution branch box to obtain the dynamic topology diagram at the first moment. Step 6: Obtain the topology connection relationship and electrical distance from the dynamic topology diagram at the first moment. Based on the topology connection relationship and electrical distance, quantify the correlation coefficient between each distribution branch box through the branch correlation degree calculation model to obtain the branch correlation degree matrix. Step 7: Obtain standardized monitoring data at the second moment. Extract fault feature vectors for each distribution branch box based on the standardized monitoring data at the second moment and the dynamic topology diagram at the first moment. Use a multi-dimensional fault decision algorithm to fuse and analyze the branch correlation matrix and fault feature vectors. Combine the topological connection relationship of nodes in the dynamic topology diagram to determine whether the alarm signal of the current fault indicator is an interference signal caused by the fault of an adjacent node. If the correlation coefficient is less than the feature matching threshold, it is determined to be a fault of the current node and a fault alarm message is generated. If the correlation coefficient is greater than or equal to the feature matching threshold, it is determined to be an interference signal caused by the fault of an adjacent node and the alarm is blocked. Query and verification are performed. In step 7, when the multi-dimensional fault decision algorithm determines whether the alarm signal is an interference signal caused by the fault of an adjacent node, it includes: extracting the fault feature vector of the current fault indicator and the real-time fault feature vector of the distribution branch box ranked first in the correlation coefficient, and calculating the feature similarity. If the feature similarity is greater than or equal to the feature matching threshold, it is determined to be an interference signal caused by the failure of adjacent nodes and the alarm is blocked; if the feature similarity is less than the feature matching threshold, the determination is made by combining the comparison result of the correlation coefficient and the feature matching threshold.

2. The power distribution branch box monitoring method based on fault indicator according to claim 1, characterized in that, In step 7, after generating fault alarm information or masking alarms, the judgment result is fed back in real time to the fault indicator of the current fault indicator and the fault indicator of the power distribution branch box ranked first in the correlation coefficient. After receiving the feedback, each fault indicator updates the adjacent correlation feature data stored locally.

3. The power distribution branch box monitoring method based on a fault indicator according to claim 2, characterized in that, In step 7, when generating fault alarm information or masking alarms, a short-term prediction model is constructed using the branch correlation matrix and historical fault feature vectors. Based on the real-time monitoring data of each distribution branch box, the probability value of the distribution branch box with the highest correlation coefficient in the future preset time period is predicted. If the probability value is greater than the warning threshold, a fault warning message will be output and synchronized to the corresponding branch box fault indicator.

4. The power distribution branch box monitoring method based on a fault indicator according to claim 3, characterized in that, In step 3, when exchanging standardized monitoring data through the neighbor discovery protocol, if the fault indicator detects that the number of signal transmission failures exceeds the preset number, it automatically switches to low-power communication mode, uses frequency hopping spread spectrum to retry the interaction, and writes the signal transmission failure status and switching mode information into the neighbor association feature data. In step 7, during fault diagnosis, the historical correlation coefficient stored locally is used as the primary auxiliary factor in determining the fault, taking into account the signal transmission failure status and the node connection redundancy in the dynamic topology diagram.

5. The power distribution branch box monitoring method based on a fault indicator according to claim 4, characterized in that, In step 3, when exchanging standardized monitoring data through the neighbor discovery protocol, if the signal transmission power of the fault indicator is lower than the preset power threshold, a signal forwarding request is sent to the neighbor fault indicator with which a communication connection has been established. The forwarding request includes the standardized monitoring data to be sent and the target node identifier. After receiving a forwarding request, the neighbor fault indicator selects whether to forward the request based on its own communication load status. If the forwarding is successful, the forwarding record will be synchronized to the fault indicator that initiated the request.

6. The power distribution branch box monitoring method based on a fault indicator according to claim 5, characterized in that, In step 3, when the standardized monitoring data is exchanged through the neighbor discovery protocol, the fault indicator synchronously detects the zero-crossing point of the current and voltage it collects. The communication trigger time is taken as the zero-crossing point of the current and voltage, and the standardized monitoring data and the zero-crossing time mark information are packaged and sent. After receiving the neighbor fault indicator, it first checks the time difference between the zero-crossing time marker information and the local zero-crossing detection result. If the time difference is less than the synchronization threshold, the data is confirmed to be valid and used to extract the neighbor association feature data; otherwise, it requests retransmission.

7. The power distribution branch box monitoring method based on a fault indicator according to claim 6, characterized in that, In step 3, when each fault indicator uses carrier communication to exchange standardized monitoring data, it detects the peak and trough positions of the carrier signal in real time. When the peak amplitude is greater than the preset peak threshold or the trough amplitude is less than the preset trough threshold, it sends data and carries peak and trough detection markers in the data frame. After receiving the neighbor fault indicator, it verifies whether the amplitude of the carrier signal at the time of reception is in the peak-valley range based on the marking information. If it is in the range, the data is confirmed to be valid and used to extract the neighbor association feature data; otherwise, it is discarded and retransmission is triggered.

8. The power distribution branch box monitoring method based on a fault indicator according to claim 7, characterized in that, In step 4, the dynamic topology construction algorithm based on graph theory is the Kruskal algorithm. In step 6, the branch correlation degree calculation model adopts the hierarchical weighted fusion algorithm. In step 7, the multi-dimensional fault decision algorithm adopts the dual threshold fusion decision algorithm.

9. A power distribution branch box monitoring system based on a fault indicator, characterized in that, Used to perform the method of any one of claims 1-8.

Images