A method and system for determining an operating state of a power distribution network

Abstract

The application discloses a kind of operation state determination method and system of distribution network, applied to distribution network technical field, method includes: obtaining the topological graph and real-time operation data of target distribution network;First measurable node set is determined based on the historical operation data of the target distribution network obtained measurable data reliability;Determine the first initial predicted operation data corresponding to the first measurable node set;Measurable operation data and the first initial predicted operation data corresponding to the first measurable node set are fused based on measurable data reliability, and first node operation data is obtained;First node operation data and second initial predicted operation data are predicted based on, and the target real-time operation state of target distribution network is obtained, and target real-time operation state at least includes the predicted operation data of each node of target distribution network.The operation state determination method of distribution network provided by the application can obtain accurate real-time operation state under the condition that measurable data is not complete.

Description

Technical Field

[0001] This invention relates to the field of power distribution network technology, and in particular to a method and system for determining the operating status of a power distribution network. Background Technology

[0002] During the operation of a power distribution network, incomplete coverage of measuring devices and communication delays make it impossible to accurately obtain all real-time operational data. Therefore, determining the real-time operating status of the power distribution network provides an accurate data foundation for operations such as fault location, ensuring the operational safety and reliability of the network.

[0003] The existing method for determining the real-time operating status of a distribution network is to estimate the status based on historical data. For example, load curve extrapolation is performed based on historical operating data to obtain the power operating data of the currently unmeasured nodes. However, this method relies too much on historical operating data, which leads to a significant increase in the deviation between historical data and the current actual operating status when the distribution network load changes abruptly. Consequently, the estimation results of the unmeasured nodes cannot accurately reflect the current operating status, ultimately resulting in the inability to obtain accurate real-time operating status and severely reducing the accuracy of data sources for subsequent operations of the distribution network. Summary of the Invention

[0004] This invention provides a method and system for determining the operating status of a power distribution network, in order to solve the technical problem that the existing technology relies too much on historical operating data, resulting in the inability to obtain accurate real-time operating status, and thus achieve the effect of obtaining accurate real-time operating status.

[0005] To address the aforementioned technical problems, this invention provides a method and system for determining the operating status of a power distribution network, the method comprising: The topology map and real-time operation data of the target distribution network are obtained, and each node in the topology map is divided into a first measurable node set and a second unmeasurable node set. The reliability of the measurement data of the first measurable node set is determined based on the historical operation data of the target distribution network, and the measurement operation data of the first measurable node set is extracted from the real-time operation data. Determine the first initial prediction running data corresponding to the first measurable node set, and determine the second initial prediction running data corresponding to the second unmeasurable node set; Based on the reliability of the measurement data, the measurement operation data corresponding to the first measurable node set and the first initial prediction operation data are fused to obtain the first node operation data; Based on the first node's operating data and the second initial predicted operating data, a prediction is made to obtain the target real-time operating status of the target distribution network. The target real-time operating status includes at least the predicted operating data of each node of the target distribution network and the corresponding reliability of the predicted operating data.

[0006] Preferably, determining the reliability of the measurement data of the first measurable node set based on the acquired historical operating data of the target distribution network includes: The sub-historical operating data of each first measurable node in the first measurable node set is determined from the historical operating data. Based on a preset standard, the sub-historical operation data corresponding to each first measurable node is analyzed to obtain the reliability of the sub-measurement data. The preset standard includes at least the accuracy of the measurement device, communication latency, and data staleness. The reliability of the measurement data is determined based on the reliability of each of the sub-measurement data.

[0007] Preferably, determining the first initial prediction running data corresponding to the first measurable node set and determining the second initial prediction running data corresponding to the second unmeasurable node set includes: Power flow calculations are performed based on the real-time operating data and topology diagram of the target distribution network to obtain the operating status baseline values ​​of each node. Based on the typical operating scenarios of the target distribution network, the benchmark values ​​of each operating state are processed to obtain the first initial predicted operating data and the second initial predicted operating data.

[0008] Preferably, the step of predicting the target real-time operating status of the target distribution network based on the first node operating data and the second initial predicted operating data includes: The first node's running data and the second initial prediction running data are input into the target prediction model for processing to obtain the target's real-time running status. The processing procedure of the target prediction model includes: The initial processing state of the topology graph is determined based on the first node's running data, the second initial prediction running data, and the reliability of the measurement data. Based on the initial processing state, multi-hop propagation processing is performed, and the real-time running state of the target is obtained based on the processing results.

[0009] Preferably, determining the initial processing state of the topology graph based on the first node's running data, the second initial prediction running data, and the measurement data reliability includes: The node operation data of each node is determined based on the first node operation data and the second initial predicted operation data; The propagation credibility of each first measurable node in the first measurable node set is determined based on the credibility of the measured data. Based on the topology graph, the neighbor attention weights between each of the first measurable nodes are determined; The initial processing state is determined based on the running data of each node, the propagation credibility of each node, and the attention weight of each neighbor.

[0010] Preferably, the step of performing multi-hop propagation processing based on the initial processing state and obtaining the real-time running state of the target based on the processing result includes: The multi-hop propagation process includes N layers of propagation processing, where N is a positive integer; In the i-th layer of the N-layer propagation process, the output of the (i-1)-th layer propagation process is processed according to the node operation data of the neighboring nodes, the current propagation credibility of the sending node, and the neighbor attention weight to obtain the output of the i-th layer propagation process, where 1≤i≤N; When i=1, the output of the (i-1)th layer propagation process is the initial processing state; The real-time operating status of the target is determined by the output of the Nth layer propagation process.

[0011] Preferably, the target real-time operating status includes at least the predicted operating data of each node in the target distribution network and the corresponding reliability of the predicted operating data, including: The first node's running data and the second initial prediction running data are processed using a hybrid density technique to obtain the corresponding confidence level of the prediction running data; Based on the predicted running data and the confidence level, the real-time running status of the target is determined.

[0012] Preferably, after predicting the target real-time operating state of the target distribution network based on the first node operating data and the second initial predicted operating data, wherein the target real-time operating state includes at least the predicted operating data of each node of the target distribution network and the corresponding confidence level of the predicted operating data, the method further includes: The probability of exceeding the limit for each node is determined based on the predicted operational data. Based on each of the aforementioned exceedance probabilities, the urgency of regulation for the target distribution network is determined; Based on the aforementioned urgency of regulation and the probabilities of exceeding the limits, the target regulation scheme is obtained.

[0013] Preferably, determining the probability of exceeding the limit for each node based on the predicted running data includes: Determine the operational boundaries of each of the aforementioned nodes; Based on the operating boundary and the predicted operating data, the mutual exclusion probability of the three states of each node is determined, and the three states include low voltage state, overvoltage state and normal state. The corresponding out-of-limit probability is determined based on each of the mutual exclusion probabilities.

[0014] Another aspect of the present invention provides a system for determining the operating status of a power distribution network, comprising: The acquisition module is used to acquire the topology map and real-time operation data of the target distribution network, and divide each node in the topology map into a first measurable node set and a second unmeasurable node set. The credibility module is used to determine the credibility of the measurement data of the first measurable node set based on the historical operation data of the target distribution network, and to extract the measurement operation data of the first measurable node set from the real-time operation data. The determination module is used to determine the first initial prediction running data corresponding to the first measurable node set and to determine the second initial prediction running data corresponding to the second unmeasurable node set. The fusion module is used to fuse the measurement operation data corresponding to the first measurable node set and the first initial prediction operation data based on the reliability of the measurement data to obtain the first node operation data. The prediction module is used to make predictions based on the first node's operating data and the second initial predicted operating data to obtain the target real-time operating status of the target distribution network. The target real-time operating status includes at least the predicted operating data of each node of the target distribution network and the corresponding reliability of the predicted operating data.

[0015] Compared with the prior art, the beneficial effects of the present invention are at least one of the following: This invention acquires the topology and real-time operational data of a target distribution network, dividing each node in the topology into a first measurable node set and a second unmeasurable node set. Based on the acquired historical operational data of the target distribution network, the reliability of the measurement data for the first measurable node set is determined. Measurement operational data for the first measurable node set is extracted from the real-time operational data, and a first initial predicted operational data corresponding to the first measurable node set is determined. A second initial predicted operational data corresponding to the second unmeasurable node set is also determined. Finally, based on the reliability of the measurement data, the measurement operational data corresponding to the first measurable node set and the first initial predicted operational data are fused. By obtaining the operational data of the first node and then making predictions based on the first node operational data and the second initial predicted operational data, the target real-time operational state of the target distribution network is obtained. This target real-time operational state includes at least the predicted operational data of each node in the target distribution network and the corresponding credibility of the predicted operational data. This solves the problem in the prior art where over-reliance on historical operational data leads to a significant increase in the deviation between historical data and the current actual operational state when the distribution network load changes abruptly. Consequently, the estimation results of unmeasured nodes cannot accurately reflect the current operational state, ultimately failing to obtain an accurate real-time operational state and severely reducing the accuracy of subsequent operation data sources for the distribution network. This invention introduces the credibility of measured data and merges real-time measured operational data with initial predicted operational data. Then, by combining the first node operational data and the second initial predicted operational data for prediction, a target real-time operational state containing the predicted operational data of each node and its corresponding credibility can be obtained. This makes the estimation results of unmeasured nodes more accurately reflect the current operational state, thereby obtaining a more accurate real-time operational state. This provides a more accurate data foundation for subsequent operations such as fault location, ensuring the operational safety and reliability of the distribution network. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a method for determining the operating status of a power distribution network in one embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a power distribution network operation status determination system in one embodiment of the present invention; Figure label: Among them, 11. Acquisition module; 12. Credibility module; 13. Determination module; 14. Fusion module; 15. Prediction module. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] In the description of this invention, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0019] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0020] In the description of this invention, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0021] In distribution network operation, due to incomplete coverage of measuring devices and communication delays, it is difficult to accurately obtain all real-time operational data. Therefore, it is necessary to determine the real-time operating status of the distribution network to provide an accurate data foundation for operations such as fault location, ensuring operational safety and reliability. Existing methods mostly rely on historical data for state estimation, such as extrapolating the current power data of unmeasured nodes through load curves. However, this method is overly dependent on historical data. When the load changes abruptly, the deviation between historical data and the current actual state increases significantly, causing the estimation results of unmeasured nodes to fail to accurately reflect the current operating status. Ultimately, it is difficult to obtain accurate real-time operating status, severely reducing the accuracy of data sources for subsequent distribution network operations.

[0022] One embodiment of the present invention provides a method for determining the operating status of a power distribution network. For details, please refer to [link to relevant documentation]. Figure 1 , Figure 1 The diagram shown is a flowchart illustrating a method for determining the operating status of a power distribution network according to one embodiment of the present invention, including: S1. Obtain the topology map and real-time operation data of the target distribution network, and divide each node in the topology map into a first measurable node set and a second unmeasurable node set; S2. Determine the reliability of the measurement data of the first measurable node set based on the historical operation data of the target distribution network obtained, and extract the measurement operation data of the first measurable node set from the real-time operation data. S3. Determine the first initial prediction running data corresponding to the first measurable node set, and determine the second initial prediction running data corresponding to the second unmeasurable node set; S4. Based on the reliability of the measurement data, the measurement operation data corresponding to the first measurable node set and the first initial prediction operation data are fused to obtain the first node operation data. S5. Based on the first node operation data and the second initial predicted operation data, a prediction is made to obtain the target real-time operation status of the target distribution network. The target real-time operation status includes at least the predicted operation data of each node of the target distribution network and the corresponding reliability of the predicted operation data.

[0023] First, the topology map and real-time operation data of the target distribution network are obtained, and each node in the topology map is divided into a first measurable node set and a second unmeasurable node set.

[0024] The target distribution network refers to the specific low-voltage distribution network object that needs to be analyzed and its state estimated; the topology diagram refers to the graph structure representation of the connection relationship between nodes and lines in the distribution network, where nodes correspond to busbars or user nodes and edges correspond to lines in operation; real-time operating data refers to the operating information collected from the distribution network at the current moment, including node power, voltage, current, and the status of measuring devices; the first measurable node set refers to the set of nodes in the topology diagram that are equipped with measuring devices and can obtain real-time measuring data; the second unmeasurable node set refers to the set of nodes in the topology diagram that are not equipped with measuring devices or cannot obtain real-time measuring data, i.e., the set of blind zone nodes.

[0025] Step S1 first acquires the topology map of the target distribution network. This topology map can be exported from a distribution network geographic information system or a dispatch automation system, and includes information such as node numbers, inter-node line connections, line impedance, switch status, and phase. Simultaneously, it acquires real-time operational data, which can be collected from distribution network measurement devices, distribution automation terminals, or phasor measurement devices. This data includes the voltage, current, active power, reactive power, measurement timestamps, accuracy levels, and communication status of each node at the current moment. Then, based on whether each node in the topology map has a measurement device installed or can acquire real-time measurement data, all nodes are divided into a first measurable node set and a second unmeasurable node set. The first measurable node set contains all nodes capable of acquiring real-time measurement data, while the second unmeasurable node set contains all nodes that cannot acquire real-time measurement data, i.e., blind zone nodes. After the division, the first measurable node set is used to provide actual measurement operational data, while the second unmeasurable node set serves as the set of blind zone nodes whose operational status needs to be estimated later.

[0026] Preferably, the reliability of the measurement data of the first measurable node set is determined based on the historical operation data of the target distribution network, and the measurement operation data of the first measurable node set is extracted from the real-time operation data.

[0027] Historical operating data refers to the operating records accumulated by the target distribution network over a period of time, including information such as voltage, current, power, timestamps, accuracy levels, and communication status of measuring devices at each node; measurement data reliability refers to a continuous index used to measure the quality of measurement data at each node in the first set of measurable nodes, comprehensively reflecting factors such as device accuracy, communication latency, refresh rate, and data obsolescence; measured operating data refers to the measured voltage, current, power, and other operating data extracted from real-time operating data that belong to the first set of measurable nodes.

[0028] First, the historical operation data of each node in the first measurable node set over a past period is read from the historical operation database of the target distribution network. This includes information such as the accuracy level of the measuring device, communication delay records, data refresh rate, and data timestamp. Then, the historical operation data of each first measurable node is analyzed according to preset standards. The device accuracy quality index, communication delay quality index, refresh rate quality index, and data obsolescence quality index are calculated separately. These are then weighted and summed with non-negative weights equal to 1 to obtain the reliability of the sub-measurement data of that node. The reliability of the measurement data of the first measurable node set is then aggregated based on the reliability of each sub-measurement data. Simultaneously, measured data such as voltage, current, active power, and reactive power belonging to the first measurable node set are extracted from the real-time operation data at the current moment according to the node number. These measured data are then unified to the same reference time scale, phase system, and per-unit standard to obtain the measurement operation data of the first measurable node set.

[0029] Specifically, determining the reliability of measurement data for the first set of measurable nodes based on the acquired historical operating data of the target distribution network includes: determining sub-historical operating data for each first measurable node in the first set of measurable nodes from the historical operating data; analyzing the sub-historical operating data corresponding to each first measurable node based on preset standards to obtain the reliability of the sub-measurement data, wherein the preset standards include at least the accuracy of the measurement device, communication latency, and data obsolescence; and determining the reliability of the measurement data based on the reliability of each sub-measurement data.

[0030] The first measurable node refers to each specific node in the set of first measurable nodes; sub-historical operation data refers to the historical operation records belonging to a certain first measurable node, which are separated from the overall historical operation data by node; preset standard refers to the established criteria used to evaluate the quality of measurement data, which includes at least the accuracy of the measurement device, communication delay, and data staleness. Among them, the accuracy of the measurement device reflects the magnitude of the measurement error of the device itself, the communication delay reflects the time delay from data acquisition to reception, and the data staleness reflects the length of time interval between the data and the current moment; sub-measurement data reliability refers to the continuous index of the quality of the measurement data of a single first measurable node after analyzing its sub-historical operation data according to the preset standard; measurement data reliability refers to the index formed by combining the reliability of the sub-measurement data of each first measurable node in the set of first measurable nodes, which is used to characterize the overall reliability of the measurement data of the entire set of first measurable nodes.

[0031] First, the historical operation database of the target distribution network is retrieved by node number to find the sub-historical operation data of each first measurable node in the first measurable node set. Each record includes the accuracy level of the measuring device, communication delay, data refresh rate, data timestamp, and corresponding voltage, current, and power measurement values. Then, for each first measurable node, four quality indicators are calculated according to preset standards: the device accuracy quality indicator is determined by the ratio of the device accuracy level to the reference accuracy; the communication delay quality indicator is determined by the ratio of the actual communication delay to the maximum allowable delay; the refresh rate quality indicator is determined by the ratio of the actual refresh rate to the reference refresh rate; and the data staleness quality indicator is determined by the ratio of the difference between the current time and the most recent valid data timestamp to the maximum allowable staleness time. Each quality indicator is normalized to the range of 0 to 1, and then weighted and summed according to non-negative weights with a total weight of 1 to obtain the reliability of the sub-measurement data of the first measurable node.

[0032] The following explanation uses a specific first measurable node as an example.

[0033] Assuming the accuracy class of a measurement device at a first measurable node is 0.5 and the baseline accuracy is 0.2, then the device accuracy quality index is 0.2 divided by 0.5, which equals 0.4. The actual communication latency is 200 milliseconds, and the maximum allowable latency is 500 milliseconds, so the communication latency quality index is 1 - (200 / 500), which equals 0.6. The actual refresh rate is 1 second per cycle, and the baseline refresh rate is 0.5 seconds per cycle, so the refresh rate quality index is 0.5 divided by 1, which equals 0.5. The difference between the current time and the most recent valid data timestamp is 30 seconds, and the maximum allowable obsolescence time is 120 seconds, so the data obsolescence quality index is 1 - (30 / 120), which equals 0.75. All four quality indices are in the range of 0 to 1.

[0034] Then, a weighted sum is performed using weights that are non-negative and sum to 1. Assuming the weight for device accuracy is 0.4, communication latency is 0.2, refresh rate is 0.2, and data staleness is 0.2, the reliability of the sub-measurement data of the first measurable node is 0.4×0.4 + 0.2×0.6 + 0.2×0.5 + 0.2×0.75 = 0.53. This value, between 0 and 1, indicates that the measurement data quality of this node is at a medium level.

[0035] After obtaining the reliability of the sub-measurement data of each first measurable node, this scheme determines the reliability of the measurement data of the first measurable node set based on the reliability of each sub-measurement data. Specifically, aggregation can be performed using methods such as weighted average based on node importance, weighted average based on the number of measurement channels, or minimum value. If weighted average is used, non-negative weights are assigned based on the electrical location, load importance, or measurement coverage criticality of each first measurable node in the distribution network, and the reliability of each sub-measurement data is weighted and summed to obtain the overall measurement data reliability. If minimum value is used, the minimum value among the reliability of each sub-measurement data is taken as the overall measurement data reliability to conservatively reflect the worst measurement quality. The final measurement data reliability is a continuous value between 0 and 1, which is used for subsequent fusion with measurement operation data and initial prediction operation data.

[0036] Preferably, a first initial predicted operating data corresponding to the first measurable node set is determined, and a second initial predicted operating data corresponding to the second unmeasurable node set is determined. Power flow calculations are performed based on the real-time operating data and the topology diagram of the target distribution network to obtain the operating state baseline value of each node; the operating state baseline values ​​are then processed based on typical operating scenarios of the target distribution network to obtain the first initial predicted operating data and the second initial predicted operating data.

[0037] The first initial predicted operating data refers to the predicted initial operating state value pre-formed for the first measurable node set, which has not yet been integrated with the actual measured data. It can be derived from power flow calculations or typical scenario extrapolation. The second initial predicted operating data refers to the predicted initial operating state value pre-formed for the second unmeasurable node set. It can also be derived from power flow calculations or typical scenario extrapolation. Real-time operating data refers to the operating information collected from the distribution network at the current moment. The topology diagram refers to the graph structure representation of the connection relationship between nodes and lines in the distribution network. Power flow calculation refers to the calculation process of solving the steady-state voltage, current, and power distribution of the distribution network based on the topology diagram, node power, and line parameters. The operating state reference value refers to the basic reference value of the operating state of each node, such as voltage, current, and power, obtained through power flow calculation. The typical operating scenario refers to the representative operating conditions of the target distribution network in terms of season, load level, photovoltaic output, or topology.

[0038] The active and reactive power measurements of the first measurable node set are extracted from real-time operational data. For the second unmeasurable node set, typical load values ​​and typical photovoltaic output values ​​under the same season and load level are read from the historical database. The net injection is calculated as the photovoltaic output minus the load consumption to synthesize the node injection power. If a node's power is missing, the historical average value for the same period is taken and marked as low reliability. Then, the resistance, reactance, ground admittance, length, phase, and switch status of each branch are read from the topology diagram. Disconnected branches are deleted, and voltage and capacity base values ​​are selected. The line impedance, node power, voltage, and current are all converted to per-unit values, with the initial root node voltage value set to 1.0 per unit.

[0039] Then, power flow calculations are performed: For radial low-voltage distribution networks, a forward-backward substitution method is used. Starting from the end nodes, the branch currents and power losses are calculated backward in the order from downstream to upstream. Then, starting from the root nodes, the voltages of each node are calculated forward in the order from upstream to downstream. This process is repeated iteratively until the maximum difference between two adjacent voltage amplitudes is less than 10 to the power of -6 per unit. If a loop network exists or convergence is difficult, the Newton-Raphson method is used for iterative solution. After convergence, the voltage amplitude, voltage phase angle, active power and reactive power injected into each node, as well as the active power, reactive power, current amplitude and power loss of each branch are recorded as the operating status reference values ​​for each node. Subsequent physical constraint weights are set according to the power source marking.

[0040] Finally, initial predicted operating data is obtained based on typical operating scenarios: a library of typical operating scenarios is constructed in advance according to season, load level, photovoltaic output, and topology status. Each scenario corresponds to a set of node load scaling factors, photovoltaic output scaling factors, voltage corrections, and branch power corrections. Scenario parameters are obtained by clustering historical operating data. At the current moment, scenarios are matched by month, time period, total load level, and total photovoltaic output. If multiple scenarios are matched, the smallest Euclidean distance between total load and photovoltaic output is selected, or a weighted combination is used. The baseline values ​​of each node's operating status are scaled and corrected according to the scaling factors and corrections of the selected scenarios. The results corresponding to the first set of measurable nodes are used as the first initial predicted operating data, and the results corresponding to the second set of unmeasurable nodes are used as the second initial predicted operating data. Subsequently, it is verified whether the voltage amplitude, branch power, and root node injection exceed the limits or are significantly unbalanced. If not, the scaling factors are adjusted and recalculated. After verification, the scenario number and correction amount are output and recorded.

[0041] Preferably, based on the reliability of the measurement data, the measurement operation data corresponding to the first measurable node set and the first initial prediction operation data are fused to obtain the first node operation data.

[0042] The first node running data refers to the running data of the first measurable node set obtained after fusion and used for subsequent prediction. The higher the credibility, the larger the proportion of the measurable running data; the lower the credibility, the larger the proportion of the first initial prediction running data.

[0043] In one embodiment of the present invention, for each node in the first measurable node set, its measurement operation data and first initial prediction operation data are taken respectively, and weighted and fused according to the confidence level ρ of the measurement data. The fused first node operation data is equal to ρ multiplied by the measurement operation data, plus 1 minus ρ multiplied by the first initial prediction operation data. If a channel measurement is missing or the confidence level is lower than a preset lower limit, the first initial prediction operation data is taken directly for that channel, and the confidence level mark of that channel is retained. In actual operation, the measurement operation data is constructed as the observation complex amplitude value according to the continuous measurement confidence encoding method, and the first initial prediction operation data is constructed as the prior complex amplitude value according to the historical statistics or pseudo measurement method. The two form the observation feature and the prior feature respectively. The observation initial hidden state and the prior initial hidden state are obtained through parameter non-sharing mapping. Then, the real part and the imaginary part of the observation complex amplitude value and the prior complex amplitude value are weighted and fused with ρ as the gate weight. The first node operation data is restored from the fused complex amplitude value. If the measurement of a certain channel is missing or the confidence is lower than the preset lower limit, the first initial prediction operation data is directly taken for that channel, and the confidence mark of that channel is retained for subsequent multi-hop propagation.

[0044] Because the accuracy, communication latency, refresh rate, and data staleness of measurement devices vary significantly in low-voltage distribution networks, directly processing them with hard masks (whether measured or not) would lose the continuous differences in measurement quality. Using only real measurement values ​​would ignore information such as phase and missing values. Therefore, a fusion approach at the complex amplitude encoding level is adopted. First, observed complex amplitude values ​​and prior complex amplitude values ​​are constructed separately. Then, two independent initial hidden states are obtained through independent initialization and parameter-non-sharing mapping of observed and prior features. This approach simultaneously preserves measurement quality, phase information, observation masks, and prior information, avoiding the mixing of observed and prior evidence at the input. Furthermore, it maintains structural consistency with subsequent observation and prior dual-branch layer-by-layer confidence-gated propagation, thus more effectively suppressing the contamination of blind zone nodes by low-confidence measurements during multi-hop propagation.

[0045] Finally, based on the first node operating data and the second initial predicted operating data, a prediction is made to obtain the target real-time operating state of the target distribution network. The target real-time operating state includes at least the predicted operating data of each node in the target distribution network and the corresponding confidence level of the predicted operating data. The first node operating data and the second initial predicted operating data are input into the target prediction model for processing to obtain the target real-time operating state. The processing of the target prediction model includes: determining the initial processing state of the topology based on the first node operating data, the second initial predicted operating data, and the confidence level of the measurement data; performing multi-hop propagation processing based on the initial processing state; and obtaining the target real-time operating state based on the processing result.

[0046] The first node operating data refers to the node operating data obtained by fusing the first measurable node set with the first initial predicted operating data; the second initial predicted operating data refers to the initial operating state prediction value corresponding to the second unmeasurable node set; the target real-time operating state refers to the operating state of the target distribution network to be determined by this scheme at the current moment, including at least the predicted operating data of each node and its corresponding credibility; the target prediction model refers to the model used to make predictions based on the first node operating data, the second initial predicted operating data, and the credibility of the measured data; the initial processing state refers to the graph state representation obtained after the target prediction model initializes the topology graph based on the first node operating data, the second initial predicted operating data, and the credibility of the measured data; multi-hop propagation processing refers to the process of multi-layer information transmission and aggregation along the connection relationship between nodes on the topology graph, so that high-credibility information can propagate across multiple nodes to the blind zone node; the predicted operating data refers to the predicted operating state value of each node output by the target prediction model, and the credibility refers to the reliability of the predicted operating data.

[0047] The first node operation data of the first measurable node set and the second initial prediction operation data of the second unmeasurable node set are input into the target prediction model. The target prediction model first determines the initial processing state of the topology graph: the first node operation data and the second initial prediction operation data are concatenated according to the node number to form the node operation data of each node, and the reliability of the measured data is used as the propagation reliability of each node in the first measurable node set. The propagation reliability of the second unmeasurable node set is initialized to the prior reliability corresponding to its node operation data. At the same time, the neighbor attention weights between each node and its physical neighbors and self-loops are determined according to the topology graph. Then, the node operation data, each propagation reliability, and each neighbor attention weight are jointly encoded into the initial processing state. Then, the target prediction model performs multi-hop propagation processing based on the initial processing state. This processing includes N layers of propagation. In the i-th layer, the output results of the (i-1)-th layer are weighted, aggregated, and updated according to the node operation data of the neighbor nodes, the current propagation reliability of the sending node, and the neighbor attention weights to obtain the output result of the i-th layer. The input of the first layer is the initial processing state, and the output result of the N-th layer determines the real-time operation state of the target. Finally, the target prediction model uses a hybrid density technique to process the output results of the Nth layer to obtain the predicted operating data of each node and its corresponding confidence level, thereby outputting the target real-time operating status of the target distribution network. This status includes at least the predicted operating data of each node and the confidence level corresponding to the predicted operating data.

[0048] Specifically, determining the initial processing state of the topology graph based on the first node running data, the second initial predicted running data, and the measurement data reliability includes: determining the node running data of each node based on the first node running data and the second initial predicted running data; determining the propagation reliability of each first measurable node in the first measurable node set based on the measurement data reliability; determining the neighbor attention weights among each first measurable node based on the topology graph; and determining the initial processing state based on each node running data, each propagation reliability, and each neighbor attention weight.

[0049] Node running data refers to the running data covering all nodes in the topology graph, formed by concatenating the first node running data and the second initial predicted running data according to node numbers; Measurement data credibility refers to a continuous index used to measure the quality of measurement data in the first measurable node set; Propagation credibility refers to the degree of credibility of each node in the first measurable node set in the subsequent multi-hop propagation process for gating message transmission, determined by the measurement data credibility; Neighbor attention weight refers to the weight of adjacent nodes in the topology graph during information aggregation, calculated based on node characteristics, power similarity, electrical distance, and switch status; Initial processing state refers to the initial graph state representation of the topology graph before entering multi-hop propagation processing, formed by encoding each node running data, each propagation credibility, and each neighbor attention weight.

[0050] The following steps are performed to determine the initial processing state. First, read the first node operating data of the first measurable node set and the second initial predicted operating data of the second unmeasurable node set. Establish a node data table according to the node numbering order in the topology diagram, with each node corresponding to one row. For the first measurable nodes, fill in the fused voltage, current, active power, and reactive power; for the second unmeasurable nodes, fill in the voltage, current, active power, and reactive power from the second initial predicted operating data, obtaining node operating data covering all nodes. Second, read the measurement data reliability of the first measurable node set and directly write the measurement data reliability of each first measurable node into the propagation reliability field of that node in the node data table. For the second unmeasurable node, initialize its propagation reliability field to zero or a preset prior reliability, and simultaneously fill in 1 in the measurement missing marker field. The third step involves traversing each edge in the topology graph. For each edge's two endpoints, nodes i and j, a connection vector is formed using the node operation data and propagation confidence of nodes i and j. This vector is mapped using the query weight matrix and key weight matrix, then multiplied by the dot product and divided by the square root of the dimension to obtain the feature attention score. Next, the normalized power similarity between nodes i and j is calculated. The reciprocal of the line impedance amplitude between the two nodes is taken as the electrical distance prior, and the bias value corresponding to the switch state is taken. These four items are added together to obtain the unnormalized score. Then, the unnormalized score is Softmax normalized on the physical neighbors and self-loop sets of node i to obtain the neighbor attention weights between nodes i and j. The fourth step involves organizing the node operation data, propagation confidence field, measurement missing marker field, and neighbor attention weights between each node and its neighbors from the node data table into a graph structure tensor based on node number and graph adjacency relationship. This graph structure tensor represents the initial processing state of the topology graph and serves as the input for subsequent multi-hop propagation processing.

[0051] Specifically, multi-hop propagation processing is performed based on the initial processing state, and the real-time running state of the target is obtained based on the processing result. This includes: the multi-hop propagation processing includes N layers of propagation processing, where N is a positive integer; in the i-th layer of propagation processing of the N layers, the result output by the (i-1)-th layer of propagation processing is processed according to the node running data of neighboring nodes, the current propagation credibility of the sending node, and the neighbor attention weight to obtain the result output by the i-th layer of propagation processing, where 1≤i≤N; when i=1, the result output by the (i-1)-th layer of propagation processing is the initial processing state; the real-time running state of the target is determined by the result output by the N-th layer of propagation processing.

[0052] Multi-hop propagation processing refers to the process of transmitting and aggregating information across multiple layers along the connections between nodes in a topology graph; N-layer propagation processing refers to the number of layers in the multi-hop propagation process, where N is a positive integer, and each layer corresponds to one information aggregation; the i-th layer propagation processing refers to any one of the N layers, where i is the layer number and satisfies 1≤i≤N; neighbor nodes refer to nodes in the topology graph that have a physical connection or self-loop relationship with the current node; node running data refers to the running data covering all nodes formed by concatenating the first node running data and the second initial predicted running data according to the node number; sending node refers to a neighbor node that sends a message to the current node in a certain layer of propagation; current propagation Credibility refers to the propagation credibility of a sending node when it enters this layer of propagation, used to gate the strength of the sending node's message; Neighbor attention weight refers to the normalized weight between the current node and the sending node when information is aggregated; The output of the (i-1)th layer of propagation processing refers to the graph state representation obtained after the previous layer of propagation processing, and when i equals 1, the result is the initial processing state; The output of the ith layer of propagation processing refers to the updated graph state representation after this layer of propagation processing; The target real-time operating state refers to the current operating state of the target distribution network determined by the output of the Nth layer of propagation processing, which includes at least the predicted operating data of each node and its corresponding credibility.

[0053] The multi-hop propagation process is executed in the following steps. First, the number of propagation layers N is set to a positive integer, for example, N equals 3, and the initial processing state is taken as the output of the 0th layer propagation process. The initial processing state includes the node running data, propagation confidence and neighbor attention weights of each node, and each node retains two independent branches: the observed hidden state and the prior hidden state.

[0054] Then, propagation processing is performed layer by layer starting from layer 1. For layer i, where i ranges from 1 to N, propagation processing is performed. In layer i, for each node i in the topology graph, the set of its physical neighbors and self-loops is read first. For each neighbor node j in the set, the following operations are performed: the observed hidden state and prior hidden state of neighbor node j are extracted from the output of layer (i-1). Using the neighbor attention weight between node i and node j as the aggregation weight, the observed message and prior message are calculated respectively. The observed message is calculated by mapping the observed hidden state of neighbor node j through the observation weight matrix, multiplying it by the propagation confidence of neighbor node j, multiplying it by the neighbor attention weight, and summing it over all neighbors and self-loops. The prior message is calculated by mapping the prior hidden state of neighbor node j through the prior weight matrix, multiplying it by 1 minus the propagation confidence of neighbor node j, multiplying it by the neighbor attention weight, and summing it over all neighbors and self-loops. If a multi-head attention mechanism is used, the above process is performed separately on each attention head, and then the messages from each attention head are concatenated.

[0055] Next, for each node i, its observed hidden state at layer i-1 and the observed message obtained from aggregation at this layer are concatenated and sent to the observation update unit to obtain the observed hidden state at layer i. Its prior hidden state at layer i-1 and the prior message obtained from aggregation at this layer are concatenated and sent to the prior update unit to obtain the prior hidden state at layer i. The observation update unit and the prior update unit adopt a network structure with non-shared parameters, such as independent multilayer perceptrons or gated recurrent units. Simultaneously, the propagation credibility of neighboring nodes is weighted and averaged according to the neighbor attention weights to obtain the propagation credibility of node i after updating at layer i. This propagation credibility is maintained between 0 and 1. Then, using the updated propagation credibility as a gating coefficient, the observed hidden state and the prior hidden state of node i are weighted and fused. That is, the fused state equals the propagation credibility multiplied by the observed hidden state, plus 1 minus the propagation credibility multiplied by the prior hidden state, to obtain the output result of node i at layer i.

[0056] After completing the calculations for all nodes in layer i, the observed hidden state, prior hidden state, propagation confidence, and fusion state of each node in layer i are combined into the output of the propagation processing in layer i, and used as the input for the propagation processing in layer i+1. This process is repeated until the propagation processing in layer N is completed. Finally, the fusion state of each node in the output of the propagation processing in layer N is sent to the mixing density output head, which outputs the mixing coefficient, mean, and scale parameter of the voltage amplitude of that node. Then, the voltage amplitude point estimate, variance, low voltage probability, normal probability, overvoltage probability, risk value, and urgency are calculated to obtain the target real-time operating state of the target distribution network. This state includes at least the predicted operating data of each node and the confidence level corresponding to the predicted operating data.

[0057] Specifically, the target real-time operating status includes at least the predicted operating data of each node of the target distribution network and the corresponding confidence level of the predicted operating data, including: processing the first node operating data and the second initial predicted operating data using a hybrid density technique to obtain the corresponding confidence level of the predicted operating data; and determining the target real-time operating status based on the predicted operating data and the confidence level.

[0058] Hybrid density technology refers to a method of modeling and predicting the uncertainty of operational data by using a weighted combination of multiple probability distribution components, such as a mixture of Gaussian distributions. Predicted operational data refers to the predicted values ​​of operational status such as voltage, current, and power of each node output by the model. Credibility refers to the reliability of the predicted operational data, which can be calculated from the weight, mean, and variance of each component in the hybrid density distribution. The target real-time operational status refers to the current operational status of the target distribution network, which is composed of the predicted operational data of each node and their corresponding credibility.

[0059] After completing N-layer multi-hop propagation, the target prediction model takes the fused state output from the Nth layer and sends it to the mixing density output head. The mixing density output head outputs C Gaussian components of raw parameters for each node, including raw mixing coefficients, raw mean, and raw scale, where C can be 3 to 5. Then, the raw mixing coefficients are Softmax normalized to obtain the weights π_ic for each component, with the sum of all weights being 1. The raw scale is Softplus transformed to obtain the standard deviation σ_ic for each component, ensuring it is positive. The raw mean is directly used as the mean μ_ic for each component. This yields the mixed Gaussian probability density of the voltage amplitude at each node, as shown in the formula. Next, the expectation of the mixing distribution is used as the point estimate of the voltage amplitude, i.e., the sum of the weights of each component multiplied by their corresponding means. The variance of the mixing distribution is used to quantify the uncertainty, i.e., the sum of the weights of each component multiplied by their corresponding variances and the squares of the means, minus the squares of the point estimates. Simultaneously, the probabilities of low voltage, normal operation, and overvoltage are calculated based on the cumulative distribution function of the Gaussian mixture distribution, thus obtaining the total over-limit probability. Furthermore, the risk value and urgency of regulation are calculated by combining the three-state normalized information entropy and the normalized expected over-limit amplitude. Finally, the voltage amplitude point estimate, variance, confidence interval, low voltage probability, normal operation probability, overvoltage probability, risk value, and urgency of each node, along with the corresponding measurement data reliability and propagation reliability, are written into the output record to constitute the target real-time operating status of the target distribution network.

[0060] The model used in this embodiment of the invention is based on a graph attention network architecture, which is a readily available standard multi-layer graph attention network. Each layer updates the hidden state of the current node by aggregating information from neighboring nodes through attention weights. This solution does not change the basic architecture of the graph attention network, but rather makes adaptive adjustments around the input, output, and intermediate states to accommodate the needs of incomplete measurements and voltage probability estimation of blind zone nodes.

[0061] Regarding input, standard graph attention networks typically only accept node features and a fixed adjacency matrix. This scheme expands the input to include an initial processing state, specifically comprising three parts: node operation data for each node, which is formed by concatenating the first node operation data fused from the first measurable node set and the second initial predicted operation data from the second unmeasurable node set according to node number; the propagation confidence of each node, determined by the confidence of the measured data from the first measurable node set, with the second unmeasurable node initialized to zero or a preset prior confidence; and the neighbor attention weights between each node and its physical neighbors and self-loops, obtained by adding the node feature attention score, normalized power similarity, electrical distance prior, and switch state bias, followed by Softmax normalization.

[0062] Regarding intermediate states, standard graph attention networks have only a single hidden state per node. This scheme organizes the state of each node into three parts: the observed hidden state, the prior hidden state, and the propagation confidence. The observed hidden state is initialized by observation features, which include the real part, imaginary part, mask, and confidence of the actual observation value after weighting by the observation complex amplitude. The prior hidden state is initialized by prior features, which include the real part, imaginary part, missing value, and unreliability of historical statistics or pseudo-measurements after weighting by the prior complex amplitude. The initial value of the propagation confidence is determined by the confidence of the measurement data. During each layer of propagation, the observed hidden state and the prior hidden state are aggregated with neighbor messages and updated separately. The propagation confidence is updated by weighting the neighbor confidence according to the neighbor attention weights. At the output of each layer, the two hidden states are gated and fused according to the updated propagation confidence. The next layer still reads the observed hidden state, the prior hidden state, and the propagation confidence separately, without treating the fused result as a new observation representation.

[0063] In terms of output, standard graph attention networks typically output a single hidden state or point estimate. This scheme, after propagation at layer N, sends the fused state of each node to the mixing density output head, outputting the mixing coefficients, mean, and scale parameters of C Gaussian components. This yields the voltage amplitude point estimate, variance, low voltage probability, normal probability, overvoltage probability, risk value, and urgency level. These, along with the measurement data reliability and propagation reliability of each node, are written into the output record, constituting the target real-time operating state of the target distribution network.

[0064] Following step S5, the method further includes: determining the over-limit probability of each node based on the predicted operating data; determining the control urgency of the target distribution network based on each over-limit probability; obtaining a target control scheme based on the control urgency and each over-limit probability; and determining the target real-time operating status based on the predicted operating data and the confidence level.

[0065] The probability of exceeding the limit refers to the probability that the predicted operating data of a node exceeds the safe operating boundary, which can be calculated by the cumulative distribution function of the mixed density distribution; the urgency of regulation refers to the non-negative index used to characterize the degree to which a node needs to be prioritized for regulation, which is obtained by comprehensively considering the probability of exceeding the limit of each node and its uncertainty, the magnitude of the exceeding the limit and the electrical coupling effect of adjacent nodes; the target regulation scheme refers to the scheme formed based on the urgency of regulation and the probability of exceeding the limit of each node, which is used to guide operation and maintenance or scheduling to prioritize which nodes to handle, such as prioritizing verification, adding temporary measurements or performing reactive power regulation.

[0066] Read the pre-set lower and upper voltage safety limits, such as 0.95 per unit and 1.05 per unit. For each node, use the Gaussian mixture distribution of its voltage amplitude to calculate the probability of low voltage, normal voltage, and overvoltage. The sum of these three probabilities is 1. Then, add the low voltage probability and the overvoltage probability to obtain the total over-limit probability. At the same time, retain the low voltage probability and the overvoltage probability to distinguish the direction of the over-limit. Then, determine the urgency of regulation based on each over-limit probability. Specifically, for each node, take the normalized information entropy of its three-state probabilities as an uncertainty measure, and calculate the original expected over-limit amplitude, that is, calculate the expectation of the absolute value of the voltage deviation for the low voltage range and the overvoltage range respectively, and then normalize it according to the positive reference amplitude. Combine the over-limit probability, the normalized information entropy, and the normalized expected over-limit amplitude with non-negative weights to obtain the risk value. Further consider the upward trend of uncertainty and the electrical coupling effect of adjacent nodes, construct non-negative electrical coupling weights according to the line impedance amplitude, and add the weighted sum of the risk values ​​of adjacent nodes to the risk value of this node to obtain the regulation urgency of the node. Finally, based on the urgency of regulation and the probabilities of each limit exceedance, the target regulation scheme is obtained. Specifically, nodes are sorted from highest to lowest urgency, and the direction of each limit exceedance is recorded. When the urgency exceeds the threshold and the overvoltage probability is greater than the undervoltage probability, it is marked as high overvoltage urgency. When the urgency exceeds the threshold and the undervoltage probability is not less than the overvoltage probability, it is marked as high undervoltage urgency. When uncertainty is high but the total limit exceedance probability is low, it is recommended to prioritize adding temporary measurements or verifying data quality. The sorting and marking results are output as the target regulation scheme, allowing the existing safety verification module to select specific regulation actions based on equipment capacity and operational constraints. Simultaneously, the predicted operating data, corresponding reliability, limit exceedance probability, risk value, and urgency of each node are written into the output record, constituting the target real-time operating status of the target distribution network.

[0067] Specifically, determining the over-limit probability of each node based on the predicted operating data includes: determining the operating boundary of each node; determining the mutual exclusion probability of the three states of each node based on the operating boundary and the predicted operating data, wherein the three states include a low voltage state, an overvoltage state, and a normal state; and determining the corresponding over-limit probability based on each mutual exclusion probability.

[0068] Operating boundaries refer to the upper and lower voltage limits allowed for safe operation of a node, such as a low voltage threshold of 0.95 per unit and an overvoltage threshold of 1.05 per unit. Three states refer to dividing the node voltage operating state into three mutually exclusive states: low voltage state, normal state, and overvoltage state. Mutually exclusive probability refers to the probability that the node voltage falls into the low voltage range, normal range, and overvoltage range, respectively. The sum of the three is 1, which can be calculated by the cumulative distribution function of the mixed Gaussian distribution of voltage amplitude.

[0069] First, determine the operating boundary for each node by reading the pre-set lower voltage safety limit v_min and upper voltage safety limit v_max. For example, in a low-voltage distribution network, v_min can be set to 0.95 per unit and v_max to 1.05 per unit. Store this boundary in a boundary table, corresponding one-to-one with the node number. Then, determine the three-state mutual exclusion probabilities based on the operating boundaries and predicted operating data. Specifically, extract the Gaussian mixture distribution parameters of the voltage amplitude for each node from the target prediction model output, including the mixing coefficient π_ic, mean μ_ic, and standard deviation σ_ic of each Gaussian component. For each node, use the cumulative distribution function Φ of the Gaussian mixture distribution to calculate the low-voltage probability, over-voltage probability, and normal probability. The low-voltage probability is equal to the sum of the mixing coefficients of each component multiplied by the value of Φ at v_min; the over-voltage probability is equal to the sum of the mixing coefficients of each component multiplied by 1 minus the value of Φ at v_max; and the normal probability is equal to 1 minus the low-voltage probability minus the over-voltage probability. The sum of these three is 1, and all are clipped to the interval between 0 and 1. Finally, the over-limit probability is determined based on the mutual exclusion probabilities. That is, for each node, the low voltage probability and the over voltage probability are added together to obtain the total over-limit probability of the node. At the same time, the low voltage probability and the over voltage probability are retained to distinguish the over-limit direction. The total over-limit probability, low voltage probability and over voltage probability of each node, together with the node number, are output for subsequent calculation of the urgency of regulation and generation of target regulation scheme.

[0070] The above embodiments provide the overall process for determining the operating status of a distribution network based on incomplete measurement data, including node division, determination of measurement data reliability, determination of first and second initial predicted operating data, fusion of measurement operating data and first initial predicted operating data to obtain first node operating data, and prediction of the target real-time operating status based on the first node operating data and second initial predicted operating data. The following embodiment is a specific implementation of the above embodiments. Based on the above overall process, it further provides the specific construction of the observation mask and continuous reliability in continuous measurement reliability encoding, the complex amplitude encoding method of observation components and prior components, the independent initialization method of observation features and prior features, and the specific implementation of multi-hop dynamic graph inference with observation and prior dual-branch layer-by-layer reliability gating. Therefore, the following embodiment is a specific implementation and refinement of the above embodiments in the scenario of voltage probability estimation of blind zone nodes in low-voltage distribution networks. The two belong to the relationship between the overall description and specific implementation of the same technical solution.

[0071] Another embodiment of the present invention provides a method for voltage probability estimation of blind zone nodes in low-voltage distribution networks based on incomplete measurement data. At the current moment, using available partial measurement data, node power information, network topology, and switch states, the method estimates the voltage probability of unmeasured nodes at the same time, and provides point estimates, uncertainties, and risk ranking for exceeding limits. The overall technical approach includes continuous measurement reliability encoding, multi-hop dynamic graph inference with observation and prior dual-branch layer-by-layer reliability gating, hybrid density probability output, DistFlow (Distribution Flow) physical soft constraints, information-theoretic risk assessment, and optional online adaptive updates.

[0072] First, at the input representation level, this invention no longer employs simple hard masks with and without measurements, but instead constructs a quantum-inspired continuous measurement reliability encoding. For each node's voltage, current, active power, and reactive power channels, observation masks and continuous reliability are set separately. The observation mask distinguishes between true zero values ​​and unobserved values, while the continuous reliability integrates quality indicators such as device accuracy, communication latency, refresh rate, and data staleness to form differentiable weights in the 0-1 range. Furthermore, the observed and prior components of each channel are encoded as complex amplitude values, where the observed components are determined by the reliability and staleness phase, and the prior components are obtained from historical statistics or pseudo-measurements. Observational features and prior features are initialized separately, forming two independent initial hidden states, thereby avoiding the mixing of observational evidence and prior evidence at the input.

[0073] Secondly, at the topology reasoning level, this embodiment of the invention constructs a distribution network diagram structure, where nodes correspond to buses or user nodes, edges correspond to operational lines, and disconnected edges are deleted and self-loops are added based on switch states. In each propagation layer, the observation branch and the prior branch generate messages respectively: the observation message is generated only from the observation hidden state and is gated by the propagation credibility of the sending node; the prior message is generated only from the prior hidden state and is gated by the complement of the credibility of the sending node. The edge attention score is composed of feature attention, normalized power similarity, electrical distance prior, and switch state bias, and is normalized on the physical neighbor and self-loop sets. The propagation credibility is updated layer by layer according to the weighted average of the neighbor credibility. The two branches evolve independently using update units with non-shared parameters, and gating fusion is performed only at the output end of each layer according to the updated propagation credibility. Thus, high-credibility measurement information can propagate across continuous blind zones, low-credibility measurements are suppressed in multi-hop propagation, and the prior branch undertakes backoff when necessary, avoiding the problem of empty neighbors, zero denominators, or priors being mistakenly referred to as measured observations.

[0074] At the output level, this embodiment of the invention employs a hybrid density network to probabilistically model the voltage amplitude of unmeasured nodes, outputting the mixing coefficients, mean, and scale parameters of multiple Gaussian components. The expected value of the hybrid distribution is used as the point estimate of the voltage amplitude, and the variance of the hybrid distribution is used to quantify the uncertainty. Simultaneously, to introduce physical constraints on AC power flow, this method defines a variable with positive net load consumption in the DistFlow equation, maintaining consistency with the positive injection convention at the input end through an explicit negative sign conversion. Branch residuals are constructed based on the branch power recursive relationship, voltage drop relationship, and line current square relationship of the radial distribution network, and are used only as soft physical constraints during the training phase. The total loss consists of the negative log-likelihood of the voltage probability, the DistFlow physical residual, and the consistency term of available observations. For nodes or branches lacking reliable power sources, the corresponding physical weights are reduced or set to zero to avoid inaccurate pseudo-measurements forming strong equality constraints.

[0075] Furthermore, this embodiment of the invention constructs a risk assessment chain based on the probability output. The probabilities of three mutually exclusive states—low voltage, normal, and overvoltage—are calculated using the cumulative distribution function of a Gaussian mixture distribution, and a total over-limit probability is defined. Treating the three states as discrete states, a normalized information entropy is calculated to measure uncertainty; simultaneously, the original expected over-limit magnitude is calculated and normalized to obtain the normalized over-limit magnitude. Combining the over-limit probability, the uncertainty of the three states, and the normalized expected over-limit magnitude, a non-negative risk value is obtained; then, considering the upward trend of uncertainty and the electrical coupling effect of adjacent nodes, the urgency of regulation is calculated. Based on the direction and probability of the over-limit, high urgency for overvoltage or high urgency for low voltage can be marked; when uncertainty is high and the over-limit probability is low, it is recommended to prioritize adding temporary measurements or verifying data quality. Finally, a risk ranking is output for existing safety verification modules to select regulatory actions, rather than directly replacing scheduling instructions.

[0076] Furthermore, embodiments of the present invention also provide optional meta-learning and online adaptive mechanisms. In the offline phase, multiple tasks can be constructed according to season, load level, photovoltaic level, or topology state. Each task is strictly divided into a support set and a query set, and initialized using MAML (Model-Agnostic Meta-Learning). In the online phase, drift detection is performed by calculating the KL (Kullback-Leibler) divergence of the measured node voltage distribution or residual distribution using a sliding window. Candidate updates are only allowed when the drift exceeds a threshold for multiple consecutive windows and meets the minimum adaptation interval. The online loss does not use the true voltage label of the blind zone, but consists of the observation consistency loss of the current phasor nodes and the physical residuals of the entire network. Candidate updates preferably only adjust the probability output header, phase header, or upper-level parameters of the graph network, and utilize the latest measured nodes as the validation set. The current model is replaced only when the observation loss of the candidate model decreases and the physical residual does not increase beyond the allowable limit; otherwise, the update is rolled back and frozen to the next minimum interval, thereby avoiding cumulative drift caused by unlabeled self-updates.

[0077] This embodiment verifies the voltage point estimation accuracy, limit violation identification, robustness under sparse measurement conditions, adaptability to operational disturbances, probability distribution quality, and risk ranking capability. The main parameter configurations are shown in Table 1, the point estimation and adaptability results are shown in Tables 2 to 4, and the probability estimation and risk ranking results are shown in Tables 5 and 6.

[0078] Table 1 Experimental parameter configuration The configuration in Table 1 provides a unified experimental benchmark for subsequent verification in six aspects: voltage point estimation accuracy, limit violation identification, robustness under measurement sparsity conditions, adaptability to operational disturbances, probability distribution quality, and risk ranking capability.

[0079] Table 2. Experimental Results of Voltage Estimation and Limit Exceedance Identification Table 2 shows that when using the complete technical solution, the MAPE (Mean Absolute Percentage Error) for measurement nodes is 0.8%, the MAPE for blind nodes is 1.0%, and the F1 score for out-of-bounds recognition is 0.92. Compared with the probabilistic graphical network without continuous confidence gating, the MAPE for blind nodes decreases from 2.2% to 1.0%, a reduction of approximately 54.5%, and the F1 score increases from 0.80 to 0.92; compared with the hard-masked GNN (Graph Neural Network), the MAPE for blind nodes decreases by approximately 64.3%. Table 3 shows that when the measurement coverage decreases to 15%, the MAPE for blind nodes in this solution is still 1.5%, a reduction of approximately 64.3% compared to the 4.2% of the hard-masked GNN, indicating that multi-hop confidence propagation can maintain high estimation accuracy in continuous blind zones. As shown in Table 4, under the scenarios of sudden load increase, change in photovoltaic penetration rate and topology reconfiguration, MAPE decreased by 65.6%, 67.9% and 71.1% respectively after enabling online adaptive mode, and the update time did not exceed 10 seconds.

[0080] Table 3. Robustness test results under different measurement coverage rates. Table 3 shows the robustness comparison results of the present invention and the hard-masked GNN in dead zone node voltage estimation under different measurement coverage rates. As the measurement coverage rate gradually decreases from 50% to 30%, 24.2%, and 15%, the MAPE of the present invention remains at 0.6%, 0.9%, 1.0%, and 1.5%, respectively, while the MAPE of the hard-masked GNN increases from 1.5% to 2.5%, 2.8%, and 4.2%. At an extremely low coverage rate of 15%, the MAPE of the present invention is still only 1.5%, which is about 64.3% lower than that of the hard-masked GNN (4.2%). This indicates that the multi-hop confidence propagation mechanism can effectively maintain the estimation accuracy in continuous dead zones and has significantly stronger robustness to measurement sparsity conditions.

[0081] Table 4 Results of the online adaptive experiment Table 4 shows the experimental performance of the online adaptive mechanism under three operational disturbance scenarios: load surge, photovoltaic penetration change, and network topology reconfiguration. Without adaptation, the MAPE for the three scenarios were 3.2%, 2.8%, and 4.5%, respectively. After enabling online adaptation, the MAPE decreased to 1.1%, 0.9%, and 1.3%, respectively, with reductions of approximately 65.6%, 67.9%, and 71.1%, and update times of 5 seconds, 8 seconds, and 10 seconds, all not exceeding 10 seconds. This result demonstrates that the online adaptive mechanism can achieve safe updates by utilizing the observation consistency of measured nodes and the physical residuals of the entire network without relying on the true voltage labels in the blind zone, thereby quickly restoring estimation accuracy after load, photovoltaic, and topology changes.

[0082] This embodiment uses NLL (Negative Log-Likelihood) and CRPS (Continuous Ranked Probability Score) to evaluate the predicted probability distribution. The smaller the values ​​of these two metrics, the better the predicted distribution fits the actual voltage. 95% PICP (Prediction Interval Coverage Probability) and MPIW (Mean Prediction Interval Width) are used to evaluate interval effectiveness. The closer the PICP is to 95% and the smaller the MPIW, the more concentrated the predicted intervals are while meeting the target coverage.

[0083] This embodiment uses the Brier score and ECE (Expected Calibration Error) to evaluate the probability of exceeding the limit. The smaller the value of both, the more consistent the predicted probability of exceeding the limit is with the actual frequency of exceeding the limit. The Spearman correlation coefficient, Kendall correlation coefficient, and Top-10% hit rate are used to evaluate the risk ranking. The larger the value, the closer the risk priority output by the model is to the actual risk order.

[0084] To evaluate the effectiveness of the model's output probability distribution, this embodiment constructs a 33-node radial low-voltage distribution network case study, setting up operating conditions such as heavy load, photovoltaic backfeed, measurement noise, outdated communication, and a small number of gross errors. The average voltage measurement coverage is 26.65%. The samples are divided into training, validation, and independent test sets with 1500 / 700 / 800 scenarios, and the probability scale is calibrated only using the validation set. Five fixed random seeds are used for repeated experiments, with each test containing approximately 19,400 blind zone node samples. The average limit violation rate of the test set is 2.30%, and the low-voltage threshold and high-voltage threshold are 0.95 pu and 1.05 pu, respectively. All results are expressed as the mean ± standard deviation of the five experiments.

[0085] Table 5. Experimental results of probability distribution and probability of exceeding limits (mean ± standard deviation of 5 trials) As shown in Table 5, when using the complete technical solution, the NLL is -5.166±0.016, the CRPS is 0.000731±0.000014 pu, the 95% prediction interval coverage is 94.77%±0.31%, and the average interval width is 0.00513±0.00014 pu. Compared with the probabilistic graphical model without layer-by-layer confidence gating, the CRPS is reduced by approximately 39.2%, the average interval width is reduced by approximately 39.9%, the Brier score is reduced by approximately 36.1%, and the ECE is reduced by approximately 53.2%. Therefore, layer-by-layer confidence gating not only improves the accuracy of point estimation but also suppresses the cumulative contamination of low-confidence measurements during multi-hop propagation, making the probability distribution more concentrated while maintaining target coverage.

[0086] Compared to the model where confidence is used only as the initial continuous input, the complete technical solution reduces CRPS by approximately 12.1%, average interval width by approximately 10.9%, Bridge score by approximately 7.6%, and ECE by approximately 26.3%. The Spearman correlation coefficient increases from 0.9849 to 0.9899, ​​and the Kendall correlation coefficient increases from 0.9059 to 0.9176, indicating that recalculating confidence gating at each graph propagation layer is superior to using confidence only once at the input. Compared to the model without the DistFlow physical baseline, the complete technical solution reduces CRPS by approximately 2.3%, average interval width by approximately 2.8%, Bridge score by approximately 8.8%, and ECE by approximately 27.0%, demonstrating that the physical baseline can further constrain the probability distribution of blind nodes and improve the calibration of out-of-limit probabilities.

[0087] Table 6. Results of the risk ranking experiment (mean ± standard deviation of 5 trials) Spearman ρ represents the Spearman rank correlation coefficient. It measures the degree of monotonic correlation between two variables, typically ranging from -1 to 1; a larger absolute value indicates a stronger correlation. Kendall τ represents the Kendall rank correlation coefficient. It also measures the degree of agreement between two variables, with values ​​ranging from -1 to 1.

[0088] As shown in Table 6, the Spearman correlation coefficient of the complete technical solution is 0.9899±0.0005, the Kendall correlation coefficient is 0.9176±0.0027, and the hit rate of the Top-10% high-risk nodes is 95.68%±0.58%. This indicates that the output over-limit probability and comprehensive risk value can accurately reflect the true risk order, allowing maintenance personnel to prioritize checking high-risk blind spot nodes instead of relying solely on single voltage point estimation or fixed threshold alarms. For continuous probability density, the density value can be greater than 1, so NLL can be negative, and its evaluation still indicates that the smaller the value, the better the probability distribution.

[0089] Based on the above results, this scheme has the following technical advantages: First, when measurement coverage is insufficient or measurement quality is inconsistent, the observation representation and prior representation are propagated separately and fused through layer-by-layer confidence gating, which can reduce the propagation impact of abnormal measurements on continuous blind zones. Second, the DistFlow physical baseline provides blind zone nodes with constraints consistent with network topology and power distribution, making the probability interval more concentrated and the over-limit probability easier to calibrate. Third, the simultaneous output of probability distribution, over-limit probability, and risk ranking can provide a quantitative basis for voltage over-limit early warning, measurement device maintenance, and reactive power regulation priority. Fourth, the online adaptive mechanism can restore estimation accuracy after load, photovoltaic, and topology changes, thereby improving the applicability of the method under different operating conditions.

[0090] Another embodiment of the present invention provides a system for determining the operating status of a power distribution network. For details, please refer to [link to relevant documentation]. Figure 2 , Figure 2 The diagram shown illustrates the structure of a power distribution network operation status determination system according to one embodiment of the present invention, comprising: The acquisition module 11 is used to acquire the topology map and real-time operation data of the target distribution network, and divide each node in the topology map into a first measurable node set and a second unmeasurable node set. The credibility module 12 is used to determine the credibility of the measurement data of the first measurable node set based on the historical operation data of the target distribution network, and to extract the measurement operation data of the first measurable node set from the real-time operation data. The determination module 13 is used to determine the first initial prediction running data corresponding to the first measurable node set and to determine the second initial prediction running data corresponding to the second unmeasurable node set. The fusion module 14 is used to fuse the measurement operation data corresponding to the first measurable node set and the first initial prediction operation data based on the reliability of the measurement data to obtain the first node operation data. The prediction module 15 is used to make predictions based on the first node operation data and the second initial predicted operation data to obtain the target real-time operation status of the target distribution network. The target real-time operation status includes at least the predicted operation data of each node of the target distribution network and the corresponding confidence level of the predicted operation data.

[0091] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0092] Accordingly, embodiments of the present invention provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform steps in the power distribution network operation status determination method of the above embodiments, for example... Figure 1 Steps S1 to S5 as described above.

[0093] In determining the real-time operating status of a target distribution network, this invention does not simply rely on historical operating data for load curve extrapolation. Instead, it first acquires the topology and real-time operating data of the target distribution network, and divides each node in the topology into a first measurable node set and a second unmeasurable node set. Then, based on the acquired historical operating data of the target distribution network, it determines the reliability of the measurement data of the first measurable node set, and extracts the measurement operating data of the first measurable node set from the real-time operating data. At the same time, it determines the first initial predicted operating data corresponding to the first measurable node set and the second initial predicted operating data corresponding to the second unmeasurable node set. Based on this, it fuses the measurement operating data corresponding to the first measurable node set and the first initial predicted operating data based on the reliability of the measurement data to obtain the first node operating data. Based on the first node operating data and the second initial predicted operating data, it makes a prediction to obtain the target real-time operating status of the target distribution network. This target real-time operating status includes at least the predicted operating data of each node of the target distribution network and the corresponding reliability of the predicted operating data. Therefore, in the face of sudden changes in distribution network load, the embodiments of the present invention fuse and correct real-time measured operation data and initial predicted operation data by measuring the reliability of measurement data, thereby reducing the excessive reliance on historical operation data. This enables the estimation results of unmeasured nodes to truly reflect the current operation status, thereby obtaining accurate real-time operation status, providing an accurate data foundation for subsequent operations such as fault location, and ensuring the operation safety and reliability of the distribution network.

[0094] The above embodiments are preferred embodiments of the present invention. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principle of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for determining the operating status of a power distribution network, characterized in that, include: The topology map and real-time operation data of the target distribution network are obtained, and each node in the topology map is divided into a first measurable node set and a second unmeasurable node set. The reliability of the measurement data of the first measurable node set is determined based on the historical operation data of the target distribution network, and the measurement operation data of the first measurable node set is extracted from the real-time operation data. Determine the first initial prediction running data corresponding to the first measurable node set, and determine the second initial prediction running data corresponding to the second unmeasurable node set; Based on the reliability of the measurement data, the measurement operation data corresponding to the first measurable node set and the first initial prediction operation data are fused to obtain the first node operation data; Based on the first node's operating data and the second initial predicted operating data, a prediction is made to obtain the target real-time operating status of the target distribution network. The target real-time operating status includes at least the predicted operating data of each node of the target distribution network and the corresponding reliability of the predicted operating data.

2. The method for determining the operating status of a power distribution network as described in claim 1, characterized in that, The step of determining the reliability of the measurement data of the first measurable node set based on the acquired historical operating data of the target distribution network includes: The sub-historical operating data of each first measurable node in the first measurable node set is determined from the historical operating data. Based on a preset standard, the sub-historical operation data corresponding to each first measurable node is analyzed to obtain the reliability of the sub-measurement data. The preset standard includes at least the accuracy of the measurement device, communication latency, and data staleness. The reliability of the measurement data is determined based on the reliability of each of the sub-measurement data.

3. The method for determining the operating status of a power distribution network as described in claim 1, characterized in that, The step of determining the first initial prediction running data corresponding to the first measurable node set and determining the second initial prediction running data corresponding to the second unmeasurable node set includes: Based on the real-time operating data and topology diagram of the target distribution network, power flow calculation is performed to obtain the operating status baseline value of each node; Based on the typical operating scenarios of the target distribution network, the benchmark values ​​of each operating state are processed to obtain the first initial predicted operating data and the second initial predicted operating data.

4. The method for determining the operating status of a power distribution network as described in claim 1, characterized in that, The step of predicting the target real-time operating status of the target distribution network based on the first node's operating data and the second initial predicted operating data includes: The first node's running data and the second initial prediction running data are input into the target prediction model for processing to obtain the target's real-time running status. The processing procedure of the target prediction model includes: The initial processing state of the topology graph is determined based on the first node's running data, the second initial prediction running data, and the reliability of the measurement data. Based on the initial processing state, multi-hop propagation processing is performed, and the real-time running state of the target is obtained based on the processing results.

5. The method for determining the operating status of a distribution network as described in claim 4, characterized in that, The step of determining the initial processing state of the topology graph based on the first node's running data, the second initial prediction running data, and the reliability of the measurement data includes: The node operation data of each node is determined based on the first node operation data and the second initial predicted operation data; The propagation credibility of each first measurable node in the first measurable node set is determined based on the credibility of the measured data. Based on the topology graph, the neighbor attention weights between each of the first measurable nodes are determined; The initial processing state is determined based on the running data of each node, the propagation credibility of each node, and the attention weight of each neighbor.

6. The method for determining the operating status of a power distribution network as described in claim 5, characterized in that, The multi-hop propagation process based on the initial processing state, and the obtaining of the target's real-time running state based on the processing results, includes: The multi-hop propagation process includes N layers of propagation processing, where N is a positive integer; In the i-th layer of the N-layer propagation process, the output of the (i-1)-th layer propagation process is processed according to the node operation data of the neighboring nodes, the current propagation credibility of the sending node, and the neighbor attention weight to obtain the output of the i-th layer propagation process, where 1≤i≤N; When i=1, the output of the (i-1)th layer propagation process is the initial processing state; The real-time operating status of the target is determined by the output of the Nth layer propagation process.

7. The method for determining the operating status of a distribution network as described in claim 1, characterized in that, The target real-time operating status includes at least the predicted operating data of each node in the target distribution network and the corresponding reliability of the predicted operating data, including: The first node's running data and the second initial prediction running data are processed using a hybrid density technique to obtain the corresponding confidence level of the prediction running data; Based on the predicted running data and the confidence level, the real-time running status of the target is determined.

8. The method for determining the operating status of a distribution network as described in claim 1, characterized in that, After predicting the target real-time operating state of the target distribution network based on the first node operating data and the second initial predicted operating data, wherein the target real-time operating state includes at least the predicted operating data of each node of the target distribution network and the corresponding confidence level of the predicted operating data, the method further includes: The probability of exceeding the limit for each node is determined based on the predicted operational data. Based on each of the aforementioned exceedance probabilities, the urgency of regulation for the target distribution network is determined; Based on the aforementioned urgency of regulation and the probabilities of exceeding the limits, the target regulation scheme is obtained.

9. The method for determining the operating status of a distribution network as described in claim 8, characterized in that, Determining the probability of exceeding the limit for each node based on the predicted operational data includes: Determine the operational boundaries of each of the aforementioned nodes; Based on the operating boundary and the predicted operating data, the mutual exclusion probability of the three states of each node is determined, and the three states include low voltage state, overvoltage state and normal state. The corresponding out-of-limit probability is determined based on each of the mutual exclusion probabilities.

10. A system for determining the operating status of a power distribution network, characterized in that, include: The acquisition module is used to acquire the topology map and real-time operation data of the target distribution network, and divide each node in the topology map into a first measurable node set and a second unmeasurable node set. The credibility module is used to determine the credibility of the measurement data of the first measurable node set based on the historical operation data of the target distribution network, and to extract the measurement operation data of the first measurable node set from the real-time operation data. The determination module is used to determine the first initial prediction running data corresponding to the first measurable node set and to determine the second initial prediction running data corresponding to the second unmeasurable node set. The fusion module is used to fuse the measurement operation data corresponding to the first measurable node set and the first initial prediction operation data based on the reliability of the measurement data to obtain the first node operation data. The prediction module is used to make predictions based on the first node's operating data and the second initial predicted operating data to obtain the target real-time operating status of the target distribution network. The target real-time operating status includes at least the predicted operating data of each node of the target distribution network and the corresponding reliability of the predicted operating data.

Images