A distribution network load transaction early warning method and device

Abstract

The application provides a power distribution network load abnormality early warning method and device, relates to the technical field of load abnormality early warning, and comprises the following steps: collecting power consumption parameters of each node in a power distribution network in real time, forming a whole network node state matrix at the current time, sliding updating the node state matrix in a continuous time window, and calculating the trajectory entropy of different nodes; constructing a state confluence area, identifying a complete abnormal event propagation chain graph, and cross- verifying the abnormal event propagation chain graph with a preset impact load characteristic rule. Based on the whole network node state matrix and the trajectory entropy calculation, the spatiotemporal correlation characteristics of load evolution can be dynamically captured; through state confluence area identification and abnormal propagation chain graph backtracking, the source node of the electric vehicle load aggregation can be accurately located, the abnormal diffusion path and the influence range can be tracked; and the set cross-verification mechanism can effectively distinguish different types of impact loads such as electric vehicle charging and capacitor switching.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of load anomaly early warning, in particular to a distribution network load anomaly early warning method and device. BACKGROUND

[0002] With the large-scale access of electric vehicles, the distribution network is facing the severe challenge of new type of impact load. The traditional load anomaly detection method mainly relies on the threshold value judgment of single node voltage and current, and it is difficult to capture the multi-node space-time coupling power mutation caused by the concentrated charging of electric vehicles. Therefore, timely and effective identification of load anomaly in the distribution network has become a problem to be solved in the current electric vehicle industry.

[0003] In the prior art, the disclosure number CN106849356A discloses that by compiling a distribution network load anomaly judgment module on the basis of the OPEN3000 main network regulation and control system, automatic searching and judging of the distribution network load anomaly circuit is realized. However, this method does not consider the application scenario of electric vehicle power connection. The space-time randomness of electric vehicle access leads to high dynamicization of load impact characteristics. The existing steady-state threshold rule cannot distinguish between normal fluctuations and real anomalies. The electrical coupling relationship between distribution network nodes is complex. Local anomalies are easy to spread through the topology path, and there is a lack of complete traceability ability for anomaly propagation chain. Impact load events are often accompanied by active-reactive coordinated fluctuations, and traditional current or power single indicator detection is easy to produce false alarms due to noise interference. The above problems lead to the difficulty of the distribution network in realizing the rapid positioning and accurate classification of abnormal events in the high penetration scenario of electric vehicles, which restricts the active early warning and safety control ability. Therefore, it is necessary to consider the dynamic capture of load anomaly conditions in multiple dimensions.

[0004] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] The purpose of the present application is to provide a distribution network load anomaly early warning method and device to solve the problems raised in the background.

[0006] To achieve the above purpose, the present application provides the following technical scheme:

[0007] A distribution network load anomaly early warning method, the specific steps comprising:

[0008] S1: Real-time acquisition of power consumption parameters of each node in the distribution network, the power consumption parameters including voltage, active power, reactive power and current, construction of a multi-dimensional state vector of each node, and formation of a full network node state matrix at the current time with node topology number as the index, the state matrix representing the overall operation state of the distribution system at a specific time;

[0009] S2: The node state matrix is ​​updated by sliding within a continuous time window to generate the state evolution trajectory of each node within the time window, constructing a set of node trajectories, and calculating the trajectory entropy of different nodes to capture the dynamic evolution characteristics of the distribution network load state.

[0010] S3: Perform aggregation analysis on the set of state trajectories to identify the aggregation of multiple node trajectories in space and time. When the preset aggregation threshold is met, determine the corresponding region as the state confluence region. The state confluence region is an abnormal load aggregation region where electric vehicles are concentratedly connected.

[0011] S4: Backtrack each node in the state confluence region and its temporal evolution trajectory, identify the starting node, expansion path and impact boundary of the abnormal evolution, and construct a complete abnormal event propagation chain map, which includes the abnormal starting node, the abnormal impact chain and the event propagation duration;

[0012] S5: Cross-validate the abnormal event propagation chain map with the preset impact load characteristic rules. When the rule matching conditions are met, confirm that the abnormality type is an impact load abnormality caused by the centralized access of electric vehicles.

[0013] Furthermore, the current state matrix of all network nodes is formed, and the specific process is as follows:

[0014] by The sampling interval is used to sample the power consumption parameters, and the number of sampling points is set to [number]. Continuously collect power consumption parameters from each node, including the topology number of each node. Based on progressively increasing encoding from the power supply end to the load end; at each sampling point within the time window , will node voltage Active power reactive power and current The four-dimensional state vectors are stacked vertically in node number order to form the entire network node state matrix. Its dimensions are Its rows correspond to nodes, and its columns represent voltage, active power, reactive power, and current parameters in sequence.

[0015] Furthermore, the process of generating the state evolution trajectory of each node within the time window to construct the node trajectory set includes the following steps:

[0016] A sliding update is performed on the state matrix of all network nodes to construct the energy coefficients of current and voltage, as well as the power coefficients of active and reactive power. The dimensionality of the state evolution trajectory equation is reduced to construct a set of node trajectories.

[0017] ;

[0018] ;

[0019] wherein, denotes the steepness coefficient of the electric quantity coefficient; denotes the voltage and the current electric quantity; denotes the power coefficient of the active power and the reactive power ;

[0020] The node trajectory set after dimension reduction is:

[0021] ;

[0022] wherein, denotes the trajectory matrix of the node .

[0023] Further, the calculation of the trajectory entropy of different nodes captures the dynamic evolution characteristics of the power grid load state, including the following steps:

[0024] Calculate the mean and standard deviation for each parameter column in , and the standardized trajectory matrix is:

[0025] ;

[0026] wherein, denotes the standardized electric quantity coefficient ; denotes the standardized power coefficient ;

[0027] Calculate the covariance matrix:

[0028] ;

[0029] wherein, denotes the transpose matrix of ; denotes the covariance matrix of the th node;

[0030] Solve the eigenvalues and eigenvectors of the covariance matrix:

[0031] ;

[0032] wherein, denotes the eigenvalue of the th power consumption parameter; denotes the eigenvector of the th power consumption parameter;

[0033] Calculate trajectory entropy:

[0034] ;

[0035] in,

[0036] ;

[0037] in, Indicates the index of the parameter variable; Represents a node The trajectory entropy; Indicates the first Normalized weights for each feature value;

[0038] if This indicates that the power consumption parameters are approaching a steady state.

[0039] if This indicates that the power consumption parameters are trending abnormally.

[0040] Furthermore, identifying the spatial and temporal aggregation of multiple node trajectories includes the following steps:

[0041] Obtain the trajectory entropy sequence of all nodes within the time window, and standardize all... sequence:

[0042] ;

[0043] in, Represents a node The mean trajectory entropy within the time window; Represents a node Standard deviation of trajectory entropy within the time window; express The standardized value;

[0044] The mutual information entropy method is used to calculate the different nodes. and Dynamic similarity score of sequences :

[0045] ;

[0046] Among them, if A score approaching 1 indicates that the two nodes have highly synchronized evolution patterns; a score approaching 0 indicates that their evolution patterns are unrelated. express and Mutual information; express Information entropy; express Information entropy; express The standardized value;

[0047] Based on dynamic similarity score Construct the matrix: Furthermore, based on the power grid topology, the spatial proximity relationship of nodes is defined, and an adjacency matrix is ​​constructed. ,in Represents a node and Directly connected, otherwise Combined with similarity matrix Adjacency Matrix Construct the fused spatiotemporal feature matrix:

[0048] ;

[0049] Node groups are formed using spectral clustering algorithm: , ,express One candidate merging region; This represents element-wise multiplication; Represents a node With nodes The spatiotemporal fusion matrix; Indicates the sequence number of the candidate merging region;

[0050] For each cluster Calculate the overall score:

[0051] ;

[0052] in,

[0053] ;

[0054] ;

[0055] ;

[0056] in, Indicates the first The overall score of each cluster; Indicates the strength of temporal consistency; Indicates spatial clustering; Weights representing the strength of temporal consistency; Weights representing spatial clustering; Indicates group The total number of nodes included; Indicates the spatial distance between nodes;

[0057] if , then mark as a state confluence region, where, represents a preset comprehensive score threshold.

[0058] Further, a complete abnormal event propagation chain graph is constructed, specifically including the following steps:

[0059] For each node in the state confluence region, the abnormal starting point is located by the following steps:

[0060] Calculate the local mutation index of node in the first window:

[0061] ;

[0062] where, represents the local mutation index of node in the first window; represents the average trajectory entropy of node in the first window; represents the standard deviation of the trajectory entropy sequence of node from the first window to the first window; represents the window number index;

[0063] If , then mark as an abnormality triggered in the first window; represents a preset local mutation threshold, and the Granger causality test is used to verify whether the time series significantly affects other nodes;

[0064] Establish a dynamic propagation rule for extended path tracking, and the propagation from node to node needs to meet:

[0065] ;

[0066] represents the number of windows passed from node to node ; represents the power grid fluctuation propagation speed; represents the minimum topological path length of node and node ; ​Indicates the window time normalization factor; Indicates the sampling interval; Indicates the number of sampling points;

[0067] If node In the The window is marked as abnormal, and the correlation coefficient of the trajectory entropy between windows satisfies:

[0068] ;

[0069] Then mark For effective transmission pathways, Represents a node In the The average standardized trajectory entropy within each window; Represents a node In the Standardized trajectory entropy within a time window; Represents the Pearson correlation coefficient; iterative expansion continues until no new node meets the condition;

[0070] If node In extended path tracing, the following conditions must be met:

[0071] ;

[0072] in, Represents a node Peak trajectory entropy across all windows; This represents the preset trajectory entropy boundary threshold. This indicates the total number of time windows.

[0073] Furthermore, the abnormal event propagation chain map is cross-validated with the preset impact load characteristic rules, including the following steps:

[0074] Check whether the preset impact load characteristic rules are triggered from the abnormal propagation chain graph. The preset impact load characteristic rules include multi-node power surge, voltage linkage, and dual index jump.

[0075] Let the set of propagation chain nodes be: The set of propagation chain paths is as follows: ; The total number of nodes in the propagation chain;

[0076] The multi-node power surge includes counting the number of active power surges at each node in the propagation chain:

[0077] ;

[0078] in,

[0079] ;

[0080] This represents the minimum number of nodes required to trigger the active power surge rule; Indicates an indicator function; Represents a node In the The difference between the active power at each sampling point and the active power at the previous sampling point; Represents a node In the Active power at each sampling point; Point In the Active power at each sampling point; This represents the evaluation index value for a sudden increase in multiple nodes;

[0081] The voltage linkage includes the propagation path. Calculate the correlation coefficient of node voltages Phase difference:

[0082] ;

[0083] Represents a node With nodes The Pearson correlation coefficient; This indicates the phase difference of voltage fluctuations; Indicates the path proportion threshold; This indicates the voltage linkage evaluation index value;

[0084] The dual-metric leap includes improvements at each node in the propagation chain. Detect its active power With reactive power ,

[0085] ;

[0086] in, This represents the threshold ratio of active power to reactive power. Indicates the node proportion threshold; This indicates the evaluation index value for the leap in both indicators;

[0087] The confidence level is calculated by determining the number of nodes and the proportion of paths in the statistical propagation chain that satisfy the above-preset impact load characteristic rules. The calculation formula is as follows:

[0088] ;

[0089] in, Indicates the mean of the indicator; if If so, it is determined that the propagation chain was caused by the impact load, triggering an early warning.

[0090] The present invention also provides a distribution network load anomaly early warning device, the early warning device being used to execute the above-mentioned early warning method, comprising:

[0091] The data acquisition unit is used to collect the power consumption parameters of each node in the power distribution network in real time. The power consumption parameters include voltage, active power, reactive power and current. It constructs a multi-dimensional state vector for each node and forms a state matrix of all nodes in the network at the current moment by indexing the node topology number. The state matrix represents the overall operating state of the power distribution system at a specific moment.

[0092] The trajectory identification unit is used to perform sliding updates on the node state matrix within a continuous time window, generate the state evolution trajectory of each node within the time window, construct a set of node trajectories, and calculate the trajectory entropy of different nodes to capture the dynamic evolution characteristics of the distribution network load state.

[0093] The node aggregation unit is used to perform aggregation analysis on the set of node trajectories, identify the aggregation of multiple node trajectories in space and time, and determine the corresponding area as the state confluence area when the preset aggregation threshold is met. The state confluence area is an abnormal load aggregation area where electric vehicles are concentratedly accessing the area.

[0094] An anomaly backtracking unit is used to backtrack each node in the state confluence region and its temporal evolution trajectory, identify the starting node, expansion path and impact boundary of the anomaly evolution, and construct a complete anomaly event propagation chain map, which includes the anomaly starting node, the anomaly impact chain and the event propagation duration;

[0095] The mobile analysis unit cross-validates the abnormal event propagation chain map with the preset impact load characteristic rules. When the rule matching conditions are met, the anomaly type is confirmed as an impact load anomaly caused by the centralized access of electric vehicles.

[0096] Compared with the prior art, the beneficial effects of the present invention are:

[0097] This invention systematically solves the problem of electric vehicle impact load detection through multi-dimensional state matrix construction, trajectory entropy analysis, and anomaly propagation chain modeling. Based on the state matrix of all network nodes and trajectory entropy calculation using a sliding time window, it can dynamically capture the spatiotemporal correlation characteristics of load evolution, significantly improving the sensitivity to the disordered charging-ordered propagation mode of electric vehicle clusters. Through state confluence area identification and anomaly propagation chain map backtracking, it can accurately locate the source node of electric vehicle load aggregation, track the abnormal diffusion path, and the scope of impact. By integrating active and reactive power dual-index jump rules, voltage linkage characteristics, and cross-validation mechanisms for multi-node power surges, it can effectively distinguish different types of impact loads such as electric vehicle charging and capacitor switching, reducing the risk of protection maloperation caused by disordered electric vehicle access. Attached Figure Description

[0098] Figure 1 This is a schematic diagram of the distribution network load anomaly early warning method in this invention;

[0099] Figure 2 This is a schematic diagram of the power surge data of multiple nodes in the distribution network load anomaly early warning system of this invention;

[0100] Figure 3 This is a schematic diagram of voltage linkage data in the early warning of distribution network load anomalies in this invention;

[0101] Figure 4 This is a schematic diagram of the dual-indicator jump data in the distribution network load anomaly early warning system of this invention;

[0102] Figure 5 This is a structural block diagram of the distribution network load anomaly early warning device in this invention. Detailed Implementation

[0103] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0104] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0105] Example:

[0106] Please see Figures 1-4 The present invention provides a technical solution:

[0107] A method for early warning of abnormal load changes in a distribution network, comprising the following steps:

[0108] S1: Real-time acquisition of power consumption parameters of each node in the power distribution network, including voltage, active power, reactive power and current, constructing a multi-dimensional state vector of each node, and forming a state matrix of all nodes in the network at the current moment by indexing the node topology number, and characterizing the overall operating state of the power distribution system at a specific moment through the state matrix.

[0109] The specific process for forming the current state matrix of all network nodes is as follows:

[0110] by The sampling interval is used to sample the power consumption parameters, and the number of sampling points is set to [number]. Continuously collect power consumption parameters from each node, including the topology number of each node. Based on progressively increasing encoding from the power supply end to the load end; at each sampling point within the time window , will node voltage Active power reactive power and current The four-dimensional state vectors are stacked vertically in node number order to form the entire network node state matrix. Its dimensions are Its rows correspond to nodes, and its columns represent voltage, active power, reactive power, and current parameters in sequence.

[0111] The topology number of each node The encoding is based on its spatial location within the power grid hierarchy, specifically as follows:

[0112] Topology number of each node Sequential coding can be performed according to the physical hierarchy of the power grid from the power source to the load. That is, starting from the main substation or generation node on the power source side, incremental integer numbers are assigned sequentially to the downstream distribution transformer, branch lines, and finally to the physical location of the end user load, so that nodes with smaller numbers are closer to the upstream of the power source level, and nodes with larger numbers are located downstream of the load level. Adjacent feeder nodes at the same level are numbered consecutively.

[0113] The node is specifically a topological unit in the distribution network that can independently measure power consumption parameters. The power consumption parameters are specifically the current passing through the node, the voltage at both ends, the active power, and the reactive power.

[0114] The specific logic for forming the current state matrix of all network nodes is as follows:

[0115] In time At any given time, four power consumption parameters are collected at each monitoring point, and the node state vector can be represented as:

[0116] ;

[0117] in, Indicates the node topology number; Indicates the first The node The voltage at each sampling point; Indicates the first The node Active power at each sampling point; Indicates the first The node Reactive power at each sampling point; Indicates the first The node The current at each sampling point; Indicates the first The node State vector of each sampling point; Indicates the node index; Indicates the sampling point number;

[0118] Based on node state vectors, constructing The state matrix of all nodes in the network at any given time:

[0119] ;

[0120] in, Represents the state matrix of all nodes in the network; Indicates the total number of nodes; Indicates the first The node The voltage at each sampling point; Indicates the first The node Active power at each sampling point; Indicates the first The node The sampling point of the first sampling point The reactive power of each node; Indicates the first The sampling point of the first sampling point The current at each node.

[0121] S2: Slide update the node state matrix within a continuous time window to generate the state evolution trajectory of each node within the time window, construct a set of node trajectories, and calculate the trajectory entropy of different nodes to capture the dynamic evolution characteristics of the distribution network load state.

[0122] The process of generating the state evolution trajectory of each node within the time window and constructing the node trajectory set includes the following steps:

[0123] The number of sampling points is The state matrix of all nodes in the network is updated by sliding. For any node... Its state evolution trajectory equation within the time window is:

[0124] ;

[0125] Represents a node The trajectory matrix; Represents a node The The voltage at each sampling point; Represents a node The Active power at each sampling point; Represents a node The Reactive power at each sampling point; node The The voltage at each sampling point;

[0126] By constructing the quantity coefficients of current and voltage, as well as the power coefficients of active and reactive power, the dimensionality of the state evolution trajectory equations is reduced, and a set of node trajectories is constructed.

[0127] ;

[0128] ;

[0129] in, Indicates the kurtosis coefficient of the power coefficient; Indicates voltage and current Power consumption coefficient; Indicates active power and reactive power The power coefficient;

[0130] In the above formula, the dependent variable Reflects the first The node at the th The electricity coefficient at each sampling point is essentially obtained through... With current The instantaneous product divided by a basis A nonlinear adjustment factor for the ratio is used to suppress distortion in power calculations and improve the robustness of results when the voltage-current ratio is abnormal, such as a sudden voltage drop or current surge; independent variable and Directly affects the dependent variable: on the one hand, and An increase in will directly boost the numerator. On the one hand, they show a positive correlation; on the other hand, the ratio of the two adjusts the overall result through the exponent in the denominator. When the voltage increases or the current decreases, the exponent approaches 0 and the denominator approaches 1, and the power calculation retains its original value; conversely, if the current increases sharply, the ratio will shrink, the exponent will increase significantly, and the denominator will expand, making the ratio... It is actively suppressed, avoiding numerical overflow or misjudgment in extreme cases of traditional algorithms;

[0131] The dependent variable in the above formula Indicates the first The node at the th The power coefficient at each sampling point represents the active power of that node. Divide by its apparent power magnitude Thus The absolute value is mapped to the interval between 0 and 1, reflecting the relative proportion of active power in the total apparent power. Normalization eliminates the influence of different equipment capacities or load levels on power analysis, facilitating horizontal comparisons of power utilization efficiency; independent variables... and Together they determine the dependent variable: An increase in will directly boost the numerator and denominator through changes in the same direction. However, because the denominator contains and , right The effects show a positive correlation; while An increase in will cause the denominator to expand, making The decrease indicates a negative correlation.

[0132] The set of node trajectories after dimensionality reduction is as follows:

[0133] ;

[0134] in, Represents a node The trajectory matrix.

[0135] The calculation of trajectory entropy at different nodes to capture the dynamic evolution characteristics of the distribution network load state includes the following steps:

[0136] right The mean and standard deviation of each parameter column are calculated, and the standardized trajectory matrix is:

[0137] ;

[0138] in, express Standardized power coefficient; express Standardized power coefficient;

[0139] Calculate the covariance matrix:

[0140] ;

[0141] in, express The transpose of the matrix; Indicates the first The covariance matrix of the nodes;

[0142] Find the eigenvalues ​​and eigenvectors of the covariance matrix:

[0143] ;

[0144] in, Indicates the first Characteristic values ​​of individual electricity consumption parameters; Indicates the first Feature vector of each electricity consumption parameter;

[0145] Calculate trajectory entropy:

[0146] ;

[0147] in,

[0148] ;

[0149] in, Indicates the index of the parameter variable; Represents a node The trajectory entropy; Indicates the first Normalized weights for each feature value;

[0150] if This indicates that the power consumption parameters are approaching a steady state.

[0151] if This indicates that the power consumption parameters are trending abnormally.

[0152] S3: Perform aggregation analysis on the set of state trajectories to identify the aggregation of multiple node trajectories in space and time. When the preset aggregation threshold is met, determine the corresponding region as the state confluence region. The state confluence region is an abnormal load aggregation region where electric vehicles are concentratedly connected.

[0153] The identification of the aggregation of multiple node trajectories in space and time includes the following steps:

[0154] Obtain the trajectory entropy sequence of all nodes within the time window, and standardize all... sequence:

[0155] ;

[0156] in, Represents a node The mean trajectory entropy within the time window; Represents a node Standard deviation of trajectory entropy within the time window; express The standardized value;

[0157] The mutual information entropy method is used to calculate the different nodes. and Dynamic similarity score of sequences :

[0158] ;

[0159] Among them, if A score approaching 1 indicates that the two nodes have highly synchronized evolution patterns; a score approaching 0 indicates that their evolution patterns are unrelated. express and Mutual information; express Information entropy; express Information entropy; express The standardized value;

[0160] The mutual information entropy method is specifically as follows:

[0161] set up , ,in, and For category symbols, and The sets of symbols representing the two sequences, respectively. , ; Indicates the first Type symbols ; Indicates the first Type symbols ;

[0162] Calculate univariate probabilities and joint probabilities:

[0163] ;

[0164] ;

[0165] ;

[0166] Symbols The probability of occurrence; Symbols The probability of occurrence; express The joint probability;

[0167] Calculate information entropy:

[0168] ;

[0169] ;

[0170] express Information entropy; express Information entropy;

[0171] Calculate mutual information:

[0172] ;

[0173] express and Mutual information;

[0174] Based on dynamic similarity score Construct the matrix: Furthermore, based on the power grid topology, the spatial proximity relationship of nodes is defined, and an adjacency matrix is ​​constructed. ,in Represents a node and Directly connected, otherwise Combined with similarity matrix Adjacency Matrix Construct the fused spatiotemporal feature matrix:

[0175] ;

[0176] Node groups are formed using spectral clustering algorithm: , ,express One candidate merging region; This represents element-wise multiplication; Represents a node With nodes The spatiotemporal fusion matrix;

[0177] The spectral clustering algorithm is specifically as follows:

[0178] Obtain the fused spatiotemporal feature matrix: Construct a degree matrix, where the diagonal elements are the node degrees. Perform Laplace normalization on the degree matrix: ,in, Degree matrix; Represents the normalized Laplace matrix;

[0179] right Solve the characteristic equation: Before selecting The smallest non-zero eigenvalues: The corresponding feature vector is: Based on the feature vectors, a low-dimensional embedding matrix is ​​constructed. : ;

[0180] matrix Each row is used as a low-dimensional representation of a node. The clustering process is as follows: Random initialization Cluster centers, The number of merging regions is preset; the center point is iteratively updated, and nodes are assigned to the nearest center until convergence, outputting a set of node groups. Each group is a candidate merging region.

[0181] For each cluster Calculate the overall score:

[0182] ;

[0183] in,

[0184] ;

[0185] ;

[0186] ;

[0187] in, Indicates the first The overall score of each cluster; Indicates the strength of temporal consistency; Indicates spatial clustering; Weights representing the strength of temporal consistency; Weights representing spatial clustering; Indicates group The total number of nodes included; Indicates the spatial distance between nodes;

[0188] if Then mark This is the merging region, where... This indicates the preset overall score threshold.

[0189] S4: Backtrack each node in the state confluence region and its temporal evolution trajectory, identify the starting node, expansion path and impact boundary of the abnormal evolution, and construct a complete abnormal event propagation chain map, which includes the abnormal starting node, the abnormal impact chain and the event propagation duration;

[0190] The construction of a complete abnormal event propagation chain graph specifically includes the following steps:

[0191] For each node within the state merging region To locate the origin of the anomaly, follow these steps:

[0192] compute nodes In the Local mutation index within a window:

[0193] ;

[0194] in, Represents a node In the Local mutation index within a window; Represents a node In the The average trajectory entropy within each window; Represents a node No. The window to the first The standard deviation of the trajectory entropy of the sequence consisting of all trajectory entropies of each window; Indicates the window number index;

[0195] if Then mark In the An exception was triggered within a single window; This represents a preset local mutation threshold, which is verified using the Granger causality test. of Does the timing significantly affect other nodes?

[0196] The Granger causality test specifically refers to:

[0197] Set the target node as Verification required. To determine whether there is a causal relationship, the steps are as follows:

[0198] Construct an unconstrained model, including Historical information:

[0199] ;

[0200] Represents the intercept term; Represents a node Regression coefficient of its own historical values; Represents a node In the The trajectory entropy of a time window; Represents a node Regression coefficients of historical values;

[0201] Construct a constrained model, excluding Historical information:

[0202] ;

[0203] in, Indicates the maximum lag order; Represents the residual term;

[0204] use test:

[0205] ;

[0206] in, Indicates the total number of time windows; This represents the sum of squared residuals of a constrained model; This represents the sum of squared residuals of an unconstrained model.

[0207] if If the value exceeds the critical value, then it is considered... yes Granger's cause;

[0208] Establish dynamic propagation rules to extend path tracing, starting from nodes. To the node The spread of [something] needs to meet the following conditions:

[0209] ;

[0210] Indicates from node To the node The number of windows passed through; Indicates the propagation speed of power grid fluctuations; Represents a node With nodes The minimum topological path length; Indicates the window time normalization factor; Indicates the sampling interval; Indicates the number of sampling points;

[0211] If node In the The window is marked as abnormal, and the correlation coefficient of the trajectory entropy between windows satisfies:

[0212] ;

[0213] Then mark For effective transmission pathways, Represents a node In the The average standardized trajectory entropy within each window; Represents a node In the Standardized trajectory entropy within a time window; Represents the Pearson correlation coefficient; iterative expansion continues until no new node meets the condition;

[0214] The correlation coefficient is calculated as follows:

[0215] Collection Nodes and nodes The standardized trajectory entropy mean values ​​over multiple consecutive time windows form two sets of sequences:

[0216] sequence :node In the time window The mean of the standardized trajectory entropy:

[0217] ;

[0218] Sequence D: Node In the window The mean of the standardized trajectory entropy:

[0219] ;

[0220] in, This represents the number of data points.

[0221] Global standardization:

[0222] ;

[0223] in, Represents a sequence The mean; Represents a sequence The mean;

[0224] If node In extended path tracing, the following conditions must be met:

[0225] ;

[0226] in, Represents a node Peak trajectory entropy across all windows; This represents the preset trajectory entropy boundary threshold. This indicates the total number of time windows.

[0227] S5: Cross-validate the abnormal event propagation chain map with the preset impact load characteristic rules. When the rule matching conditions are met, confirm that the abnormality type is an impact load abnormality caused by the centralized access of electric vehicles.

[0228] The preset impact load characteristic rules include multi-node power surge, voltage linkage, and dual index jump;

[0229] Let the set of propagation chain nodes be: The set of propagation chain paths is as follows: ; Indicates the total number of nodes in the propagation chain;

[0230] The multi-node power surge includes counting the number of active power surges at each node in the propagation chain:

[0231] ;

[0232] in,

[0233] ;

[0234] This represents the minimum number of nodes required to trigger the active power surge rule; Indicates an indicator function; Represents a node In the The difference between the active power at each sampling point and the active power at the previous sampling point; Represents a node In the Active power at each sampling point; Point In the Active power at each sampling point; This represents the evaluation index value for a sudden increase in multiple nodes;

[0235] In this embodiment, 10 time points and 5 groups of nodes were selected to simulate the power surge of multiple nodes. The experimental measurement data are shown in Table 1.

[0236] Table 1: Multi-node power surge data detection table

[0237]

[0238] Table 1 shows that the active power of multiple nodes was collected and divided into 10 groups of data according to the time nodes. Among them, the active power of nodes 1 to 5 showed a sudden increase trend, which is consistent with the distribution network load change caused by the connection of electric vehicles, that is, the sudden increase of power of multiple nodes.

[0239] The voltage linkage includes the propagation path. Calculate the correlation coefficient of node voltages Phase difference:

[0240] ;

[0241] Represents a node With nodes The Pearson correlation coefficient; This indicates the phase difference of voltage fluctuations; Indicates the path proportion threshold; Represents a node The voltage; Represents a node The voltage; This indicates the voltage linkage evaluation index value;

[0242] ;

[0243] in,

[0244] ;

[0245] ;

[0246] Calculate covariance:

[0247] ;

[0248] express and covariance; Indicates the sampling point index variable; express The mean; express The mean;

[0249] Calculate the standard deviation:

[0250] ;

[0251] ;

[0252] express Standard deviation; express Standard deviation;

[0253] Calculate the mean:

[0254] ;

[0255] ;

[0256] Phase difference Represents a node With nodes The time lag of voltage fluctuations can be calculated using the cross-correlation function method:

[0257] ;

[0258] in, Indicates the time offset; Indicates the sampling interval;

[0259] In this embodiment, 10 time points and 5 sets of node voltages were selected for simulated voltage linkage, and the data measurements are shown in Table 2:

[0260] Table 2: Voltage Linkage Data Measurement Table

[0261]

[0262] Table 2 shows that the voltage of multiple nodes was collected and divided into 10 groups of data according to the time nodes. Among them, the voltage of nodes 1 to 5 showed the same upward trend, and the voltage values ​​of each node showed similar values ​​at the same time point, which is consistent with the distribution network load change caused by the connection of electric vehicles, i.e. voltage linkage.

[0263] The dual-metric leap includes improvements at each node in the propagation chain. Detect its active power With reactive power ,

[0264] ;

[0265] in, This represents the threshold ratio of active power to reactive power. Indicates the node proportion threshold; This indicates the evaluation index value for the leap in both indicators;

[0266] In this embodiment, 10 time points and 5 sets of node power ratios were selected to simulate the jump in the two indicators. The measurement data are shown in Table 3.

[0267] Table 3: Measurement Table of Two-Indicator Leap Data

[0268]

[0269] Table 3 shows that the voltage of multiple nodes was collected and divided into 10 groups of data according to the time nodes. Among them, the ratio of active power to reactive power in nodes 1 to 5 shows an upward trend, which is consistent with the distribution network load change caused by electric vehicle connection, i.e., the double index jump.

[0270] The confidence level is calculated by determining the number of nodes and the proportion of paths in the statistical propagation chain that satisfy the above-preset impact load characteristic rules. The calculation formula is as follows:

[0271] ;

[0272] in, Indicates the mean of the indicator; if If so, it is determined that the propagation chain was caused by the impact load, triggering an early warning.

[0273] Please see Figure 5 The present invention also provides a distribution network load anomaly early warning device, the early warning device being used to execute the above-mentioned early warning method, comprising:

[0274] The data acquisition unit is used to collect the power consumption parameters of each node in the power distribution network in real time. The power consumption parameters include voltage, active power, reactive power and current. It constructs a multi-dimensional state vector for each node and forms a state matrix of all nodes in the network at the current moment by indexing the node topology number. The state matrix represents the overall operating state of the power distribution system at a specific moment.

[0275] The trajectory identification unit is used to perform sliding updates on the node state matrix within a continuous time window, generate the state evolution trajectory of each node within the time window, construct a set of node trajectories, and calculate the trajectory entropy of different nodes to capture the dynamic evolution characteristics of the distribution network load state.

[0276] The node aggregation unit is used to perform aggregation analysis on the set of node trajectories, identify the aggregation of multiple node trajectories in space and time, and determine the corresponding area as the state confluence area when the preset aggregation threshold is met. The state confluence area is an abnormal load aggregation area where electric vehicles are concentratedly accessing the area.

[0277] An anomaly backtracking unit is used to backtrack each node in the state confluence region and its temporal evolution trajectory, identify the starting node, expansion path and impact boundary of the anomaly evolution, and construct a complete anomaly event propagation chain map, which includes the anomaly starting node, the anomaly impact chain and the event propagation duration;

[0278] The mobile analysis unit cross-validates the abnormal event propagation chain map with the preset impact load characteristic rules. When the rule matching conditions are met, the anomaly type is confirmed as an impact load anomaly caused by the centralized access of electric vehicles.

[0279] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0280] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0281] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0282] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for early warning of load anomalies in a distribution network, characterized in that, include: S1: Real-time acquisition of power consumption parameters of each node in the power distribution network, including voltage, active power, reactive power and current, constructing a multi-dimensional state vector of each node, and forming a state matrix of all nodes in the network at the current moment by indexing the node topology number, and characterizing the overall operating state of the power distribution system at a specific moment through the state matrix. S2: Slide update the node state matrix within a continuous time window to generate the state evolution trajectory of each node within the time window, construct a set of node trajectories, and calculate the trajectory entropy of different nodes to capture the dynamic evolution characteristics of the distribution network load state. S3: Perform aggregation analysis on the set of node trajectories to identify the aggregation of multiple node trajectories in space and time. When the preset aggregation threshold is met, determine the corresponding region as the state confluence region. The state confluence region is an abnormal load aggregation region where electric vehicles are concentratedly connected. S4: Backtrack each node in the state confluence region and its temporal evolution trajectory, identify the starting node, expansion path and impact boundary of the abnormal evolution, and construct a complete abnormal event propagation chain map, which includes the abnormal starting node, the abnormal impact chain and the event propagation duration; S5: Cross-validate the abnormal event propagation chain map with the preset impact load characteristic rules. When the rule matching conditions are met, confirm that the abnormality type is an impact load abnormality caused by the centralized access of electric vehicles.

2. The method for early warning of distribution network load anomalies according to claim 1, characterized in that, The specific process for forming the current state matrix of all network nodes is as follows: The power consumption parameters are sampled at sampling intervals of ΔT, and the number of sampling points is set to k. The power consumption parameters of each node are continuously collected, and the topology number i of each node is encoded by progressively increasing numbers from the power source to the load. Within each sampling point n in the time window, the voltage of node i is recorded. Active power reactive power and current The four-dimensional state vectors are stacked vertically in node number order to form the network-wide node state matrix M. (n) Its dimensions are N×4, with rows corresponding to nodes and columns representing voltage, active power, reactive power, and current parameters respectively.

3. The method for early warning of distribution network load anomalies according to claim 2, characterized in that, The process of generating the state evolution trajectory of each node within the time window and constructing the node trajectory set includes the following steps: A sliding update is performed on the state matrix of all network nodes to construct the energy coefficients of current and voltage, as well as the power coefficients of active and reactive power. The dimensionality of the state evolution trajectory equation is reduced to construct a set of node trajectories. Wherein, λ represents the kurtosis coefficient of the electric power coefficient; Indicates voltage and current Power consumption coefficient; Indicates active power and reactive power The power coefficient; The set of node trajectories after dimensionality reduction is as follows: Among them, T i (n) Let i represent the trajectory matrix of node i.

4. The method for early warning of distribution network load anomalies according to claim 1, characterized in that, The calculation of trajectory entropy at different nodes to capture the dynamic evolution characteristics of the distribution network load state includes the following steps: For T i (n) The mean and standard deviation of each parameter column are calculated, and the standardized trajectory matrix is: in, express Standardized power coefficient; express Standardized power coefficient; Calculate the covariance matrix: in, express The transpose of C; i Let represent the covariance matrix of the i-th node; Find the eigenvalues ​​and eigenvectors of the covariance matrix: Where, λ j v represents the characteristic value of the j-th electricity consumption parameter; j The feature vector representing the j-th electricity consumption parameter; Calculate trajectory entropy: in, Where l represents the parameter variable index; TEE i p represents the trajectory entropy of node i; j This represents the normalized weight of the j-th eigenvalue; If TEE i If the value is less than 0.3, it indicates that the power consumption parameters are approaching a steady state. If TEE i A value ≥0.3 indicates that the power consumption parameters are trending abnormally.

5. The method for early warning of distribution network load anomalies according to claim 1, characterized in that, The identification of the aggregation of multiple node trajectories in space and time includes the following steps: Obtain the trajectory entropy sequence of all nodes within the time window, and standardize all TEEs. i sequence: Where, μ i σ represents the mean entropy of the trajectory of node i within the time window; i This represents the standard deviation of the trajectory entropy of node i within the time window; means TEE i Standardized value; TEE i Represents the trajectory entropy of node i; The mutual information entropy method is used to calculate the different nodes. and The dynamic similarity score of the sequence S ij : Where, if S ij A score approaching 1 indicates that the two nodes have highly synchronized evolution patterns; a score approaching 0 indicates that their evolution patterns are unrelated. express and Mutual information; express Information entropy; express Information entropy; means TEE j The standardized value; Based on dynamic similarity score S ij Construct the matrix: S = [S ij Furthermore, based on the power grid topology, the spatial proximity relationship of nodes is defined, and an adjacency matrix A is constructed. ij A ij =1 indicates that nodes i and j are directly connected; otherwise, A ij =0; combining the similarity matrix S and the adjacency matrix A ij Construct the fused spatiotemporal feature matrix: W ij =S ij ⊙A ij Node groups are partitioned using spectral clustering algorithm: C m ={i1,i2,…,i m }, m = 1, 2, ..., M, representing M candidate merging regions; ⊙ represents element-wise multiplication; W ij represents the spatiotemporal fusion matrix of node i and node j; m represents the index of the candidate merging region; For each cluster C m Calculate the overall score: c m =w1a m +w2β m in, w1 + w2 = 1 Where, γ m α represents the overall score of the m-th cluster; m Indicates the strength of temporal consistency; β m w1 represents the spatial clustering strength; w2 represents the weight of the temporal consistency strength; n represents the weight of the spatial clustering strength. m Represents group C m The total number of nodes included; d ij Indicates the spatial distance between nodes; If γ m ≥γ th Then mark C m For the state confluence region, where γ th This indicates the preset overall score threshold.

6. The method for early warning of distribution network load anomalies according to claim 1, characterized in that, The construction of a complete abnormal event propagation chain graph specifically includes the following steps: For each node i within the state merging region m ∈C m To locate the origin of the anomaly, follow these steps: Calculate the local mutation index of node i within the p-th window: in, This represents the local mutation index of node i within the p-th window; This represents the mean trajectory entropy of node i within the p-th window; The standard deviation of the trajectory entropy is represented by the sequence of trajectory entropies from the (p-1)th window to the pth window of node i; p represents the window index. if Then mark i m ∈C m An exception is triggered within the p-th window; θ event This represents a preset local mutation threshold, which is verified using the Granger causality test. m of Does the timing significantly affect other nodes? To establish dynamic propagation rules for extended path tracing, the propagation from node i to node j must satisfy the following: Δp i→j This represents the number of windows traversed from node i to node j; v wave Indicates the propagation speed of power grid fluctuations; α represents the minimum topological path length between node i and node j; α represents the window time normalization factor; ΔT represents the sampling interval; k represents the number of sampling points; If node j is at the p+Δp i→j The window is marked as abnormal, and the correlation coefficient of the trajectory entropy between windows satisfies: Then mark i→j as a valid propagation path. This represents the mean of the standardized trajectory entropy of node i within the p-th window; This indicates that node j is at the p+Δp level. i→j Standardized trajectory entropy within a time window; ρ represents the Pearson correlation coefficient; iterative expansion until no new node meets the condition; If node j satisfies the following in extended path tracing: in, θ represents the peak value of the trajectory entropy of node j across all windows; boundary R represents the preset trajectory entropy boundary threshold; R represents the total number of time windows.

7. The method for early warning of distribution network load anomalies according to claim 1, characterized in that, The process of cross-validating the abnormal event propagation chain map with preset impact load characteristic rules includes the following steps: Check whether the preset impact load characteristic rules are triggered from the abnormal propagation chain graph. The preset impact load characteristic rules include multi-node power surge, voltage linkage, and dual index jump. Let the set of propagation chain nodes be: The set of propagation chain paths is as follows: B represents the total number of nodes in the propagation chain; The multi-node power surge includes counting the number of active power surges at each node in the propagation chain: in, W represents the minimum number of nodes required to trigger the active power surge rule; Indicates an indicative function; ΔP j (n) P represents the difference between the active power of node j at the nth sampling point and the active power at the previous sampling point; j (n) P represents the active power of node j at the nth sampling point; j (n-1) R1 represents the active power at point j in the (n-1)th sampling point; R1 represents the multi-node surge evaluation index value. The voltage linkage includes calculating the correlation coefficient ρ(V) of the node voltage for the propagation path i→j. i V j Phase difference: ρ(V i V j ) represents the Pearson correlation coefficient between node i and node j; Δt ij Indicates the phase difference of voltage fluctuations; X th R2 represents the path ratio threshold; R2 represents the voltage linkage evaluation index value. The dual-index jump includes detecting the active power P of each node j in the propagation chain. j (n) With reactive power Where, θ PQ Y represents the ratio threshold between active and reactive power; th R3 represents the threshold for the node ratio; R3 represents the value of the dual-indicator leap evaluation index. The confidence level is calculated by determining the number of nodes and the proportion of paths in the statistical propagation chain that satisfy the above-preset impact load characteristic rules. The calculation formula is as follows: Where θ represents the average value of the index; if θ ≥ 70%, the propagation chain is determined to be caused by the impact load, triggering an early warning.

8. A distribution network load anomaly early warning device, characterized in that: The early warning device is used to perform the early warning method according to any one of claims 1-7, including: The data acquisition unit is used to collect the power consumption parameters of each node in the power distribution network in real time. The power consumption parameters include voltage, active power, reactive power and current. It constructs a multi-dimensional state vector for each node and forms a state matrix of all nodes in the network at the current moment by indexing the node topology number. The state matrix represents the overall operating state of the power distribution system at a specific moment. The trajectory identification unit is used to perform sliding updates on the node state matrix within a continuous time window, generate the state evolution trajectory of each node within the time window, construct a set of node trajectories, and calculate the trajectory entropy of different nodes to capture the dynamic evolution characteristics of the distribution network load state. The node aggregation unit is used to perform aggregation analysis on the set of node trajectories, identify the aggregation of multiple node trajectories in space and time, and determine the corresponding area as the state confluence area when the preset aggregation threshold is met. The state confluence area is an abnormal load aggregation area where electric vehicles are concentratedly accessing the area. An anomaly backtracking unit is used to backtrack each node in the state confluence region and its temporal evolution trajectory, identify the starting node, expansion path and impact boundary of the anomaly evolution, and construct a complete anomaly event propagation chain map, which includes the anomaly starting node, the anomaly impact chain and the event propagation duration; The mobile analysis unit is used to cross-validate the abnormal event propagation chain map with the preset impact load characteristic rules. When the rule matching conditions are met, the abnormality type is confirmed as an impact load abnormality caused by the centralized access of electric vehicles.

Images